Fix typo in subheader BertForQuestionAnswering #4
+1
−1
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
exact thanks ! |
qwang70
pushed a commit
to DRL36/pytorch-pretrained-BERT
that referenced
this pull request
Mar 2, 2019
Closed
LysandreJik
added a commit
that referenced
this pull request
Apr 10, 2020
* Initial commit to get BERT + run_glue.py on TPU * Add README section for TPU and address comments. * Cleanup TPU bits from run_glue.py (#3) TPU runner is currently implemented in: https://github.com/pytorch-tpu/transformers/blob/tpu/examples/run_glue_tpu.py. We plan to upstream this directly into `huggingface/transformers` (either `master` or `tpu`) branch once it's been more thoroughly tested. * Cleanup TPU bits from run_glue.py TPU runner is currently implemented in: https://github.com/pytorch-tpu/transformers/blob/tpu/examples/run_glue_tpu.py. We plan to upstream this directly into `huggingface/transformers` (either `master` or `tpu`) branch once it's been more thoroughly tested. * No need to call `xm.mark_step()` explicitly (#4) Since for gradient accumulation we're accumulating on batches from `ParallelLoader` instance which on next() marks the step itself. * Resolve R/W conflicts from multiprocessing (#5) * Add XLNet in list of models for `run_glue_tpu.py` (#6) * Add RoBERTa to list of models in TPU GLUE (#7) * Add RoBERTa and DistilBert to list of models in TPU GLUE (#8) * Use barriers to reduce duplicate work/resources (#9) * Shard eval dataset and aggregate eval metrics (#10) * Shard eval dataset and aggregate eval metrics Also, instead of calling `eval_loss.item()` every time do summation with tensors on device. * Change defaultdict to float * Reduce the pred, label tensors instead of metrics As brought up during review some metrics like f1 cannot be aggregated via averaging. GLUE task metrics depends largely on the dataset, so instead we sync the prediction and label tensors so that the metrics can be computed accurately on those instead. * Only use tb_writer from master (#11) * Apply huggingface black code formatting * Style * Remove `--do_lower_case` as example uses cased * Add option to specify tensorboard logdir This is needed for our testing framework which checks regressions against key metrics writtern by the summary writer. * Using configuration for `xla_device` * Prefix TPU specific comments. * num_cores clarification and namespace eval metrics * Cache features file under `args.cache_dir` Instead of under `args.data_dir`. This is needed as our test infra uses data_dir with a read-only filesystem. * Rename `run_glue_tpu` to `run_tpu_glue` Co-authored-by: LysandreJik <lysandre.debut@reseau.eseo.fr>
This was referenced Nov 12, 2020
stas00
added a commit
that referenced
this pull request
Dec 9, 2020
LysandreJik
pushed a commit
that referenced
this pull request
Dec 9, 2020
* ci-doc-job-skip-take-4 * wip * wip * wip * wip * skip yaml * wip * wip * wip * wip * wip * wip * wip * wip * wip * wip * wip * wip * ready to test * yet another way * trying with HEAD * trying with head.sha * trying with head.sha fix * trying with head.sha fix wip * undo * try to switch to sha * current branch * current branch * PR number check * joy ride * joy ride * joy ride * joy ride * joy ride * joy ride * joy ride * joy ride * joy ride * joy ride * joy ride * joy ride
2 tasks
wamartin-aml
pushed a commit
to wamartin-aml/transformers
that referenced
this pull request
Nov 1, 2021
Remove data based dependencies in T5 for ORT
4 tasks
rraminen
pushed a commit
to rraminen/transformers
that referenced
this pull request
Jun 3, 2022
Co-authored-by: Michael Melesse <mmelesse@ixt-rack-60.amd.com>
jameshennessytempus
pushed a commit
to jameshennessytempus/transformers
that referenced
this pull request
Jun 1, 2023
1 task
nikolaJovisic
added a commit
to nikolaJovisic/transformers
that referenced
this pull request
Aug 23, 2023
fix binary classification for tensorflow segformer fix binary classification for tf segformer huggingface#2 fix huggingface#5 Revert "fix huggingface#5" This reverts commit 15b516055c25faa3297196095de19b41ff0149fe. Revert "fix huggingface#4" This reverts commit 0b534e62d03db5ef74f77b61837e0561a1fc129a. fix huggingface#5 fix fix fix
gante
added a commit
that referenced
this pull request
Oct 23, 2024
5 tasks
DaryaTereshchenko
added a commit
to DaryaTereshchenko/transformers
that referenced
this pull request
Dec 17, 2024
…izer fix additional tokens list and run tests
hsilva664
pushed a commit
to hsilva664/transformers
that referenced
this pull request
Feb 27, 2025
Refactor Pre - Processor and Conversion scripts
jmamou
pushed a commit
to jmamou/transformers
that referenced
this pull request
Feb 27, 2025
fix candidate_input_ids device before running SD validation
RyanMullins
pushed a commit
to RyanMullins/transformers
that referenced
this pull request
Mar 12, 2025
LysandreJik
added a commit
that referenced
this pull request
Mar 12, 2025
* Fix converter * [Broken] Adds Gemma 3 to Hugging Face Transformers * Consolidating Config and Processor params across impls * Sorting out configuration parameters. Adds qk_norm before RoPE. Still not sure if RoPE is right. * Additional plumbing for CausalLM and ConditionalGeneration variants * incomplete draft of Orbax conversion script * More complete checkpoint conversion * Supporting Gemma 3 1B checkpoints * Updating RoPE for multiple frequencies * Adjustments to rotary embedder * Proof of life for text-only operation * Updating the conversion script to handle multimodal projection weights * Fixing tet-only conversions * Cleaner conversion script with multimodal support and a simpler processor * Additional refatcors to the Gemma3Processor * Simplified Processor to work over text representations * Updated conversion script to join text and vision embeddings at converion time * Logging for debugging * Update src/transformers/models/gemma2/modeling_gemma2.py Co-authored-by: Joshua Lochner <admin@xenova.com> * Removed extraneous Config params * Switching to fast tokenizer for checkpoint conversions * isolating siglip for performance tetsing * Minor changes for debugging tests against baselines * Adding average pooling for soft tokens * Updating processor code to enable simpler embedding interleaving for arbitrary number of images in prompts * Updating conversion script for ShieldGemma 2 conversion compatibility * Allow disable_compile to be provided as a kwarg * Refresh from modular * Updated conversion script and corrected sliding window * Fix type mismatch in cache_position (#4) * Fix dtype (#5) * Fix type mismatch in cache_position * Actually fix in the modular file Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> --------- Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> * fixes for embedding table overflow and missing image_soft_token_mask from Gemma3Processor * Adding 2D pooling for image embeddings * Revert "Adding 2D pooling for image embeddings" This reverts commit 65350cf. * Gemma3 average pooling changed from 1D to 2D * Major refactor to Gemma3MultimodalInputProjection * Updating Gemm 3 Auto* registrations * Add option to save Gemma 3 chat template with tokenizer during weights conversion * Removing unused imports * Moving out-of-vocab handling from Gemma3Processor to Gemma3ForConditionalGeneration * Removing duplicate config property * Removing final logit softcapping and 1-indexing of position ids * Fixing image processor config and none --> None typo * Fixing sliding window size for 1B * Updating image_mean and image_std in Image Processor * Attention masking changed to lower triangular * Moving image special tokens to conversion script * Mirror image processor defaults from conversion script into Gemma3ProcessorKwargs * Remove special token variables from symbol space * Moving image soft token mask computation from Gemma3Processor to Gemma3ForConditionalGeneration * tie lm_head and embedding weights Co-authored-by: Matthew Douglas <38992547+matthewdouglas@users.noreply.github.com> * Correct tied weights in Gemma3CausalLM * iterative bidirectional attention * resolving merge conflicts * Reverting to Gemma 2 HybridCache with sldiing window support and a sliding_window_pattern of 6 * Correcting RoPE scaling * clean up first pass, dummy model geenration works * final clean up before fixing tests * causal lm test works, so fine * Fix conversion * Update src/transformers/models/gemma3/processing_gemma3.py * model tests are happy * processor tests are happy * image processing tests added * fixup * Fix pre-processing in conversion * Inputs merging * Do not normalize vision embeddings * Apply Ryan's (and team) changes to attention * token type ids + mask * template * move embed scale, add rope scale, fix tests * Add chat template to tokenizer * Use prefix for causal model loading * use existing code for sliding mask from gemma2 * self.embed_tokens already normalizes * Correcting Gemma3TextConfig parameters in conversion script * typo, modular overwrites my fixes * enable device map for text model * Conversion updates * ultra nit: no einsums * update image token * copy deepcopy config + some docs * add some test, still WIP * Refactoring --include_chat_tempalte logic in converter * Update src/transformers/models/gemma3/modular_gemma3.py Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> * Add eos tokens for instruct models * dump so i can work on dgx * Removing add_bos by default * dump * add fast im proc * docs for PaS + fixup * another fixup * one more fixup * fix tests * Inverting prior BOS change * ultra nit * Reverting to Tokenizer saved with add_bos_token=True and chat template starting with BOS * resize embeds, remove sqrt, add slow test outputs * FA2 but quality is meh * nit * skip FA2, no idea what happened * last bit for green CI * please, green CI for docs * T_T * Fix for Gemma3 logits * Support both options for system prompt * Update src/transformers/models/gemma3/image_processing_gemma3_fast.py Co-authored-by: Pedro Cuenca <pedro@huggingface.co> * Update docs/source/en/model_doc/gemma3.md Co-authored-by: Pedro Cuenca <pedro@huggingface.co> * Update docs/source/en/model_doc/gemma3.md Co-authored-by: Pedro Cuenca <pedro@huggingface.co> * Update docs/source/en/model_doc/gemma3.md Co-authored-by: Pedro Cuenca <pedro@huggingface.co> * Update docs/source/en/model_doc/gemma3.md Co-authored-by: Pedro Cuenca <pedro@huggingface.co> * Update docs/source/en/model_doc/gemma3.md Co-authored-by: Pedro Cuenca <pedro@huggingface.co> * Docs updates now that assets are live * Style fixes --------- Co-authored-by: Joshua Lochner <admin@xenova.com> Co-authored-by: Pedro Cuenca <pedro@huggingface.co> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Matthew Douglas <38992547+matthewdouglas@users.noreply.github.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Raushan Turganbay <raushan.turganbay@alumni.nu.edu.kz> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: Lysandre <hi@lysand.re>
RyanMullins
added a commit
to RyanMullins/transformers
that referenced
this pull request
Jun 26, 2025
* initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3p5 overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * regenerating modeling file after syncing to HEAD * Use torch.std(..., unbiased=False) for activation sparsity (huggingface#8) * Refactoring to a single QVK Norm (huggingface#13) * AltUp: support scale_corrected_output (huggingface#14) * Converts einsums to nn.Linear (huggingface#7) * Converts einsums to nn.Linear * Removing unused variables * Aligning SharedKVCache with HybridCache (huggingface#11) * Alinging SharedKVStore with HybridCache * Remove KVStore. Refactor apply_rotary_pos_emb for sharing * Addressing review comments * Supporting split modality embeddings in Gemma3n (huggingface#10) * Adding the Embedder class * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, adding audio embedding layers, integrating embedder with the remaining architecture, adding a forward method for conditional generation * Apply suggestions from code review Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, prop drilling audio and vision configs to the text config * Removing TODO's that have been addressed * Simplify Embedder init and add audio embeddings * Embeddings refactor. Adds Gemma3nAudioEmbedder and Gemma3nVisionEmbedder * Refactoring vision and audio embeddings into ConditionalGeneration model --------- Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating attention mask for Gemma 3.5 (huggingface#15) * xxx_token_index to xxx_token_id * remvoing deprecated last_cache_position * Removing references to SigLIP * Always init per-layer inputs * Using torch.finfo().min for epsilon_tensor * Gemma3nDecoderLayer inherits from Gemma3DecoderLayer. Remove gating lambdas * fix modular GEMMA3N_INPUTS_DOCSTRING * Gemma3nAttention inherits from Gemma3Attention * Modular inheritance fixes * CausalLM conversion script for 4B model (huggingface#16) * Add Gemma3n Audio Encoder (huggingface#6) * initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (#2) * Updating DecoderBlock for Gemma 3.5 (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right Gemma 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * CausalLM conversion script for 4B model * inv_timescales to non-persistent buffer * Addressing audio encoder Attention feedback * Addressing Gemma3nAudioSSCPConvBlock feedback * Addressing Gemma3nAudioConformerAttention feedback * Addressing padding feedback * Weights conversion loads audio state dict * Always use vision_config so saving works * Token id updates for configs * Stubs for interleaving audio embs * Addressing reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Fixing cache access error * Removing duplicate code from a bad merge * Gemma 3n Text + Vision Part 1 (huggingface#17) * testing utilities for numerics comparisons * Corrected einsum to nn.Linear weights conversion * Inherit scaled word embs from Gemma3 not Bart * Fixing transposes for collapsed linears * More transpose fixes * numpy api fix * RMSNorm: Explicit kwargs, scale_shift=0.0 when with_scale=True * Force AltUp to float32 * Updating debugging script for AudioEncoder debugging * Support divide_weight_by_sqrt_fan_in from JAX for per-layer inputs * Correcting attention einsum conversions * RMSNorm in type of x * Fixing douplicate laurel norm/gating * KV sharing using the right previous indices * Refactor kv shared index computation. Correct frac_shared_layers * Use num_shared_layers instead of inferring from a fraction * fixing a bug for logging * Fix shared data_ptrs in altup inits * rope: adjust proj -> norm -> rope to preserve computation (huggingface#20) * rope: adjust proj -> norm -> rope to preserve computation * Removing some breaking language model fluff in ConditionalGeneration * Consolidate query_states transforms --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Vectorize the loops in AltUp (huggingface#19) * Vectorize the loops in AltUp * fix typo * Expanding to support batched inputs * remove extra debug script * Fix AltUp.forward --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Add 'scale_shift=0.0, with_scale=True' to the final norm in TextModel * Convert norm to 1/sqrt (huggingface#21) * Convert norm to 1/sqrt * Scale shift change per Phil's rec * Adding default activation sparsity * Fixing 2B config in weights conversion script * Fixing RMSNorm parameters - adding scale_shift and with_scale * Correcting query pre-attention scaling * Adding query_rescale_scalar to text config * Adding layer_idx to MLP * Permafix for input_layernorm * Use 1/sqrt instead of rsqrt in DecoderLayer * Fix o_proj conversion * Conversion script update for vision encoder * Removing logging for debugging timm model * Fixing bugs in Gemma3nForConditionalGeneration for text generation * Generating the modeling_gemma3n.py file * Removing the addition of an erroneous line in the modeling file * Adding gemma3n text model to modeling_auto * Bugfix: Updating the interleaving of inputs_embeds and vision_embeds * Updating the modeling file with the latest bugfix changes * Updating models/auto for Gemma 3n * using AutoTokenizer in forward test * Adding processing_gemma3n.py * Gemma 3n configured for AutoModel. Conversion script updated. * Removing errant merge artifacts --------- Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Removing errant debugging statements from Gemma 3 * Gemma3n audio model (huggingface#18) * testing utilities for numerics comparisons * Implement CumulativeGroupNorm and add to SubSampleConvProjection and SSCPConvBlock * Add audio version of forward script based on RyanMullins' implementation * Updating to match encoder tests. WIP: config question needs resolving * Updates to audio classes to enable end-to-end running * Removing vestigial classes, cleaning up print statements * Adding SiLU / Swish to audio conformer feed forward block * Shifted Gemma3p5Audio naming prefix to Gemma3NanoAudio * Adding outputs to audio test * Fixes to padding in SSCP and 1D convolution, align RMS Norm with wider model * Update forward test to load from local weights * Update conversion to process / output audio layers * Update __all__ to export audio encoder * AutoModel registration for Gemma 3n Audio * Use AutoModel for ConditionalGeneration.audio_tower * Fixing input_proj_linear transpose * Fixing Gemma3NanoAudioConformerAttention.post conversion * Fixing Gemma3NanoAudioSSCPConvBlock.conv weights conversion * Correcting indentation issue on Gemma3p5RMSNorm --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Text + Vision Part 2 (huggingface#23) * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3p5.py * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Updating configs for the 2B variant in the conversion script * Using final generation config in conversion script --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Audio Integration (huggingface#12) * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma 3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding Gemma 3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemma3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3n * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * Converting sl.Frontend to FeatureExtractor * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3n.py * Update modular Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Draft of audio data in chat template * Removing image processing. Using SigLIP instead. * Audio input going end-to-end * Fixing dtype issues in audio encoder * x-lib formatting consistency * Adding example data * Save preprocessor_config.json from conversion script * Instrumentaiton for debugging * Additional instrumentation for preprocessing debugging * Updates to preprocessor, padding; produces correct end-to-end results on sample * Tackling configuraiton TODOs * Start of feature extractor refatcor * Adds Numpy version of USM extractor, removes Torch version and dependencies * Fixing AltUp.correct coef permute * Supporting batches of single audio segment inputs * Docstrings updates for config * In-lining audio feature extraction * Adjustments to conversion script and smoke test script --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: pculliton <phillipculliton@gmail.com> * Gemma 3n renaming * Removing test data and utilities * Renaming test files * Gemma 3n refactor * Fix tokenizer config in conversion script * Address reviewer feedback * FeatureExtractor returns float32 by default * Adding basic tests for audio, and input name for audio encoder * Audio integration test, updates to model_id for other integration tests * Use scales for q and k norms (huggingface#26) * Update audio integration test to use HF dataset * Reviewer feedback * Expand embedding table to full vocab size in weights conversion * Mix-n-match MatFormers for Gemma 3n (huggingface#25) * Remove in-place operations (huggingface#30) * chore: removing inplace ops * remove [tensor] * n pattern * chore: reviewer feedback in AudioEncoder and AltUp * More grad clipping * Dynamo compatibility * fix: cache slicing error * chore: simplify shared kv cache slicing * chore: vision encoder rename in timm * fix: image processor do_normalize=False * fixup: style * chore: model_doc * fix: docs for code quality * chore: repo consistency * fix: RMSNorm in float as in prior Gemmas * fix: per_layer_inputs = None * chore: Gemma3nForCausalLM from Gemma3nForConditionalGeneration checkpoint * chore: repo consistency * Add initial unit tests for Gemma3nAudioFeatureExtractor (huggingface#27) * Add initial unit tests for Gemma3nAudioFeatureExtractor * Add basic unit tests for Gemma3nProcessor (huggingface#28) Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * parameterize tests --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * chore: code style * fix: test cases * style and consistency * fix config in the test to be coherent with layer cache sharing * fix hidden states in tests and code * inits and mappings * fix modality prefixes * test order and prefixes * fix test exception * fix class order and reduce model size for faster tests * restore _checkpoint_conversion_mapping to load Caual from Conditional * fix config mapping! * fix: reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com>
ArthurZucker
pushed a commit
that referenced
this pull request
Jun 26, 2025
NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3.5 * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3p5 module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3p5Attention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com>
ArthurZucker
pushed a commit
that referenced
this pull request
Jun 26, 2025
* initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3.5 * Adds AltUp to Gemma 3.5 * Adding Gemma3p5 overall and text config with vision and audio config placeholders (#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (#2) * Updating DecoderBlock for Gemma 3.5 (#3) * Initial Gemm3p5TextModel (#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3.5 * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3p5 module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3p5Attention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3p5AudioEncoder with nn.Sequential * Implementing Gemma3p5AudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * CausalLM conversion script for 4B model * inv_timescales to non-persistent buffer * Addressing audio encoder Attention feedback * Addressing Gemma3p5AudioSSCPConvBlock feedback * Addressing Gemma3p5AudioConformerAttention feedback * Addressing padding feedback * Weights conversion loads audio state dict * Always use vision_config so saving works * Token id updates for configs * Stubs for interleaving audio embs * Addressing reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com>
ArthurZucker
pushed a commit
that referenced
this pull request
Jun 26, 2025
* initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3.5 * Adds AltUp to Gemma 3.5 * Adding Gemma3p5 overall and text config with vision and audio config placeholders (#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (#2) * Updating DecoderBlock for Gemma 3.5 (#3) * Initial Gemm3p5TextModel (#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3.5 * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3p5 module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3p5Attention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3p5AudioEncoder with nn.Sequential * Implementing Gemma3p5AudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * Converting sl.Frontend to FeatureExtractor * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3p5.py * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Draft of audio data in chat template * Removing image processing. Using SigLIP instead. * Audio input going end-to-end * Fixing dtype issues in audio encoder * x-lib formatting consistency * Adding example data * Save preprocessor_config.json from conversion script * Instrumentaiton for debugging * Additional instrumentation for preprocessing debugging * Updates to preprocessor, padding; produces correct end-to-end results on sample * Tackling configuraiton TODOs * Start of feature extractor refatcor * Adds Numpy version of USM extractor, removes Torch version and dependencies * Fixing AltUp.correct coef permute * Supporting batches of single audio segment inputs * Docstrings updates for config * In-lining audio feature extraction * Adjustments to conversion script and smoke test script --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: pculliton <phillipculliton@gmail.com>
Cyrilvallez
added a commit
that referenced
this pull request
Jun 26, 2025
* Gemma 3n * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3p5 overall and text config with vision and audio config placeholders (#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (#2) * Updating DecoderBlock for Gemma 3n (#3) * Initial Gemm3nTextModel (#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * regenerating modeling file after syncing to HEAD * Use torch.std(..., unbiased=False) for activation sparsity (#8) * Refactoring to a single QVK Norm (#13) * AltUp: support scale_corrected_output (#14) * Converts einsums to nn.Linear (#7) * Converts einsums to nn.Linear * Removing unused variables * Aligning SharedKVCache with HybridCache (#11) * Alinging SharedKVStore with HybridCache * Remove KVStore. Refactor apply_rotary_pos_emb for sharing * Addressing review comments * Supporting split modality embeddings in Gemma3n (#10) * Adding the Embedder class * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, adding audio embedding layers, integrating embedder with the remaining architecture, adding a forward method for conditional generation * Apply suggestions from code review Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, prop drilling audio and vision configs to the text config * Removing TODO's that have been addressed * Simplify Embedder init and add audio embeddings * Embeddings refactor. Adds Gemma3nAudioEmbedder and Gemma3nVisionEmbedder * Refactoring vision and audio embeddings into ConditionalGeneration model --------- Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating attention mask for Gemma 3.5 (#15) * xxx_token_index to xxx_token_id * remvoing deprecated last_cache_position * Removing references to SigLIP * Always init per-layer inputs * Using torch.finfo().min for epsilon_tensor * Gemma3nDecoderLayer inherits from Gemma3DecoderLayer. Remove gating lambdas * fix modular GEMMA3N_INPUTS_DOCSTRING * Gemma3nAttention inherits from Gemma3Attention * Modular inheritance fixes * CausalLM conversion script for 4B model (#16) * Add Gemma3n Audio Encoder (#6) * initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3n overall and text config with vision and audio config placeholders (#3) * Adding gemma3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (#2) * Updating DecoderBlock for Gemma 3.5 (#3) * Initial Gemm3nTextModel (#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right Gemma 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * CausalLM conversion script for 4B model * inv_timescales to non-persistent buffer * Addressing audio encoder Attention feedback * Addressing Gemma3nAudioSSCPConvBlock feedback * Addressing Gemma3nAudioConformerAttention feedback * Addressing padding feedback * Weights conversion loads audio state dict * Always use vision_config so saving works * Token id updates for configs * Stubs for interleaving audio embs * Addressing reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Fixing cache access error * Removing duplicate code from a bad merge * Gemma 3n Text + Vision Part 1 (#17) * testing utilities for numerics comparisons * Corrected einsum to nn.Linear weights conversion * Inherit scaled word embs from Gemma3 not Bart * Fixing transposes for collapsed linears * More transpose fixes * numpy api fix * RMSNorm: Explicit kwargs, scale_shift=0.0 when with_scale=True * Force AltUp to float32 * Updating debugging script for AudioEncoder debugging * Support divide_weight_by_sqrt_fan_in from JAX for per-layer inputs * Correcting attention einsum conversions * RMSNorm in type of x * Fixing douplicate laurel norm/gating * KV sharing using the right previous indices * Refactor kv shared index computation. Correct frac_shared_layers * Use num_shared_layers instead of inferring from a fraction * fixing a bug for logging * Fix shared data_ptrs in altup inits * rope: adjust proj -> norm -> rope to preserve computation (#20) * rope: adjust proj -> norm -> rope to preserve computation * Removing some breaking language model fluff in ConditionalGeneration * Consolidate query_states transforms --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Vectorize the loops in AltUp (#19) * Vectorize the loops in AltUp * fix typo * Expanding to support batched inputs * remove extra debug script * Fix AltUp.forward --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Add 'scale_shift=0.0, with_scale=True' to the final norm in TextModel * Convert norm to 1/sqrt (#21) * Convert norm to 1/sqrt * Scale shift change per Phil's rec * Adding default activation sparsity * Fixing 2B config in weights conversion script * Fixing RMSNorm parameters - adding scale_shift and with_scale * Correcting query pre-attention scaling * Adding query_rescale_scalar to text config * Adding layer_idx to MLP * Permafix for input_layernorm * Use 1/sqrt instead of rsqrt in DecoderLayer * Fix o_proj conversion * Conversion script update for vision encoder * Removing logging for debugging timm model * Fixing bugs in Gemma3nForConditionalGeneration for text generation * Generating the modeling_gemma3n.py file * Removing the addition of an erroneous line in the modeling file * Adding gemma3n text model to modeling_auto * Bugfix: Updating the interleaving of inputs_embeds and vision_embeds * Updating the modeling file with the latest bugfix changes * Updating models/auto for Gemma 3n * using AutoTokenizer in forward test * Adding processing_gemma3n.py * Gemma 3n configured for AutoModel. Conversion script updated. * Removing errant merge artifacts --------- Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Removing errant debugging statements from Gemma 3 * Gemma3n audio model (#18) * testing utilities for numerics comparisons * Implement CumulativeGroupNorm and add to SubSampleConvProjection and SSCPConvBlock * Add audio version of forward script based on RyanMullins' implementation * Updating to match encoder tests. WIP: config question needs resolving * Updates to audio classes to enable end-to-end running * Removing vestigial classes, cleaning up print statements * Adding SiLU / Swish to audio conformer feed forward block * Shifted Gemma3p5Audio naming prefix to Gemma3NanoAudio * Adding outputs to audio test * Fixes to padding in SSCP and 1D convolution, align RMS Norm with wider model * Update forward test to load from local weights * Update conversion to process / output audio layers * Update __all__ to export audio encoder * AutoModel registration for Gemma 3n Audio * Use AutoModel for ConditionalGeneration.audio_tower * Fixing input_proj_linear transpose * Fixing Gemma3NanoAudioConformerAttention.post conversion * Fixing Gemma3NanoAudioSSCPConvBlock.conv weights conversion * Correcting indentation issue on Gemma3p5RMSNorm --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Text + Vision Part 2 (#23) * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3p5.py * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Updating configs for the 2B variant in the conversion script * Using final generation config in conversion script --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Audio Integration (#12) * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma 3n overall and text config with vision and audio config placeholders (#3) * Adding Gemma 3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (#2) * Updating DecoderBlock for Gemma 3n (#3) * Initial Gemma3nTextModel (#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3n * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * Converting sl.Frontend to FeatureExtractor * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3n.py * Update modular Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Draft of audio data in chat template * Removing image processing. Using SigLIP instead. * Audio input going end-to-end * Fixing dtype issues in audio encoder * x-lib formatting consistency * Adding example data * Save preprocessor_config.json from conversion script * Instrumentaiton for debugging * Additional instrumentation for preprocessing debugging * Updates to preprocessor, padding; produces correct end-to-end results on sample * Tackling configuraiton TODOs * Start of feature extractor refatcor * Adds Numpy version of USM extractor, removes Torch version and dependencies * Fixing AltUp.correct coef permute * Supporting batches of single audio segment inputs * Docstrings updates for config * In-lining audio feature extraction * Adjustments to conversion script and smoke test script --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: pculliton <phillipculliton@gmail.com> * Gemma 3n renaming * Removing test data and utilities * Renaming test files * Gemma 3n refactor * Fix tokenizer config in conversion script * Address reviewer feedback * FeatureExtractor returns float32 by default * Adding basic tests for audio, and input name for audio encoder * Audio integration test, updates to model_id for other integration tests * Use scales for q and k norms (#26) * Update audio integration test to use HF dataset * Reviewer feedback * Expand embedding table to full vocab size in weights conversion * Mix-n-match MatFormers for Gemma 3n (#25) * Remove in-place operations (#30) * chore: removing inplace ops * remove [tensor] * n pattern * chore: reviewer feedback in AudioEncoder and AltUp * More grad clipping * Dynamo compatibility * fix: cache slicing error * chore: simplify shared kv cache slicing * chore: vision encoder rename in timm * fix: image processor do_normalize=False * fixup: style * chore: model_doc * fix: docs for code quality * chore: repo consistency * fix: RMSNorm in float as in prior Gemmas * fix: per_layer_inputs = None * chore: Gemma3nForCausalLM from Gemma3nForConditionalGeneration checkpoint * chore: repo consistency * Add initial unit tests for Gemma3nAudioFeatureExtractor (#27) * Add initial unit tests for Gemma3nAudioFeatureExtractor * Add basic unit tests for Gemma3nProcessor (#28) Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * parameterize tests --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * chore: code style * fix: test cases * style and consistency * fix config in the test to be coherent with layer cache sharing * fix hidden states in tests and code * inits and mappings * fix modality prefixes * test order and prefixes * fix test exception * fix class order and reduce model size for faster tests * restore _checkpoint_conversion_mapping to load Caual from Conditional * fix config mapping! * fix: reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> * fix import test * add model args * auto_docstring * replace test path * consistency * skip tests for now * fix docstring for doc builder * skip unused attr --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>
yonigozlan
pushed a commit
to yonigozlan/transformers
that referenced
this pull request
Jul 29, 2025
refactor convert
4 tasks
b-re-w
pushed a commit
to retentionlabs/transformers-ttt
that referenced
this pull request
Aug 1, 2025
Fix README import
ArthurZucker
added a commit
that referenced
this pull request
Aug 5, 2025
Expert Parallelism + TP fixes
qubvel
added a commit
to qubvel/transformers
that referenced
this pull request
Aug 14, 2025
Refactor DINOv3ViT
Guo-Chenxu
pushed a commit
to Guo-Chenxu/transformers
that referenced
this pull request
Aug 22, 2025
resolve comments
LifengWang
pushed a commit
to LifengWang/transformers
that referenced
this pull request
Sep 5, 2025
zaristei
pushed a commit
to zaristei/transformers
that referenced
this pull request
Sep 9, 2025
* Gemma 3n * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3p5 overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * regenerating modeling file after syncing to HEAD * Use torch.std(..., unbiased=False) for activation sparsity (huggingface#8) * Refactoring to a single QVK Norm (huggingface#13) * AltUp: support scale_corrected_output (huggingface#14) * Converts einsums to nn.Linear (huggingface#7) * Converts einsums to nn.Linear * Removing unused variables * Aligning SharedKVCache with HybridCache (huggingface#11) * Alinging SharedKVStore with HybridCache * Remove KVStore. Refactor apply_rotary_pos_emb for sharing * Addressing review comments * Supporting split modality embeddings in Gemma3n (huggingface#10) * Adding the Embedder class * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, adding audio embedding layers, integrating embedder with the remaining architecture, adding a forward method for conditional generation * Apply suggestions from code review Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, prop drilling audio and vision configs to the text config * Removing TODO's that have been addressed * Simplify Embedder init and add audio embeddings * Embeddings refactor. Adds Gemma3nAudioEmbedder and Gemma3nVisionEmbedder * Refactoring vision and audio embeddings into ConditionalGeneration model --------- Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating attention mask for Gemma 3.5 (huggingface#15) * xxx_token_index to xxx_token_id * remvoing deprecated last_cache_position * Removing references to SigLIP * Always init per-layer inputs * Using torch.finfo().min for epsilon_tensor * Gemma3nDecoderLayer inherits from Gemma3DecoderLayer. Remove gating lambdas * fix modular GEMMA3N_INPUTS_DOCSTRING * Gemma3nAttention inherits from Gemma3Attention * Modular inheritance fixes * CausalLM conversion script for 4B model (huggingface#16) * Add Gemma3n Audio Encoder (huggingface#6) * initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3.5 (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right Gemma 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * CausalLM conversion script for 4B model * inv_timescales to non-persistent buffer * Addressing audio encoder Attention feedback * Addressing Gemma3nAudioSSCPConvBlock feedback * Addressing Gemma3nAudioConformerAttention feedback * Addressing padding feedback * Weights conversion loads audio state dict * Always use vision_config so saving works * Token id updates for configs * Stubs for interleaving audio embs * Addressing reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Fixing cache access error * Removing duplicate code from a bad merge * Gemma 3n Text + Vision Part 1 (huggingface#17) * testing utilities for numerics comparisons * Corrected einsum to nn.Linear weights conversion * Inherit scaled word embs from Gemma3 not Bart * Fixing transposes for collapsed linears * More transpose fixes * numpy api fix * RMSNorm: Explicit kwargs, scale_shift=0.0 when with_scale=True * Force AltUp to float32 * Updating debugging script for AudioEncoder debugging * Support divide_weight_by_sqrt_fan_in from JAX for per-layer inputs * Correcting attention einsum conversions * RMSNorm in type of x * Fixing douplicate laurel norm/gating * KV sharing using the right previous indices * Refactor kv shared index computation. Correct frac_shared_layers * Use num_shared_layers instead of inferring from a fraction * fixing a bug for logging * Fix shared data_ptrs in altup inits * rope: adjust proj -> norm -> rope to preserve computation (huggingface#20) * rope: adjust proj -> norm -> rope to preserve computation * Removing some breaking language model fluff in ConditionalGeneration * Consolidate query_states transforms --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Vectorize the loops in AltUp (huggingface#19) * Vectorize the loops in AltUp * fix typo * Expanding to support batched inputs * remove extra debug script * Fix AltUp.forward --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Add 'scale_shift=0.0, with_scale=True' to the final norm in TextModel * Convert norm to 1/sqrt (huggingface#21) * Convert norm to 1/sqrt * Scale shift change per Phil's rec * Adding default activation sparsity * Fixing 2B config in weights conversion script * Fixing RMSNorm parameters - adding scale_shift and with_scale * Correcting query pre-attention scaling * Adding query_rescale_scalar to text config * Adding layer_idx to MLP * Permafix for input_layernorm * Use 1/sqrt instead of rsqrt in DecoderLayer * Fix o_proj conversion * Conversion script update for vision encoder * Removing logging for debugging timm model * Fixing bugs in Gemma3nForConditionalGeneration for text generation * Generating the modeling_gemma3n.py file * Removing the addition of an erroneous line in the modeling file * Adding gemma3n text model to modeling_auto * Bugfix: Updating the interleaving of inputs_embeds and vision_embeds * Updating the modeling file with the latest bugfix changes * Updating models/auto for Gemma 3n * using AutoTokenizer in forward test * Adding processing_gemma3n.py * Gemma 3n configured for AutoModel. Conversion script updated. * Removing errant merge artifacts --------- Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Removing errant debugging statements from Gemma 3 * Gemma3n audio model (huggingface#18) * testing utilities for numerics comparisons * Implement CumulativeGroupNorm and add to SubSampleConvProjection and SSCPConvBlock * Add audio version of forward script based on RyanMullins' implementation * Updating to match encoder tests. WIP: config question needs resolving * Updates to audio classes to enable end-to-end running * Removing vestigial classes, cleaning up print statements * Adding SiLU / Swish to audio conformer feed forward block * Shifted Gemma3p5Audio naming prefix to Gemma3NanoAudio * Adding outputs to audio test * Fixes to padding in SSCP and 1D convolution, align RMS Norm with wider model * Update forward test to load from local weights * Update conversion to process / output audio layers * Update __all__ to export audio encoder * AutoModel registration for Gemma 3n Audio * Use AutoModel for ConditionalGeneration.audio_tower * Fixing input_proj_linear transpose * Fixing Gemma3NanoAudioConformerAttention.post conversion * Fixing Gemma3NanoAudioSSCPConvBlock.conv weights conversion * Correcting indentation issue on Gemma3p5RMSNorm --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Text + Vision Part 2 (huggingface#23) * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3p5.py * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Updating configs for the 2B variant in the conversion script * Using final generation config in conversion script --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Audio Integration (huggingface#12) * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma 3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding Gemma 3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemma3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3n * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * Converting sl.Frontend to FeatureExtractor * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3n.py * Update modular Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Draft of audio data in chat template * Removing image processing. Using SigLIP instead. * Audio input going end-to-end * Fixing dtype issues in audio encoder * x-lib formatting consistency * Adding example data * Save preprocessor_config.json from conversion script * Instrumentaiton for debugging * Additional instrumentation for preprocessing debugging * Updates to preprocessor, padding; produces correct end-to-end results on sample * Tackling configuraiton TODOs * Start of feature extractor refatcor * Adds Numpy version of USM extractor, removes Torch version and dependencies * Fixing AltUp.correct coef permute * Supporting batches of single audio segment inputs * Docstrings updates for config * In-lining audio feature extraction * Adjustments to conversion script and smoke test script --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: pculliton <phillipculliton@gmail.com> * Gemma 3n renaming * Removing test data and utilities * Renaming test files * Gemma 3n refactor * Fix tokenizer config in conversion script * Address reviewer feedback * FeatureExtractor returns float32 by default * Adding basic tests for audio, and input name for audio encoder * Audio integration test, updates to model_id for other integration tests * Use scales for q and k norms (huggingface#26) * Update audio integration test to use HF dataset * Reviewer feedback * Expand embedding table to full vocab size in weights conversion * Mix-n-match MatFormers for Gemma 3n (huggingface#25) * Remove in-place operations (huggingface#30) * chore: removing inplace ops * remove [tensor] * n pattern * chore: reviewer feedback in AudioEncoder and AltUp * More grad clipping * Dynamo compatibility * fix: cache slicing error * chore: simplify shared kv cache slicing * chore: vision encoder rename in timm * fix: image processor do_normalize=False * fixup: style * chore: model_doc * fix: docs for code quality * chore: repo consistency * fix: RMSNorm in float as in prior Gemmas * fix: per_layer_inputs = None * chore: Gemma3nForCausalLM from Gemma3nForConditionalGeneration checkpoint * chore: repo consistency * Add initial unit tests for Gemma3nAudioFeatureExtractor (huggingface#27) * Add initial unit tests for Gemma3nAudioFeatureExtractor * Add basic unit tests for Gemma3nProcessor (huggingface#28) Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * parameterize tests --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * chore: code style * fix: test cases * style and consistency * fix config in the test to be coherent with layer cache sharing * fix hidden states in tests and code * inits and mappings * fix modality prefixes * test order and prefixes * fix test exception * fix class order and reduce model size for faster tests * restore _checkpoint_conversion_mapping to load Caual from Conditional * fix config mapping! * fix: reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> * fix import test * add model args * auto_docstring * replace test path * consistency * skip tests for now * fix docstring for doc builder * skip unused attr --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>
zaristei
pushed a commit
to zaristei/transformers
that referenced
this pull request
Sep 9, 2025
* Gemma 3n * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3p5 overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * regenerating modeling file after syncing to HEAD * Use torch.std(..., unbiased=False) for activation sparsity (huggingface#8) * Refactoring to a single QVK Norm (huggingface#13) * AltUp: support scale_corrected_output (huggingface#14) * Converts einsums to nn.Linear (huggingface#7) * Converts einsums to nn.Linear * Removing unused variables * Aligning SharedKVCache with HybridCache (huggingface#11) * Alinging SharedKVStore with HybridCache * Remove KVStore. Refactor apply_rotary_pos_emb for sharing * Addressing review comments * Supporting split modality embeddings in Gemma3n (huggingface#10) * Adding the Embedder class * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, adding audio embedding layers, integrating embedder with the remaining architecture, adding a forward method for conditional generation * Apply suggestions from code review Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, prop drilling audio and vision configs to the text config * Removing TODO's that have been addressed * Simplify Embedder init and add audio embeddings * Embeddings refactor. Adds Gemma3nAudioEmbedder and Gemma3nVisionEmbedder * Refactoring vision and audio embeddings into ConditionalGeneration model --------- Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating attention mask for Gemma 3.5 (huggingface#15) * xxx_token_index to xxx_token_id * remvoing deprecated last_cache_position * Removing references to SigLIP * Always init per-layer inputs * Using torch.finfo().min for epsilon_tensor * Gemma3nDecoderLayer inherits from Gemma3DecoderLayer. Remove gating lambdas * fix modular GEMMA3N_INPUTS_DOCSTRING * Gemma3nAttention inherits from Gemma3Attention * Modular inheritance fixes * CausalLM conversion script for 4B model (huggingface#16) * Add Gemma3n Audio Encoder (huggingface#6) * initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3.5 (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right Gemma 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * CausalLM conversion script for 4B model * inv_timescales to non-persistent buffer * Addressing audio encoder Attention feedback * Addressing Gemma3nAudioSSCPConvBlock feedback * Addressing Gemma3nAudioConformerAttention feedback * Addressing padding feedback * Weights conversion loads audio state dict * Always use vision_config so saving works * Token id updates for configs * Stubs for interleaving audio embs * Addressing reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Fixing cache access error * Removing duplicate code from a bad merge * Gemma 3n Text + Vision Part 1 (huggingface#17) * testing utilities for numerics comparisons * Corrected einsum to nn.Linear weights conversion * Inherit scaled word embs from Gemma3 not Bart * Fixing transposes for collapsed linears * More transpose fixes * numpy api fix * RMSNorm: Explicit kwargs, scale_shift=0.0 when with_scale=True * Force AltUp to float32 * Updating debugging script for AudioEncoder debugging * Support divide_weight_by_sqrt_fan_in from JAX for per-layer inputs * Correcting attention einsum conversions * RMSNorm in type of x * Fixing douplicate laurel norm/gating * KV sharing using the right previous indices * Refactor kv shared index computation. Correct frac_shared_layers * Use num_shared_layers instead of inferring from a fraction * fixing a bug for logging * Fix shared data_ptrs in altup inits * rope: adjust proj -> norm -> rope to preserve computation (huggingface#20) * rope: adjust proj -> norm -> rope to preserve computation * Removing some breaking language model fluff in ConditionalGeneration * Consolidate query_states transforms --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Vectorize the loops in AltUp (huggingface#19) * Vectorize the loops in AltUp * fix typo * Expanding to support batched inputs * remove extra debug script * Fix AltUp.forward --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Add 'scale_shift=0.0, with_scale=True' to the final norm in TextModel * Convert norm to 1/sqrt (huggingface#21) * Convert norm to 1/sqrt * Scale shift change per Phil's rec * Adding default activation sparsity * Fixing 2B config in weights conversion script * Fixing RMSNorm parameters - adding scale_shift and with_scale * Correcting query pre-attention scaling * Adding query_rescale_scalar to text config * Adding layer_idx to MLP * Permafix for input_layernorm * Use 1/sqrt instead of rsqrt in DecoderLayer * Fix o_proj conversion * Conversion script update for vision encoder * Removing logging for debugging timm model * Fixing bugs in Gemma3nForConditionalGeneration for text generation * Generating the modeling_gemma3n.py file * Removing the addition of an erroneous line in the modeling file * Adding gemma3n text model to modeling_auto * Bugfix: Updating the interleaving of inputs_embeds and vision_embeds * Updating the modeling file with the latest bugfix changes * Updating models/auto for Gemma 3n * using AutoTokenizer in forward test * Adding processing_gemma3n.py * Gemma 3n configured for AutoModel. Conversion script updated. * Removing errant merge artifacts --------- Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Removing errant debugging statements from Gemma 3 * Gemma3n audio model (huggingface#18) * testing utilities for numerics comparisons * Implement CumulativeGroupNorm and add to SubSampleConvProjection and SSCPConvBlock * Add audio version of forward script based on RyanMullins' implementation * Updating to match encoder tests. WIP: config question needs resolving * Updates to audio classes to enable end-to-end running * Removing vestigial classes, cleaning up print statements * Adding SiLU / Swish to audio conformer feed forward block * Shifted Gemma3p5Audio naming prefix to Gemma3NanoAudio * Adding outputs to audio test * Fixes to padding in SSCP and 1D convolution, align RMS Norm with wider model * Update forward test to load from local weights * Update conversion to process / output audio layers * Update __all__ to export audio encoder * AutoModel registration for Gemma 3n Audio * Use AutoModel for ConditionalGeneration.audio_tower * Fixing input_proj_linear transpose * Fixing Gemma3NanoAudioConformerAttention.post conversion * Fixing Gemma3NanoAudioSSCPConvBlock.conv weights conversion * Correcting indentation issue on Gemma3p5RMSNorm --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Text + Vision Part 2 (huggingface#23) * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3p5.py * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Updating configs for the 2B variant in the conversion script * Using final generation config in conversion script --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Audio Integration (huggingface#12) * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma 3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding Gemma 3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemma3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3n * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * Converting sl.Frontend to FeatureExtractor * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3n.py * Update modular Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Draft of audio data in chat template * Removing image processing. Using SigLIP instead. * Audio input going end-to-end * Fixing dtype issues in audio encoder * x-lib formatting consistency * Adding example data * Save preprocessor_config.json from conversion script * Instrumentaiton for debugging * Additional instrumentation for preprocessing debugging * Updates to preprocessor, padding; produces correct end-to-end results on sample * Tackling configuraiton TODOs * Start of feature extractor refatcor * Adds Numpy version of USM extractor, removes Torch version and dependencies * Fixing AltUp.correct coef permute * Supporting batches of single audio segment inputs * Docstrings updates for config * In-lining audio feature extraction * Adjustments to conversion script and smoke test script --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: pculliton <phillipculliton@gmail.com> * Gemma 3n renaming * Removing test data and utilities * Renaming test files * Gemma 3n refactor * Fix tokenizer config in conversion script * Address reviewer feedback * FeatureExtractor returns float32 by default * Adding basic tests for audio, and input name for audio encoder * Audio integration test, updates to model_id for other integration tests * Use scales for q and k norms (huggingface#26) * Update audio integration test to use HF dataset * Reviewer feedback * Expand embedding table to full vocab size in weights conversion * Mix-n-match MatFormers for Gemma 3n (huggingface#25) * Remove in-place operations (huggingface#30) * chore: removing inplace ops * remove [tensor] * n pattern * chore: reviewer feedback in AudioEncoder and AltUp * More grad clipping * Dynamo compatibility * fix: cache slicing error * chore: simplify shared kv cache slicing * chore: vision encoder rename in timm * fix: image processor do_normalize=False * fixup: style * chore: model_doc * fix: docs for code quality * chore: repo consistency * fix: RMSNorm in float as in prior Gemmas * fix: per_layer_inputs = None * chore: Gemma3nForCausalLM from Gemma3nForConditionalGeneration checkpoint * chore: repo consistency * Add initial unit tests for Gemma3nAudioFeatureExtractor (huggingface#27) * Add initial unit tests for Gemma3nAudioFeatureExtractor * Add basic unit tests for Gemma3nProcessor (huggingface#28) Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * parameterize tests --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * chore: code style * fix: test cases * style and consistency * fix config in the test to be coherent with layer cache sharing * fix hidden states in tests and code * inits and mappings * fix modality prefixes * test order and prefixes * fix test exception * fix class order and reduce model size for faster tests * restore _checkpoint_conversion_mapping to load Caual from Conditional * fix config mapping! * fix: reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> * fix import test * add model args * auto_docstring * replace test path * consistency * skip tests for now * fix docstring for doc builder * skip unused attr --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>
zaristei
pushed a commit
to zaristei/transformers
that referenced
this pull request
Sep 9, 2025
* Gemma 3n * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3p5 overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * regenerating modeling file after syncing to HEAD * Use torch.std(..., unbiased=False) for activation sparsity (huggingface#8) * Refactoring to a single QVK Norm (huggingface#13) * AltUp: support scale_corrected_output (huggingface#14) * Converts einsums to nn.Linear (huggingface#7) * Converts einsums to nn.Linear * Removing unused variables * Aligning SharedKVCache with HybridCache (huggingface#11) * Alinging SharedKVStore with HybridCache * Remove KVStore. Refactor apply_rotary_pos_emb for sharing * Addressing review comments * Supporting split modality embeddings in Gemma3n (huggingface#10) * Adding the Embedder class * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, adding audio embedding layers, integrating embedder with the remaining architecture, adding a forward method for conditional generation * Apply suggestions from code review Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, prop drilling audio and vision configs to the text config * Removing TODO's that have been addressed * Simplify Embedder init and add audio embeddings * Embeddings refactor. Adds Gemma3nAudioEmbedder and Gemma3nVisionEmbedder * Refactoring vision and audio embeddings into ConditionalGeneration model --------- Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating attention mask for Gemma 3.5 (huggingface#15) * xxx_token_index to xxx_token_id * remvoing deprecated last_cache_position * Removing references to SigLIP * Always init per-layer inputs * Using torch.finfo().min for epsilon_tensor * Gemma3nDecoderLayer inherits from Gemma3DecoderLayer. Remove gating lambdas * fix modular GEMMA3N_INPUTS_DOCSTRING * Gemma3nAttention inherits from Gemma3Attention * Modular inheritance fixes * CausalLM conversion script for 4B model (huggingface#16) * Add Gemma3n Audio Encoder (huggingface#6) * initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3.5 (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right Gemma 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * CausalLM conversion script for 4B model * inv_timescales to non-persistent buffer * Addressing audio encoder Attention feedback * Addressing Gemma3nAudioSSCPConvBlock feedback * Addressing Gemma3nAudioConformerAttention feedback * Addressing padding feedback * Weights conversion loads audio state dict * Always use vision_config so saving works * Token id updates for configs * Stubs for interleaving audio embs * Addressing reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Fixing cache access error * Removing duplicate code from a bad merge * Gemma 3n Text + Vision Part 1 (huggingface#17) * testing utilities for numerics comparisons * Corrected einsum to nn.Linear weights conversion * Inherit scaled word embs from Gemma3 not Bart * Fixing transposes for collapsed linears * More transpose fixes * numpy api fix * RMSNorm: Explicit kwargs, scale_shift=0.0 when with_scale=True * Force AltUp to float32 * Updating debugging script for AudioEncoder debugging * Support divide_weight_by_sqrt_fan_in from JAX for per-layer inputs * Correcting attention einsum conversions * RMSNorm in type of x * Fixing douplicate laurel norm/gating * KV sharing using the right previous indices * Refactor kv shared index computation. Correct frac_shared_layers * Use num_shared_layers instead of inferring from a fraction * fixing a bug for logging * Fix shared data_ptrs in altup inits * rope: adjust proj -> norm -> rope to preserve computation (huggingface#20) * rope: adjust proj -> norm -> rope to preserve computation * Removing some breaking language model fluff in ConditionalGeneration * Consolidate query_states transforms --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Vectorize the loops in AltUp (huggingface#19) * Vectorize the loops in AltUp * fix typo * Expanding to support batched inputs * remove extra debug script * Fix AltUp.forward --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Add 'scale_shift=0.0, with_scale=True' to the final norm in TextModel * Convert norm to 1/sqrt (huggingface#21) * Convert norm to 1/sqrt * Scale shift change per Phil's rec * Adding default activation sparsity * Fixing 2B config in weights conversion script * Fixing RMSNorm parameters - adding scale_shift and with_scale * Correcting query pre-attention scaling * Adding query_rescale_scalar to text config * Adding layer_idx to MLP * Permafix for input_layernorm * Use 1/sqrt instead of rsqrt in DecoderLayer * Fix o_proj conversion * Conversion script update for vision encoder * Removing logging for debugging timm model * Fixing bugs in Gemma3nForConditionalGeneration for text generation * Generating the modeling_gemma3n.py file * Removing the addition of an erroneous line in the modeling file * Adding gemma3n text model to modeling_auto * Bugfix: Updating the interleaving of inputs_embeds and vision_embeds * Updating the modeling file with the latest bugfix changes * Updating models/auto for Gemma 3n * using AutoTokenizer in forward test * Adding processing_gemma3n.py * Gemma 3n configured for AutoModel. Conversion script updated. * Removing errant merge artifacts --------- Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Removing errant debugging statements from Gemma 3 * Gemma3n audio model (huggingface#18) * testing utilities for numerics comparisons * Implement CumulativeGroupNorm and add to SubSampleConvProjection and SSCPConvBlock * Add audio version of forward script based on RyanMullins' implementation * Updating to match encoder tests. WIP: config question needs resolving * Updates to audio classes to enable end-to-end running * Removing vestigial classes, cleaning up print statements * Adding SiLU / Swish to audio conformer feed forward block * Shifted Gemma3p5Audio naming prefix to Gemma3NanoAudio * Adding outputs to audio test * Fixes to padding in SSCP and 1D convolution, align RMS Norm with wider model * Update forward test to load from local weights * Update conversion to process / output audio layers * Update __all__ to export audio encoder * AutoModel registration for Gemma 3n Audio * Use AutoModel for ConditionalGeneration.audio_tower * Fixing input_proj_linear transpose * Fixing Gemma3NanoAudioConformerAttention.post conversion * Fixing Gemma3NanoAudioSSCPConvBlock.conv weights conversion * Correcting indentation issue on Gemma3p5RMSNorm --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Text + Vision Part 2 (huggingface#23) * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3p5.py * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Updating configs for the 2B variant in the conversion script * Using final generation config in conversion script --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Audio Integration (huggingface#12) * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma 3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding Gemma 3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemma3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3n * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * Converting sl.Frontend to FeatureExtractor * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3n.py * Update modular Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Draft of audio data in chat template * Removing image processing. Using SigLIP instead. * Audio input going end-to-end * Fixing dtype issues in audio encoder * x-lib formatting consistency * Adding example data * Save preprocessor_config.json from conversion script * Instrumentaiton for debugging * Additional instrumentation for preprocessing debugging * Updates to preprocessor, padding; produces correct end-to-end results on sample * Tackling configuraiton TODOs * Start of feature extractor refatcor * Adds Numpy version of USM extractor, removes Torch version and dependencies * Fixing AltUp.correct coef permute * Supporting batches of single audio segment inputs * Docstrings updates for config * In-lining audio feature extraction * Adjustments to conversion script and smoke test script --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: pculliton <phillipculliton@gmail.com> * Gemma 3n renaming * Removing test data and utilities * Renaming test files * Gemma 3n refactor * Fix tokenizer config in conversion script * Address reviewer feedback * FeatureExtractor returns float32 by default * Adding basic tests for audio, and input name for audio encoder * Audio integration test, updates to model_id for other integration tests * Use scales for q and k norms (huggingface#26) * Update audio integration test to use HF dataset * Reviewer feedback * Expand embedding table to full vocab size in weights conversion * Mix-n-match MatFormers for Gemma 3n (huggingface#25) * Remove in-place operations (huggingface#30) * chore: removing inplace ops * remove [tensor] * n pattern * chore: reviewer feedback in AudioEncoder and AltUp * More grad clipping * Dynamo compatibility * fix: cache slicing error * chore: simplify shared kv cache slicing * chore: vision encoder rename in timm * fix: image processor do_normalize=False * fixup: style * chore: model_doc * fix: docs for code quality * chore: repo consistency * fix: RMSNorm in float as in prior Gemmas * fix: per_layer_inputs = None * chore: Gemma3nForCausalLM from Gemma3nForConditionalGeneration checkpoint * chore: repo consistency * Add initial unit tests for Gemma3nAudioFeatureExtractor (huggingface#27) * Add initial unit tests for Gemma3nAudioFeatureExtractor * Add basic unit tests for Gemma3nProcessor (huggingface#28) Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * parameterize tests --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * chore: code style * fix: test cases * style and consistency * fix config in the test to be coherent with layer cache sharing * fix hidden states in tests and code * inits and mappings * fix modality prefixes * test order and prefixes * fix test exception * fix class order and reduce model size for faster tests * restore _checkpoint_conversion_mapping to load Caual from Conditional * fix config mapping! * fix: reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> * fix import test * add model args * auto_docstring * replace test path * consistency * skip tests for now * fix docstring for doc builder * skip unused attr --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>
zaristei
pushed a commit
to zaristei/transformers
that referenced
this pull request
Sep 9, 2025
* Gemma 3n * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3p5 overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * regenerating modeling file after syncing to HEAD * Use torch.std(..., unbiased=False) for activation sparsity (huggingface#8) * Refactoring to a single QVK Norm (huggingface#13) * AltUp: support scale_corrected_output (huggingface#14) * Converts einsums to nn.Linear (huggingface#7) * Converts einsums to nn.Linear * Removing unused variables * Aligning SharedKVCache with HybridCache (huggingface#11) * Alinging SharedKVStore with HybridCache * Remove KVStore. Refactor apply_rotary_pos_emb for sharing * Addressing review comments * Supporting split modality embeddings in Gemma3n (huggingface#10) * Adding the Embedder class * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, adding audio embedding layers, integrating embedder with the remaining architecture, adding a forward method for conditional generation * Apply suggestions from code review Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, prop drilling audio and vision configs to the text config * Removing TODO's that have been addressed * Simplify Embedder init and add audio embeddings * Embeddings refactor. Adds Gemma3nAudioEmbedder and Gemma3nVisionEmbedder * Refactoring vision and audio embeddings into ConditionalGeneration model --------- Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating attention mask for Gemma 3.5 (huggingface#15) * xxx_token_index to xxx_token_id * remvoing deprecated last_cache_position * Removing references to SigLIP * Always init per-layer inputs * Using torch.finfo().min for epsilon_tensor * Gemma3nDecoderLayer inherits from Gemma3DecoderLayer. Remove gating lambdas * fix modular GEMMA3N_INPUTS_DOCSTRING * Gemma3nAttention inherits from Gemma3Attention * Modular inheritance fixes * CausalLM conversion script for 4B model (huggingface#16) * Add Gemma3n Audio Encoder (huggingface#6) * initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3.5 (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right Gemma 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * CausalLM conversion script for 4B model * inv_timescales to non-persistent buffer * Addressing audio encoder Attention feedback * Addressing Gemma3nAudioSSCPConvBlock feedback * Addressing Gemma3nAudioConformerAttention feedback * Addressing padding feedback * Weights conversion loads audio state dict * Always use vision_config so saving works * Token id updates for configs * Stubs for interleaving audio embs * Addressing reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Fixing cache access error * Removing duplicate code from a bad merge * Gemma 3n Text + Vision Part 1 (huggingface#17) * testing utilities for numerics comparisons * Corrected einsum to nn.Linear weights conversion * Inherit scaled word embs from Gemma3 not Bart * Fixing transposes for collapsed linears * More transpose fixes * numpy api fix * RMSNorm: Explicit kwargs, scale_shift=0.0 when with_scale=True * Force AltUp to float32 * Updating debugging script for AudioEncoder debugging * Support divide_weight_by_sqrt_fan_in from JAX for per-layer inputs * Correcting attention einsum conversions * RMSNorm in type of x * Fixing douplicate laurel norm/gating * KV sharing using the right previous indices * Refactor kv shared index computation. Correct frac_shared_layers * Use num_shared_layers instead of inferring from a fraction * fixing a bug for logging * Fix shared data_ptrs in altup inits * rope: adjust proj -> norm -> rope to preserve computation (huggingface#20) * rope: adjust proj -> norm -> rope to preserve computation * Removing some breaking language model fluff in ConditionalGeneration * Consolidate query_states transforms --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Vectorize the loops in AltUp (huggingface#19) * Vectorize the loops in AltUp * fix typo * Expanding to support batched inputs * remove extra debug script * Fix AltUp.forward --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Add 'scale_shift=0.0, with_scale=True' to the final norm in TextModel * Convert norm to 1/sqrt (huggingface#21) * Convert norm to 1/sqrt * Scale shift change per Phil's rec * Adding default activation sparsity * Fixing 2B config in weights conversion script * Fixing RMSNorm parameters - adding scale_shift and with_scale * Correcting query pre-attention scaling * Adding query_rescale_scalar to text config * Adding layer_idx to MLP * Permafix for input_layernorm * Use 1/sqrt instead of rsqrt in DecoderLayer * Fix o_proj conversion * Conversion script update for vision encoder * Removing logging for debugging timm model * Fixing bugs in Gemma3nForConditionalGeneration for text generation * Generating the modeling_gemma3n.py file * Removing the addition of an erroneous line in the modeling file * Adding gemma3n text model to modeling_auto * Bugfix: Updating the interleaving of inputs_embeds and vision_embeds * Updating the modeling file with the latest bugfix changes * Updating models/auto for Gemma 3n * using AutoTokenizer in forward test * Adding processing_gemma3n.py * Gemma 3n configured for AutoModel. Conversion script updated. * Removing errant merge artifacts --------- Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Removing errant debugging statements from Gemma 3 * Gemma3n audio model (huggingface#18) * testing utilities for numerics comparisons * Implement CumulativeGroupNorm and add to SubSampleConvProjection and SSCPConvBlock * Add audio version of forward script based on RyanMullins' implementation * Updating to match encoder tests. WIP: config question needs resolving * Updates to audio classes to enable end-to-end running * Removing vestigial classes, cleaning up print statements * Adding SiLU / Swish to audio conformer feed forward block * Shifted Gemma3p5Audio naming prefix to Gemma3NanoAudio * Adding outputs to audio test * Fixes to padding in SSCP and 1D convolution, align RMS Norm with wider model * Update forward test to load from local weights * Update conversion to process / output audio layers * Update __all__ to export audio encoder * AutoModel registration for Gemma 3n Audio * Use AutoModel for ConditionalGeneration.audio_tower * Fixing input_proj_linear transpose * Fixing Gemma3NanoAudioConformerAttention.post conversion * Fixing Gemma3NanoAudioSSCPConvBlock.conv weights conversion * Correcting indentation issue on Gemma3p5RMSNorm --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Text + Vision Part 2 (huggingface#23) * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3p5.py * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Updating configs for the 2B variant in the conversion script * Using final generation config in conversion script --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Audio Integration (huggingface#12) * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma 3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding Gemma 3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemma3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3n * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * Converting sl.Frontend to FeatureExtractor * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3n.py * Update modular Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Draft of audio data in chat template * Removing image processing. Using SigLIP instead. * Audio input going end-to-end * Fixing dtype issues in audio encoder * x-lib formatting consistency * Adding example data * Save preprocessor_config.json from conversion script * Instrumentaiton for debugging * Additional instrumentation for preprocessing debugging * Updates to preprocessor, padding; produces correct end-to-end results on sample * Tackling configuraiton TODOs * Start of feature extractor refatcor * Adds Numpy version of USM extractor, removes Torch version and dependencies * Fixing AltUp.correct coef permute * Supporting batches of single audio segment inputs * Docstrings updates for config * In-lining audio feature extraction * Adjustments to conversion script and smoke test script --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: pculliton <phillipculliton@gmail.com> * Gemma 3n renaming * Removing test data and utilities * Renaming test files * Gemma 3n refactor * Fix tokenizer config in conversion script * Address reviewer feedback * FeatureExtractor returns float32 by default * Adding basic tests for audio, and input name for audio encoder * Audio integration test, updates to model_id for other integration tests * Use scales for q and k norms (huggingface#26) * Update audio integration test to use HF dataset * Reviewer feedback * Expand embedding table to full vocab size in weights conversion * Mix-n-match MatFormers for Gemma 3n (huggingface#25) * Remove in-place operations (huggingface#30) * chore: removing inplace ops * remove [tensor] * n pattern * chore: reviewer feedback in AudioEncoder and AltUp * More grad clipping * Dynamo compatibility * fix: cache slicing error * chore: simplify shared kv cache slicing * chore: vision encoder rename in timm * fix: image processor do_normalize=False * fixup: style * chore: model_doc * fix: docs for code quality * chore: repo consistency * fix: RMSNorm in float as in prior Gemmas * fix: per_layer_inputs = None * chore: Gemma3nForCausalLM from Gemma3nForConditionalGeneration checkpoint * chore: repo consistency * Add initial unit tests for Gemma3nAudioFeatureExtractor (huggingface#27) * Add initial unit tests for Gemma3nAudioFeatureExtractor * Add basic unit tests for Gemma3nProcessor (huggingface#28) Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * parameterize tests --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * chore: code style * fix: test cases * style and consistency * fix config in the test to be coherent with layer cache sharing * fix hidden states in tests and code * inits and mappings * fix modality prefixes * test order and prefixes * fix test exception * fix class order and reduce model size for faster tests * restore _checkpoint_conversion_mapping to load Caual from Conditional * fix config mapping! * fix: reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> * fix import test * add model args * auto_docstring * replace test path * consistency * skip tests for now * fix docstring for doc builder * skip unused attr --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>
zaristei
pushed a commit
to zaristei/transformers
that referenced
this pull request
Sep 9, 2025
* Gemma 3n * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3p5 overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * regenerating modeling file after syncing to HEAD * Use torch.std(..., unbiased=False) for activation sparsity (huggingface#8) * Refactoring to a single QVK Norm (huggingface#13) * AltUp: support scale_corrected_output (huggingface#14) * Converts einsums to nn.Linear (huggingface#7) * Converts einsums to nn.Linear * Removing unused variables * Aligning SharedKVCache with HybridCache (huggingface#11) * Alinging SharedKVStore with HybridCache * Remove KVStore. Refactor apply_rotary_pos_emb for sharing * Addressing review comments * Supporting split modality embeddings in Gemma3n (huggingface#10) * Adding the Embedder class * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, adding audio embedding layers, integrating embedder with the remaining architecture, adding a forward method for conditional generation * Apply suggestions from code review Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, prop drilling audio and vision configs to the text config * Removing TODO's that have been addressed * Simplify Embedder init and add audio embeddings * Embeddings refactor. Adds Gemma3nAudioEmbedder and Gemma3nVisionEmbedder * Refactoring vision and audio embeddings into ConditionalGeneration model --------- Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating attention mask for Gemma 3.5 (huggingface#15) * xxx_token_index to xxx_token_id * remvoing deprecated last_cache_position * Removing references to SigLIP * Always init per-layer inputs * Using torch.finfo().min for epsilon_tensor * Gemma3nDecoderLayer inherits from Gemma3DecoderLayer. Remove gating lambdas * fix modular GEMMA3N_INPUTS_DOCSTRING * Gemma3nAttention inherits from Gemma3Attention * Modular inheritance fixes * CausalLM conversion script for 4B model (huggingface#16) * Add Gemma3n Audio Encoder (huggingface#6) * initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3.5 (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right Gemma 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * CausalLM conversion script for 4B model * inv_timescales to non-persistent buffer * Addressing audio encoder Attention feedback * Addressing Gemma3nAudioSSCPConvBlock feedback * Addressing Gemma3nAudioConformerAttention feedback * Addressing padding feedback * Weights conversion loads audio state dict * Always use vision_config so saving works * Token id updates for configs * Stubs for interleaving audio embs * Addressing reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Fixing cache access error * Removing duplicate code from a bad merge * Gemma 3n Text + Vision Part 1 (huggingface#17) * testing utilities for numerics comparisons * Corrected einsum to nn.Linear weights conversion * Inherit scaled word embs from Gemma3 not Bart * Fixing transposes for collapsed linears * More transpose fixes * numpy api fix * RMSNorm: Explicit kwargs, scale_shift=0.0 when with_scale=True * Force AltUp to float32 * Updating debugging script for AudioEncoder debugging * Support divide_weight_by_sqrt_fan_in from JAX for per-layer inputs * Correcting attention einsum conversions * RMSNorm in type of x * Fixing douplicate laurel norm/gating * KV sharing using the right previous indices * Refactor kv shared index computation. Correct frac_shared_layers * Use num_shared_layers instead of inferring from a fraction * fixing a bug for logging * Fix shared data_ptrs in altup inits * rope: adjust proj -> norm -> rope to preserve computation (huggingface#20) * rope: adjust proj -> norm -> rope to preserve computation * Removing some breaking language model fluff in ConditionalGeneration * Consolidate query_states transforms --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Vectorize the loops in AltUp (huggingface#19) * Vectorize the loops in AltUp * fix typo * Expanding to support batched inputs * remove extra debug script * Fix AltUp.forward --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Add 'scale_shift=0.0, with_scale=True' to the final norm in TextModel * Convert norm to 1/sqrt (huggingface#21) * Convert norm to 1/sqrt * Scale shift change per Phil's rec * Adding default activation sparsity * Fixing 2B config in weights conversion script * Fixing RMSNorm parameters - adding scale_shift and with_scale * Correcting query pre-attention scaling * Adding query_rescale_scalar to text config * Adding layer_idx to MLP * Permafix for input_layernorm * Use 1/sqrt instead of rsqrt in DecoderLayer * Fix o_proj conversion * Conversion script update for vision encoder * Removing logging for debugging timm model * Fixing bugs in Gemma3nForConditionalGeneration for text generation * Generating the modeling_gemma3n.py file * Removing the addition of an erroneous line in the modeling file * Adding gemma3n text model to modeling_auto * Bugfix: Updating the interleaving of inputs_embeds and vision_embeds * Updating the modeling file with the latest bugfix changes * Updating models/auto for Gemma 3n * using AutoTokenizer in forward test * Adding processing_gemma3n.py * Gemma 3n configured for AutoModel. Conversion script updated. * Removing errant merge artifacts --------- Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Removing errant debugging statements from Gemma 3 * Gemma3n audio model (huggingface#18) * testing utilities for numerics comparisons * Implement CumulativeGroupNorm and add to SubSampleConvProjection and SSCPConvBlock * Add audio version of forward script based on RyanMullins' implementation * Updating to match encoder tests. WIP: config question needs resolving * Updates to audio classes to enable end-to-end running * Removing vestigial classes, cleaning up print statements * Adding SiLU / Swish to audio conformer feed forward block * Shifted Gemma3p5Audio naming prefix to Gemma3NanoAudio * Adding outputs to audio test * Fixes to padding in SSCP and 1D convolution, align RMS Norm with wider model * Update forward test to load from local weights * Update conversion to process / output audio layers * Update __all__ to export audio encoder * AutoModel registration for Gemma 3n Audio * Use AutoModel for ConditionalGeneration.audio_tower * Fixing input_proj_linear transpose * Fixing Gemma3NanoAudioConformerAttention.post conversion * Fixing Gemma3NanoAudioSSCPConvBlock.conv weights conversion * Correcting indentation issue on Gemma3p5RMSNorm --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Text + Vision Part 2 (huggingface#23) * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3p5.py * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Updating configs for the 2B variant in the conversion script * Using final generation config in conversion script --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Audio Integration (huggingface#12) * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma 3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding Gemma 3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemma3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3n * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * Converting sl.Frontend to FeatureExtractor * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3n.py * Update modular Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Draft of audio data in chat template * Removing image processing. Using SigLIP instead. * Audio input going end-to-end * Fixing dtype issues in audio encoder * x-lib formatting consistency * Adding example data * Save preprocessor_config.json from conversion script * Instrumentaiton for debugging * Additional instrumentation for preprocessing debugging * Updates to preprocessor, padding; produces correct end-to-end results on sample * Tackling configuraiton TODOs * Start of feature extractor refatcor * Adds Numpy version of USM extractor, removes Torch version and dependencies * Fixing AltUp.correct coef permute * Supporting batches of single audio segment inputs * Docstrings updates for config * In-lining audio feature extraction * Adjustments to conversion script and smoke test script --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: pculliton <phillipculliton@gmail.com> * Gemma 3n renaming * Removing test data and utilities * Renaming test files * Gemma 3n refactor * Fix tokenizer config in conversion script * Address reviewer feedback * FeatureExtractor returns float32 by default * Adding basic tests for audio, and input name for audio encoder * Audio integration test, updates to model_id for other integration tests * Use scales for q and k norms (huggingface#26) * Update audio integration test to use HF dataset * Reviewer feedback * Expand embedding table to full vocab size in weights conversion * Mix-n-match MatFormers for Gemma 3n (huggingface#25) * Remove in-place operations (huggingface#30) * chore: removing inplace ops * remove [tensor] * n pattern * chore: reviewer feedback in AudioEncoder and AltUp * More grad clipping * Dynamo compatibility * fix: cache slicing error * chore: simplify shared kv cache slicing * chore: vision encoder rename in timm * fix: image processor do_normalize=False * fixup: style * chore: model_doc * fix: docs for code quality * chore: repo consistency * fix: RMSNorm in float as in prior Gemmas * fix: per_layer_inputs = None * chore: Gemma3nForCausalLM from Gemma3nForConditionalGeneration checkpoint * chore: repo consistency * Add initial unit tests for Gemma3nAudioFeatureExtractor (huggingface#27) * Add initial unit tests for Gemma3nAudioFeatureExtractor * Add basic unit tests for Gemma3nProcessor (huggingface#28) Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * parameterize tests --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * chore: code style * fix: test cases * style and consistency * fix config in the test to be coherent with layer cache sharing * fix hidden states in tests and code * inits and mappings * fix modality prefixes * test order and prefixes * fix test exception * fix class order and reduce model size for faster tests * restore _checkpoint_conversion_mapping to load Caual from Conditional * fix config mapping! * fix: reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> * fix import test * add model args * auto_docstring * replace test path * consistency * skip tests for now * fix docstring for doc builder * skip unused attr --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>
zaristei
pushed a commit
to zaristei/transformers
that referenced
this pull request
Sep 9, 2025
* Gemma 3n * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3p5 overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * regenerating modeling file after syncing to HEAD * Use torch.std(..., unbiased=False) for activation sparsity (huggingface#8) * Refactoring to a single QVK Norm (huggingface#13) * AltUp: support scale_corrected_output (huggingface#14) * Converts einsums to nn.Linear (huggingface#7) * Converts einsums to nn.Linear * Removing unused variables * Aligning SharedKVCache with HybridCache (huggingface#11) * Alinging SharedKVStore with HybridCache * Remove KVStore. Refactor apply_rotary_pos_emb for sharing * Addressing review comments * Supporting split modality embeddings in Gemma3n (huggingface#10) * Adding the Embedder class * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, adding audio embedding layers, integrating embedder with the remaining architecture, adding a forward method for conditional generation * Apply suggestions from code review Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, prop drilling audio and vision configs to the text config * Removing TODO's that have been addressed * Simplify Embedder init and add audio embeddings * Embeddings refactor. Adds Gemma3nAudioEmbedder and Gemma3nVisionEmbedder * Refactoring vision and audio embeddings into ConditionalGeneration model --------- Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating attention mask for Gemma 3.5 (huggingface#15) * xxx_token_index to xxx_token_id * remvoing deprecated last_cache_position * Removing references to SigLIP * Always init per-layer inputs * Using torch.finfo().min for epsilon_tensor * Gemma3nDecoderLayer inherits from Gemma3DecoderLayer. Remove gating lambdas * fix modular GEMMA3N_INPUTS_DOCSTRING * Gemma3nAttention inherits from Gemma3Attention * Modular inheritance fixes * CausalLM conversion script for 4B model (huggingface#16) * Add Gemma3n Audio Encoder (huggingface#6) * initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3.5 (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right Gemma 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * CausalLM conversion script for 4B model * inv_timescales to non-persistent buffer * Addressing audio encoder Attention feedback * Addressing Gemma3nAudioSSCPConvBlock feedback * Addressing Gemma3nAudioConformerAttention feedback * Addressing padding feedback * Weights conversion loads audio state dict * Always use vision_config so saving works * Token id updates for configs * Stubs for interleaving audio embs * Addressing reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Fixing cache access error * Removing duplicate code from a bad merge * Gemma 3n Text + Vision Part 1 (huggingface#17) * testing utilities for numerics comparisons * Corrected einsum to nn.Linear weights conversion * Inherit scaled word embs from Gemma3 not Bart * Fixing transposes for collapsed linears * More transpose fixes * numpy api fix * RMSNorm: Explicit kwargs, scale_shift=0.0 when with_scale=True * Force AltUp to float32 * Updating debugging script for AudioEncoder debugging * Support divide_weight_by_sqrt_fan_in from JAX for per-layer inputs * Correcting attention einsum conversions * RMSNorm in type of x * Fixing douplicate laurel norm/gating * KV sharing using the right previous indices * Refactor kv shared index computation. Correct frac_shared_layers * Use num_shared_layers instead of inferring from a fraction * fixing a bug for logging * Fix shared data_ptrs in altup inits * rope: adjust proj -> norm -> rope to preserve computation (huggingface#20) * rope: adjust proj -> norm -> rope to preserve computation * Removing some breaking language model fluff in ConditionalGeneration * Consolidate query_states transforms --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Vectorize the loops in AltUp (huggingface#19) * Vectorize the loops in AltUp * fix typo * Expanding to support batched inputs * remove extra debug script * Fix AltUp.forward --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Add 'scale_shift=0.0, with_scale=True' to the final norm in TextModel * Convert norm to 1/sqrt (huggingface#21) * Convert norm to 1/sqrt * Scale shift change per Phil's rec * Adding default activation sparsity * Fixing 2B config in weights conversion script * Fixing RMSNorm parameters - adding scale_shift and with_scale * Correcting query pre-attention scaling * Adding query_rescale_scalar to text config * Adding layer_idx to MLP * Permafix for input_layernorm * Use 1/sqrt instead of rsqrt in DecoderLayer * Fix o_proj conversion * Conversion script update for vision encoder * Removing logging for debugging timm model * Fixing bugs in Gemma3nForConditionalGeneration for text generation * Generating the modeling_gemma3n.py file * Removing the addition of an erroneous line in the modeling file * Adding gemma3n text model to modeling_auto * Bugfix: Updating the interleaving of inputs_embeds and vision_embeds * Updating the modeling file with the latest bugfix changes * Updating models/auto for Gemma 3n * using AutoTokenizer in forward test * Adding processing_gemma3n.py * Gemma 3n configured for AutoModel. Conversion script updated. * Removing errant merge artifacts --------- Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Removing errant debugging statements from Gemma 3 * Gemma3n audio model (huggingface#18) * testing utilities for numerics comparisons * Implement CumulativeGroupNorm and add to SubSampleConvProjection and SSCPConvBlock * Add audio version of forward script based on RyanMullins' implementation * Updating to match encoder tests. WIP: config question needs resolving * Updates to audio classes to enable end-to-end running * Removing vestigial classes, cleaning up print statements * Adding SiLU / Swish to audio conformer feed forward block * Shifted Gemma3p5Audio naming prefix to Gemma3NanoAudio * Adding outputs to audio test * Fixes to padding in SSCP and 1D convolution, align RMS Norm with wider model * Update forward test to load from local weights * Update conversion to process / output audio layers * Update __all__ to export audio encoder * AutoModel registration for Gemma 3n Audio * Use AutoModel for ConditionalGeneration.audio_tower * Fixing input_proj_linear transpose * Fixing Gemma3NanoAudioConformerAttention.post conversion * Fixing Gemma3NanoAudioSSCPConvBlock.conv weights conversion * Correcting indentation issue on Gemma3p5RMSNorm --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Text + Vision Part 2 (huggingface#23) * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3p5.py * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Updating configs for the 2B variant in the conversion script * Using final generation config in conversion script --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Audio Integration (huggingface#12) * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma 3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding Gemma 3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemma3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3n * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * Converting sl.Frontend to FeatureExtractor * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3n.py * Update modular Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Draft of audio data in chat template * Removing image processing. Using SigLIP instead. * Audio input going end-to-end * Fixing dtype issues in audio encoder * x-lib formatting consistency * Adding example data * Save preprocessor_config.json from conversion script * Instrumentaiton for debugging * Additional instrumentation for preprocessing debugging * Updates to preprocessor, padding; produces correct end-to-end results on sample * Tackling configuraiton TODOs * Start of feature extractor refatcor * Adds Numpy version of USM extractor, removes Torch version and dependencies * Fixing AltUp.correct coef permute * Supporting batches of single audio segment inputs * Docstrings updates for config * In-lining audio feature extraction * Adjustments to conversion script and smoke test script --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: pculliton <phillipculliton@gmail.com> * Gemma 3n renaming * Removing test data and utilities * Renaming test files * Gemma 3n refactor * Fix tokenizer config in conversion script * Address reviewer feedback * FeatureExtractor returns float32 by default * Adding basic tests for audio, and input name for audio encoder * Audio integration test, updates to model_id for other integration tests * Use scales for q and k norms (huggingface#26) * Update audio integration test to use HF dataset * Reviewer feedback * Expand embedding table to full vocab size in weights conversion * Mix-n-match MatFormers for Gemma 3n (huggingface#25) * Remove in-place operations (huggingface#30) * chore: removing inplace ops * remove [tensor] * n pattern * chore: reviewer feedback in AudioEncoder and AltUp * More grad clipping * Dynamo compatibility * fix: cache slicing error * chore: simplify shared kv cache slicing * chore: vision encoder rename in timm * fix: image processor do_normalize=False * fixup: style * chore: model_doc * fix: docs for code quality * chore: repo consistency * fix: RMSNorm in float as in prior Gemmas * fix: per_layer_inputs = None * chore: Gemma3nForCausalLM from Gemma3nForConditionalGeneration checkpoint * chore: repo consistency * Add initial unit tests for Gemma3nAudioFeatureExtractor (huggingface#27) * Add initial unit tests for Gemma3nAudioFeatureExtractor * Add basic unit tests for Gemma3nProcessor (huggingface#28) Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * parameterize tests --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * chore: code style * fix: test cases * style and consistency * fix config in the test to be coherent with layer cache sharing * fix hidden states in tests and code * inits and mappings * fix modality prefixes * test order and prefixes * fix test exception * fix class order and reduce model size for faster tests * restore _checkpoint_conversion_mapping to load Caual from Conditional * fix config mapping! * fix: reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> * fix import test * add model args * auto_docstring * replace test path * consistency * skip tests for now * fix docstring for doc builder * skip unused attr --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>
zaristei
pushed a commit
to zaristei/transformers
that referenced
this pull request
Sep 9, 2025
* Gemma 3n * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3p5RMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3p5 overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3p5 text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * regenerating modeling file after syncing to HEAD * Use torch.std(..., unbiased=False) for activation sparsity (huggingface#8) * Refactoring to a single QVK Norm (huggingface#13) * AltUp: support scale_corrected_output (huggingface#14) * Converts einsums to nn.Linear (huggingface#7) * Converts einsums to nn.Linear * Removing unused variables * Aligning SharedKVCache with HybridCache (huggingface#11) * Alinging SharedKVStore with HybridCache * Remove KVStore. Refactor apply_rotary_pos_emb for sharing * Addressing review comments * Supporting split modality embeddings in Gemma3n (huggingface#10) * Adding the Embedder class * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, adding audio embedding layers, integrating embedder with the remaining architecture, adding a forward method for conditional generation * Apply suggestions from code review Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Update modular Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> * Addressing review comments, prop drilling audio and vision configs to the text config * Removing TODO's that have been addressed * Simplify Embedder init and add audio embeddings * Embeddings refactor. Adds Gemma3nAudioEmbedder and Gemma3nVisionEmbedder * Refactoring vision and audio embeddings into ConditionalGeneration model --------- Co-authored-by: Ryan Mullins <ryan@ryanmullins.org> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating attention mask for Gemma 3.5 (huggingface#15) * xxx_token_index to xxx_token_id * remvoing deprecated last_cache_position * Removing references to SigLIP * Always init per-layer inputs * Using torch.finfo().min for epsilon_tensor * Gemma3nDecoderLayer inherits from Gemma3DecoderLayer. Remove gating lambdas * fix modular GEMMA3N_INPUTS_DOCSTRING * Gemma3nAttention inherits from Gemma3Attention * Modular inheritance fixes * CausalLM conversion script for 4B model (huggingface#16) * Add Gemma3n Audio Encoder (huggingface#6) * initial commit of Gemma 3.5 scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding gemma3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3.5 (huggingface#3) * Initial Gemm3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3.5 * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right Gemma 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3.5 * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * CausalLM conversion script for 4B model * inv_timescales to non-persistent buffer * Addressing audio encoder Attention feedback * Addressing Gemma3nAudioSSCPConvBlock feedback * Addressing Gemma3nAudioConformerAttention feedback * Addressing padding feedback * Weights conversion loads audio state dict * Always use vision_config so saving works * Token id updates for configs * Stubs for interleaving audio embs * Addressing reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Fixing cache access error * Removing duplicate code from a bad merge * Gemma 3n Text + Vision Part 1 (huggingface#17) * testing utilities for numerics comparisons * Corrected einsum to nn.Linear weights conversion * Inherit scaled word embs from Gemma3 not Bart * Fixing transposes for collapsed linears * More transpose fixes * numpy api fix * RMSNorm: Explicit kwargs, scale_shift=0.0 when with_scale=True * Force AltUp to float32 * Updating debugging script for AudioEncoder debugging * Support divide_weight_by_sqrt_fan_in from JAX for per-layer inputs * Correcting attention einsum conversions * RMSNorm in type of x * Fixing douplicate laurel norm/gating * KV sharing using the right previous indices * Refactor kv shared index computation. Correct frac_shared_layers * Use num_shared_layers instead of inferring from a fraction * fixing a bug for logging * Fix shared data_ptrs in altup inits * rope: adjust proj -> norm -> rope to preserve computation (huggingface#20) * rope: adjust proj -> norm -> rope to preserve computation * Removing some breaking language model fluff in ConditionalGeneration * Consolidate query_states transforms --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Vectorize the loops in AltUp (huggingface#19) * Vectorize the loops in AltUp * fix typo * Expanding to support batched inputs * remove extra debug script * Fix AltUp.forward --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Add 'scale_shift=0.0, with_scale=True' to the final norm in TextModel * Convert norm to 1/sqrt (huggingface#21) * Convert norm to 1/sqrt * Scale shift change per Phil's rec * Adding default activation sparsity * Fixing 2B config in weights conversion script * Fixing RMSNorm parameters - adding scale_shift and with_scale * Correcting query pre-attention scaling * Adding query_rescale_scalar to text config * Adding layer_idx to MLP * Permafix for input_layernorm * Use 1/sqrt instead of rsqrt in DecoderLayer * Fix o_proj conversion * Conversion script update for vision encoder * Removing logging for debugging timm model * Fixing bugs in Gemma3nForConditionalGeneration for text generation * Generating the modeling_gemma3n.py file * Removing the addition of an erroneous line in the modeling file * Adding gemma3n text model to modeling_auto * Bugfix: Updating the interleaving of inputs_embeds and vision_embeds * Updating the modeling file with the latest bugfix changes * Updating models/auto for Gemma 3n * using AutoTokenizer in forward test * Adding processing_gemma3n.py * Gemma 3n configured for AutoModel. Conversion script updated. * Removing errant merge artifacts --------- Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> * Removing errant debugging statements from Gemma 3 * Gemma3n audio model (huggingface#18) * testing utilities for numerics comparisons * Implement CumulativeGroupNorm and add to SubSampleConvProjection and SSCPConvBlock * Add audio version of forward script based on RyanMullins' implementation * Updating to match encoder tests. WIP: config question needs resolving * Updates to audio classes to enable end-to-end running * Removing vestigial classes, cleaning up print statements * Adding SiLU / Swish to audio conformer feed forward block * Shifted Gemma3p5Audio naming prefix to Gemma3NanoAudio * Adding outputs to audio test * Fixes to padding in SSCP and 1D convolution, align RMS Norm with wider model * Update forward test to load from local weights * Update conversion to process / output audio layers * Update __all__ to export audio encoder * AutoModel registration for Gemma 3n Audio * Use AutoModel for ConditionalGeneration.audio_tower * Fixing input_proj_linear transpose * Fixing Gemma3NanoAudioConformerAttention.post conversion * Fixing Gemma3NanoAudioSSCPConvBlock.conv weights conversion * Correcting indentation issue on Gemma3p5RMSNorm --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Text + Vision Part 2 (huggingface#23) * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3p5.py * Update src/transformers/models/gemma3p5/modular_gemma3p5.py Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Updating configs for the 2B variant in the conversion script * Using final generation config in conversion script --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Audio Integration (huggingface#12) * initial commit of Gemma 3n scaffold * Fixing param pass through on Gemm3nRMSNorm * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Adds AltUp to Gemma 3n * Adding Gemma 3n overall and text config with vision and audio config placeholders (huggingface#3) * Adding Gemma 3n text configs * Adding audio config placeholders * Adding a placeholder for vision configs * Updating MobileNetVisionConfig, inheriting TimmWrapperConfig * Updating text configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Removing altup configs to accept the suggested configs * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating altup config * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Addressing review comments and updating text configs * Adding a config for activation sparsity * Updating configs to pass through options to super class init and adjust some name prefixes * Updating laurel and altup with corrected config values * Normalizing sub_config initializers --------- Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Updating MLP with activation sparsity (huggingface#2) * Updating DecoderBlock for Gemma 3n (huggingface#3) * Initial Gemma3nTextModel (huggingface#4) NOTE: This implementation WILL CHANGE in the coming weeks, however, changes will be strictly additive and this will remain a suitable baseline for downstream implementations to reference. * Adding KV Cache Sharing * Adds Einsum layer to Gemma 3n * Updating EinsumLayer API * Refactored kv cache sharing in attention * Adding KVStore for cache sharing * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update modular Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Update src/transformers/cache_utils.py Co-authored-by: Ryan Mullins <ryanmullins@google.com> * Undoing erroneous force push * Reverting RMSNorm to with_scale by default * Adds LAuReL to Gemma 3n * Updating KV Cache Sharing implementation * Updating the q and k norm definitions in the attention module * Fixing name error for q,k,v RMS norm to use the right 3n module * Updating MLP with activation sparsity * Updating DecoderBlock for Gemma 3n * Updating kv cache sharing implementation with the use of a cache buffer and refactoring some lines of code * Isolating KV Cache logic to relevant components * Fixing logic error in Gemma3nAttention.forward * Refactoring caching contributions and fixing kv_store initialization * Simplifying Configs * Remove errant self from super init call * Bug fix in the Attention module - changing self.head_dim to config.head_dim * Bug fixes in the LaurelBlock and RMS Norm super init call * removing redundant code from a merge * Adding per_layer_inputs to TextModel * Adding preprocess embeddings with altup * Adds per-layer-to-single output and a host of TODOs * Integrating altup predict with the model workflow and other minor bug fixes * Using nn.Embedding temporarily for text model * It goes forward * Minor refactor of attention sparsity and RoPE initialization * Fixing duplicate rope_scaling param bug when loading from pretrained --------- Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Normalizing on altup_num_inputs config option * Adding audio encoder config * Adds high-level components for Audio Encoder * Implement uniform reducer for Audio Encoder * Adding placeholders for Conformer components in Audio Encoder * Adding placeholders for SubSampleConvProjection components in Audio Encoder * Adding SequenceLayer component placeholders * Implementing Gemma3nAudioEncoder with nn.Sequential * Implementing Gemma3nAudioSubSampleConvProjection with nn.Sequential * Implementing Conformer model with SequenceLayers * Use OrderedDict in nn.Sequential initializers * Implements sl.Residual in Torch with nn.Sequential and OrderedDict * Adopting a base SequenceLayer class with default forward() method * Implementing sl.GatedLinearUnit in Torch * Implementing sl.Swish in Torch * Implementing sl.ReLU in Torch * Implementing sl.Scale in Torch * Removing sl.Dropout after tree-shaking * Implementing sl.RMSNorm in Torch with fake shape * Implementing sl.GroupNorm in Torch * Implementing sl.Conv2d in Torch * Implementing sl.Dense in Torch * Removing sl.Delay layers, which act as pass-throughs * Connecting shapes to configs in initializers * Removing sl.Emit * Implementing sl.ExpandDims in Torch * Adding sl.GradientClipping to Torch * Implementing sl.DenseShaped in Torch * Implementing sl.LDPA in Torch * Removing unused sl.CombinedQKVProj class * Fixing erroneous type hint * Implemnenting sl.DepthwiseConv1D in Torch * Implementing sl.MaskInvalid in Torch * Fixes for initialization * Fixes for saving weights * Removing einsums per feedback from HF staff * Removing Sequence Layers idioms from audio encoder * Fixes for reviewer comments * Converting sl.Frontend to FeatureExtractor * Updates for ConditionalGeneration.get_image_features * Adding a WIP draft of image_processing_gemma3n.py * Update modular Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> * Modular conversion after github suggested change * Text + image gives good results * Fixing image size preset * Draft of audio data in chat template * Removing image processing. Using SigLIP instead. * Audio input going end-to-end * Fixing dtype issues in audio encoder * x-lib formatting consistency * Adding example data * Save preprocessor_config.json from conversion script * Instrumentaiton for debugging * Additional instrumentation for preprocessing debugging * Updates to preprocessor, padding; produces correct end-to-end results on sample * Tackling configuraiton TODOs * Start of feature extractor refatcor * Adds Numpy version of USM extractor, removes Torch version and dependencies * Fixing AltUp.correct coef permute * Supporting batches of single audio segment inputs * Docstrings updates for config * In-lining audio feature extraction * Adjustments to conversion script and smoke test script --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: pculliton <phillipculliton@gmail.com> * Gemma 3n renaming * Removing test data and utilities * Renaming test files * Gemma 3n refactor * Fix tokenizer config in conversion script * Address reviewer feedback * FeatureExtractor returns float32 by default * Adding basic tests for audio, and input name for audio encoder * Audio integration test, updates to model_id for other integration tests * Use scales for q and k norms (huggingface#26) * Update audio integration test to use HF dataset * Reviewer feedback * Expand embedding table to full vocab size in weights conversion * Mix-n-match MatFormers for Gemma 3n (huggingface#25) * Remove in-place operations (huggingface#30) * chore: removing inplace ops * remove [tensor] * n pattern * chore: reviewer feedback in AudioEncoder and AltUp * More grad clipping * Dynamo compatibility * fix: cache slicing error * chore: simplify shared kv cache slicing * chore: vision encoder rename in timm * fix: image processor do_normalize=False * fixup: style * chore: model_doc * fix: docs for code quality * chore: repo consistency * fix: RMSNorm in float as in prior Gemmas * fix: per_layer_inputs = None * chore: Gemma3nForCausalLM from Gemma3nForConditionalGeneration checkpoint * chore: repo consistency * Add initial unit tests for Gemma3nAudioFeatureExtractor (huggingface#27) * Add initial unit tests for Gemma3nAudioFeatureExtractor * Add basic unit tests for Gemma3nProcessor (huggingface#28) Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * parameterize tests --------- Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> * chore: code style * fix: test cases * style and consistency * fix config in the test to be coherent with layer cache sharing * fix hidden states in tests and code * inits and mappings * fix modality prefixes * test order and prefixes * fix test exception * fix class order and reduce model size for faster tests * restore _checkpoint_conversion_mapping to load Caual from Conditional * fix config mapping! * fix: reviewer feedback --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> * fix import test * add model args * auto_docstring * replace test path * consistency * skip tests for now * fix docstring for doc builder * skip unused attr --------- Co-authored-by: SindhuRaghuram97 <114270661+SindhuRaghuram97@users.noreply.github.com> Co-authored-by: Sindhu Raghuram <sindhuraghuram@google.com> Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Mayank Chaturvedi <imayank@google.com> Co-authored-by: Douglas Reid <douglas-reid@users.noreply.github.com> Co-authored-by: Douglas Reid <21148125+douglas-reid@users.noreply.github.com> Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> Co-authored-by: pculliton <phillipculliton@gmail.com> Co-authored-by: Aritra Roy Gosthipaty <aritra.born2fly@gmail.com> Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com> Co-authored-by: Arthur <arthur.zucker@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Should say
BertForQuestionAnswering, but saysBertForSequenceClassification.