Skip to content

Revert "Fuse routed scaling factor in topk_reduce kernel (#6220)" #6968

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Jun 8, 2025
Merged

Revert "Fuse routed scaling factor in topk_reduce kernel (#6220)" #6968

merged 1 commit into from
Jun 8, 2025

Conversation

zhyncs
Copy link
Member

@zhyncs zhyncs commented Jun 8, 2025

This reverts commit 515ef4f.

Motivation

Modifications

Checklist

@zhyncs zhyncs merged commit 1fb76eb into main Jun 8, 2025
0 of 37 checks passed
@zhyncs zhyncs deleted the zhyncs/r branch June 8, 2025 04:02
gemini-code-assist[bot]
gemini-code-assist bot reviewed Jun 8, 2025
View reviewed changes
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hello @zhyncs, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

Summary of Changes

Hello! Gemini here, providing a summary of this pull request. This PR is a revert of commit 515ef4facbc89cd7c093c198386a8817fce856d6. The original commit aimed to fuse the routed_scaling_factor into the topk_reduce kernel within the fused MoE implementation. This revert removes that fusion, specifically pulling the scaling factor out of the MoE kernel logic and applying it separately in the model's forward pass for the DeepseekV2 model. It also removes the custom Triton and torch.compile kernels previously used for the sum reduction step when topk > 2, replacing them with a standard torch.sum operation.

Highlights

  • Revert Fusion: Reverts the previous change that fused the routed_scaling_factor into the MoE kernel.
  • Scaling Factor Location: Moves the application of the routed_scaling_factor from within the fused MoE kernel logic back to the model's forward pass (specifically for DeepseekV2).
  • Sum Reduction Kernel Removal: Removes the custom Triton kernel (_moe_sum_reduce_kernel) and its Python wrapper (moe_sum_reduce_triton), as well as the moe_sum_reduce_torch_compile function, which were used for the sum reduction step.
  • Standard Sum Reduction: Replaces the custom sum reduction kernels with a standard torch.sum call within the fused_experts_impl function when topk > 2.

Changelog

Click here to see the changelog
  • python/sglang/srt/layers/moe/fused_moe_triton/fused_moe.py
    • Removed the routed_scaling_factor parameter from inplace_fused_experts (L1158), inplace_fused_experts_fake (L1206), outplace_fused_experts (L1240), outplace_fused_experts_fake (L1289), fused_experts (L1324), and fused_experts_impl (L1492) function signatures.
    • Removed the custom Triton kernel _moe_sum_reduce_kernel (L1379-L1426) and its Python wrapper moe_sum_reduce_triton (L1428-L1461).
    • Removed the moe_sum_reduce_torch_compile function (L1464-L1467).
    • Modified fused_experts_impl (L1677-L1701) to remove the logic for applying routed_scaling_factor and to use torch.sum (L1582-L1586) for the sum reduction when topk_ids.shape[1] > 2, replacing the calls to the removed custom kernels.
  • python/sglang/srt/layers/moe/fused_moe_triton/layer.py
    • Removed the routed_scaling_factor parameter from the forward_cuda method signature (L228) and its call to fused_moe (L228).
  • python/sglang/srt/layers/quantization/blockwise_int8.py
    • Removed the routed_scaling_factor argument from the call to fused_moe in the apply method (L414).
  • python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py
    • Removed the routed_scaling_factor argument from the call to fused_moe in the apply method (L320).
  • python/sglang/srt/layers/quantization/fp8.py
    • Removed the routed_scaling_factor argument from the call to fused_moe in the apply method (L1033).
  • python/sglang/srt/layers/quantization/moe_wna16.py
    • Removed the routed_scaling_factor argument from the call to fused_moe in the apply method (L391).
  • python/sglang/srt/layers/quantization/w8a8_fp8.py
    • Removed the routed_scaling_factor argument from the call to fused_moe in the apply method (L331).
  • python/sglang/srt/layers/quantization/w8a8_int8.py
    • Removed the routed_scaling_factor argument from the call to fused_moe in the apply method (L271).
  • python/sglang/srt/models/deepseek_v2.py
    • Added final_hidden_states *= self.routed_scaling_factor (L349) in the forward_normal method to apply the scaling factor after the expert outputs are combined.
Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

Feature Command Description
Code Review /gemini review Performs a code review for the current pull request in its current state.
Pull Request Summary /gemini summary Provides a summary of the current pull request in its current state.
Comment @gemini-code-assist Responds in comments when explicitly tagged, both in issue comments and review comments.
Help /gemini help Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist is currently in preview and may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments to provide feedback.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

  1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist[bot]
gemini-code-assist bot suggested changes Jun 8, 2025
View reviewed changes
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request successfully reverts the commit that fused the routed_scaling_factor into the topk_reduce kernel. The changes correctly remove the parameter and its usage from most of the affected functions and classes related to fused MoE experts. However, there appears to be an inconsistency in how the routed_scaling_factor is handled for fused shared experts, which could lead to incorrect results. Additionally, the removal of specialized kernels might impact performance.

Summary of Findings

  • Inconsistent routed_scaling_factor handling for fused shared experts: The routed_scaling_factor is now applied after the expert computation in deepseek_v2.py, but the calculation of the fused shared expert's weight in topk.py (which was not reverted) still divides by this factor. This inconsistency likely leads to incorrect results for models using fused shared experts.
  • Removal of specialized moe_sum_reduce kernels: The Triton and torch.compile kernels for summing expert outputs (moe_sum_reduce) were removed entirely, falling back to torch.sum. This might cause a performance regression for topk > 2 compared to the previous implementation.

Merge Readiness

This pull request reverts the intended change but introduces a potential correctness issue related to the routed_scaling_factor and fused shared experts due to incomplete reversion of related logic in topk.py. There is also a potential performance regression due to kernel removal. I recommend addressing the high-severity correctness issue before merging. I am unable to approve this pull request, and other reviewers should assess the changes and decide on the merge readiness.

python/sglang/srt/models/deepseek_v2.py
@@ -346,7 +346,7 @@ def forward_normal(self, hidden_states: torch.Tensor) -> torch.Tensor:
final_hidden_states = self.experts(
hidden_states=hidden_states, router_logits=router_logits
)

final_hidden_states *= self.routed_scaling_factor
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

high

The routed_scaling_factor is now applied after the expert computation and combination. This seems inconsistent with the logic in select_experts (which was not reverted), where the weight for the fused shared expert is calculated as topk_weights[:, :-1].sum(dim=-1) / routed_scaling_factor. This calculation in select_experts seems designed to counteract a scaling that was previously applied inside the expert computation. With the scaling now applied after the expert computation, this calculation in select_experts is likely incorrect and could lead to wrong results for models using fused shared experts. Could you please clarify if the logic in select_experts should also have been reverted or adjusted?

python/sglang/srt/layers/moe/fused_moe_triton/fused_moe.py
Comment on lines +1581 to +1586
elif topk_ids.shape[1] > 2:
torch.sum(
intermediate_cache3.view(*intermediate_cache3.shape),
dim=1,
out=out_hidden_states[begin_chunk_idx:end_chunk_idx],
)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

The specialized Triton and torch.compile kernels for moe_sum_reduce have been removed, and the code now falls back to torch.sum for topk > 2. While torch.sum is functionally correct, the removed kernels might have provided performance optimizations, especially for larger topk values. Was there a reason for removing these kernels entirely instead of just removing the routed_scaling_factor usage within them? This change might introduce a performance regression.

jianan-gu pushed a commit to jianan-gu/sglang that referenced this pull request Jun 12, 2025
@zhyncs @jianan-gu
Revert "Fuse routed scaling factor in topk_reduce kernel (sgl-project…
eb87dbd 
…#6220)" (sgl-project#6968)
walker-ai pushed a commit to walker-ai/sglang that referenced this pull request Jul 8, 2025
@yudian0504
PullRequest: 52 sgl_20250610_sync_tag047
a19e2e2 
Merge branch 'sgl_20250610_sync_tag047 of git@code.alipay.com:Theta/SGLang.git into main

https://code.alipay.com/Theta/SGLang/pull_requests/52


Reviewed-by: 剑川 <jianchuan.gys@antgroup.com>


* [Bugfix] Fix slice operation when chunk size mismatch (sgl-project#6697)
* [Bugfix] Fix ChatCompletion endpoint of mini_lb when stream is set (sgl-project#6703)
* [CI] Fix setup of disaggregation with different tp (sgl-project#6706)
* [PD] Remove Unnecessary Exception Handling for FastQueue.get() (sgl-project#6712)
* Fuse routed_scaling_factor in DeepSeek (sgl-project#6710)
* Overlap two kernels in DeepSeek with communication (sgl-project#6711)
* Minor refactor two-batch overlap (sgl-project#6682)
* Speed up when having padding tokens two-batch overlap (sgl-project#6668)
* [Feature] Support Flashinfer fp8 blockwise GEMM kernel on Blackwell (sgl-project#6479)
* Fix LoRA bench (sgl-project#6719)
* temp
* Fix PP for Qwen3 MoE (sgl-project#6709)
* [feat] triton kernel for get_last_loc (sgl-project#6676)
* [fix] more mem for draft_extend cuda_graph (sgl-project#6726)
* [PD] bug fix:  Update status if nixl receiver send a a dummy req. (sgl-project#6720)
* Tune memory arguments on B200 (sgl-project#6718)
* Add DeepSeek-R1-0528 function call chat template (sgl-project#6725)
* refactor(tool call): Fix BaseFormatDetector tool_index issue and refactor `parse_streaming_increment` (sgl-project#6715)
* Add draft extend CUDA graph for Triton backend (sgl-project#6705)
* refactor apply_w8a8_block_fp8_linear in fp (sgl-project#6545)
* [PD] Support completion endpoint (sgl-project#6729)
* PD Rust LB (PO2) (sgl-project#6437)
* Super tiny enable sole usage of expert distribution metrics and update doc (sgl-project#6680)
* Support picking variants of EPLB algorithms (sgl-project#6728)
* Support tuning DeepEP configs (sgl-project#6742)
* [test] add ut and bm for get_last_loc (sgl-project#6746)
* Fix mem_fraction_static for AMD CI (sgl-project#6748)
* [fix][RL] Fix DeepSeekV3ForCausalLM.post_load_weights for multiple update weight (sgl-project#6265)
* Improve EPLB logical to physical dispatch map (sgl-project#6727)
* Update DeepSeek-R1-0528 function call chat template (sgl-project#6765)
* [PD] Optimize time out logic and add env var doc for mooncake (sgl-project#6761)
* Fix aiohttp 'Chunk too big' in bench_serving (sgl-project#6737)
* Support sliding window in triton backend (sgl-project#6509)
* Fix shared experts fusion error (sgl-project#6289)
* Fix one bug in the grouped-gemm triton kernel (sgl-project#6772)
* update llama4 chat template and pythonic parser (sgl-project#6679)
* feat(tool call): Enhance Llama32Detector for improved JSON parsing in non-stream (sgl-project#6784)
* Support token-level quantization for EP MoE (sgl-project#6782)
* Temporarily lower mmlu threshold for triton sliding window backend (sgl-project#6785)
* ci: relax test_function_call_required (sgl-project#6786)
* Add intel_amx backend for Radix Attention for CPU (sgl-project#6408)
* Fix incorrect LoRA weight loading for fused gate_up_proj (sgl-project#6734)
* fix(PD-disaggregation): Can not get local ip (sgl-project#6792)
* [FIX] mmmu bench serving result display error (sgl-project#6525) (sgl-project#6791)
* Bump torch to 2.7.0 (sgl-project#6788)
* chore: bump sgl-kernel v0.1.5 (sgl-project#6794)
* Improve profiler and integrate profiler in bench_one_batch_server (sgl-project#6787)
* chore: upgrade sgl-kernel v0.1.5 (sgl-project#6795)
* [Minor] Always append newline after image token when parsing chat message (sgl-project#6797)
* Update CI tests for Llama4 models (sgl-project#6421)
* [Feat] Enable PDL automatically on Hopper architecture (sgl-project#5981)
* chore: update blackwell docker (sgl-project#6800)
* misc: cache is_hopper_arch (sgl-project#6799)
* Remove contiguous before Flashinfer groupwise fp8 gemm (sgl-project#6804)
* Correctly abort the failed grammar requests & Improve the handling of abort (sgl-project#6803)
* [EP] Add cuda kernel for moe_ep_pre_reorder (sgl-project#6699)
* Add draft extend CUDA graph for flashinfer backend  (sgl-project#6805)
* Refactor CustomOp to avoid confusing bugs (sgl-project#5382)
* Tiny log prefill time (sgl-project#6780)
* Tiny fix EPLB assertion about rebalancing period and recorder window size (sgl-project#6813)
* Add simple utility to dump tensors for debugging (sgl-project#6815)
* Fix profiles do not have consistent names (sgl-project#6811)
* Speed up rebalancing when using non-static dispatch algorithms (sgl-project#6812)
* [1/2] Add Kernel support for Cutlass based Fused FP4 MoE (sgl-project#6093)
* [Router] Fix k8s Service Discovery (sgl-project#6766)
* Add CPU optimized kernels for topk and rope fusions  (sgl-project#6456)
* fix new_page_count_next_decode (sgl-project#6671)
* Fix wrong weight reference in dynamic EPLB (sgl-project#6818)
* Minor add metrics to expert location updater (sgl-project#6816)
* [Refactor] Rename `n_share_experts_fusion` as `num_fused_shared_experts` (sgl-project#6735)
* [FEAT] Add transformers backend support  (sgl-project#5929)
* [fix] recover auto-dispatch for rmsnorm and rope (sgl-project#6745)
* fix ep_moe_reorder kernel bugs (sgl-project#6858)
* [Refactor] Multimodal data processing for VLM (sgl-project#6659)
* Decoder-only Scoring API (sgl-project#6460)
* feat: add dp-rank to KV events (sgl-project#6852)
* Set `num_fused_shared_experts` as `num_shared_experts` when shared_experts fusion is not disabled (sgl-project#6736)
* Fix one missing arg in DeepEP (sgl-project#6878)
* Support LoRA in TestOpenAIVisionServer and fix fused kv_proj loading bug. (sgl-project#6861)
* support 1 shot allreduce  in 1-node and 2-node using mscclpp (sgl-project#6277)
* Fix Qwen3MoE missing token padding optimization (sgl-project#6820)
* Tiny update error hints (sgl-project#6846)
* Support layerwise rebalancing experts (sgl-project#6851)
* Tiny allow profiler API to auto create directory (sgl-project#6865)
* Support Blackwell DeepEP docker images (sgl-project#6868)
* [EP] Add cuda kernel for moe_ep_post_reorder (sgl-project#6837)
* [theta]merge 0605
* oai: fix openAI client error with single request via batch api (sgl-project#6170)
* [PD] Fix potential perf spike caused by tracker gc and optimize doc (sgl-project#6764)
* Use deepgemm instead of triton for fused_qkv_a_proj_with_mqa (sgl-project#6890)
* [CUTLASS-FP4-MOE]  Introduce CutlassMoEParams class for easy initialization of Cutlass Grouped Gems Metadata (sgl-project#6887)
* bugfix(OAI): Fix image_data processing for jinja chat templates (sgl-project#6877)
* [CPU] enable CI for PRs, add Dockerfile and auto build task (sgl-project#6458)
* AITER backend extension and workload optimizations (sgl-project#6838)
* [theta]merge
* [theta]merge
* [Feature] Support Flashinfer fmha on Blackwell (sgl-project#6930)
* Fix a bug in abort & Improve docstrings for abort (sgl-project#6931)
* Tiny support customize DeepEP max dispatch tokens per rank (sgl-project#6934)
* Sync the changes on cuda graph runners (sgl-project#6932)
* [PD] Optimize transfer queue forward logic for dummy rank (sgl-project#6922)
* [Refactor] image data process in bench_serving (sgl-project#6879)
* [fix] logical_to_all_physical_map index 256 is out of bounds in EP parallel. (sgl-project#6767)
* Add triton fused moe kernel config for E=257 on B200 (sgl-project#6939)
* [sgl-kernel] update deepgemm (sgl-project#6942)
* chore: bump sgl-kernel v0.1.6 (sgl-project#6943)
* Minor compile fused topk (sgl-project#6944)
* [Bugfix] pipeline parallelism and Eagle Qwen2 (sgl-project#6910)
* Tiny re-introduce profile id logging (sgl-project#6912)
* Add triton version as a fused_moe_triton config search key to avoid performace decrease in different Triton version (sgl-project#5955)
* reduce torch.zeros overhead in moe align block size kernel (sgl-project#6369)
* chore: upgrade sgl-kernel v0.1.6 (sgl-project#6945)
* add fbgemm moe grouped gemm kernel benchmark (sgl-project#6924)
* [Docker] Add docker file for SGL Router (sgl-project#6915)
* Disabling mixed chunked prefill when eagle is enabled (sgl-project#6874)
* Add canary for EPLB rebalancing (sgl-project#6895)
* Refactor global_server_args_dict (sgl-project#6866)
* Fuse routed scaling factor in topk_reduce kernel (sgl-project#6220)
* Update server timeout time in AMD CI. (sgl-project#6953)
* [misc] add is_cpu() (sgl-project#6950)
* Add H20 fused MoE kernel tuning configs for DeepSeek-R1/V3 (sgl-project#6885)
* Add a CUDA kernel for fusing mapping and weighted sum for MoE. (sgl-project#6916)
* chore: bump sgl-kernel v0.1.6.post1 (sgl-project#6955)
* chore: upgrade sgl-kernel v0.1.6.post1 (sgl-project#6957)
* [DeepseekR1-FP4] Add Support for nvidia/DeepSeekR1-FP4 model (sgl-project#6853)
* Revert "Fuse routed scaling factor in topk_reduce kernel (sgl-project#6220)" (sgl-project#6968)
* [AMD] Add more tests to per-commit-amd (sgl-project#6926)
* chore: bump sgl-kernel v0.1.7 (sgl-project#6963)
* Slightly improve the sampler to skip unnecessary steps (sgl-project#6956)
* rebase h20 fused_moe config (sgl-project#6966)
* Fix CI and triton moe Configs (sgl-project#6974)
* Remove unnecessary kernels of num_token_non_padded (sgl-project#6965)
* Extend cuda graph capture bs for B200 (sgl-project#6937)
* Fuse routed scaling factor in deepseek (sgl-project#6970)
* Sync cuda graph runners (sgl-project#6976)
* Fix draft extend ut stability with flush cache (sgl-project#6979)
* Fix triton sliding window test case (sgl-project#6981)
* Fix expert distribution dumping causes OOM (sgl-project#6967)
* Minor remove one kernel for DeepSeek (sgl-project#6977)
* [perf][sgl-kernel] extend cutlass_mla_decode to support num_head < 128 (sgl-project#6929)
* Enable more unit tests for AMD CI. (sgl-project#6983)
* Use torch.compile to fuse flash attention decode metadata preparation (sgl-project#6973)
* Eliminate stream sync to speed up LoRA batch init  (sgl-project#6960)
* support qwen3 emebedding (sgl-project#6990)
* Fix torch profiler bugs for bench_offline_throughput.py (sgl-project#6557)
* chore: upgrade flashinfer v0.2.6.post1 jit (sgl-project#6958)
* cleanup tmp dir (sgl-project#7007)
* chore: update pr test xeon (sgl-project#7008)
* Fix cutlass MLA gets almost zero accuracy (sgl-project#6998)
* Update amd nightly models CI. (sgl-project#6992)
* feat: add direct routing strategy to DP worker (sgl-project#6884)
* Fallback to lower triton version for unfound fused moe configs (sgl-project#7013)
* Fix torchvision version for Blackwell (sgl-project#7015)
* Simplify prepare_extend_after_decode (sgl-project#6987)
* Migrate to assertEqual (sgl-project#6741)
* Fix torch version in blackwell dockerfile (sgl-project#7017)
* chore: update pr test xeon (sgl-project#7018)
* Update default settings for blackwell (sgl-project#7023)
* Support both approximate and exact expert distribution collection (sgl-project#6964)
* Add decode req pool (sgl-project#6980)
* [theta]merge 0610
* [theta]merge 0610
* [CI] Add CI workflow for sgl-router docker build (sgl-project#7027)
* Fix fused_moe triton configs (sgl-project#7029)
* CPU: map changes from developing branch in sgl-kernel (sgl-project#6833)
* chore: bump v0.4.7 (sgl-project#7038)
* Update README.md (sgl-project#7040)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@merrymercy merrymercy Awaiting requested review from merrymercy merrymercy is a code owner

@Ying1123 Ying1123 Awaiting requested review from Ying1123 Ying1123 is a code owner

@hnyls2002 hnyls2002 Awaiting requested review from hnyls2002

@ispobock ispobock Awaiting requested review from ispobock ispobock is a code owner

@ByronHsu ByronHsu Awaiting requested review from ByronHsu

@zhaochenyang20 zhaochenyang20 Awaiting requested review from zhaochenyang20

@HaiShaw HaiShaw Awaiting requested review from HaiShaw HaiShaw is a code owner

@ch-wan ch-wan Awaiting requested review from ch-wan ch-wan is a code owner

@BBuf BBuf Awaiting requested review from BBuf BBuf is a code owner

1 more reviewer

@gemini-code-assist gemini-code-assist[bot] gemini-code-assist[bot] requested changes

Reviewers whose approvals may not affect merge requirements
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@zhyncs