cdx1-mini #2148
Merged
cdx1-mini #2148
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
prabhu
commented
Aug 9, 2025
contrib/fine-tuning/fine-tune-mlx.sh
Outdated
| @@ -13,7 +16,7 @@ case "$TOOL_BASE_MODEL" in | |||
| esac | |||
| BASE_MODEL_MLX=${BASE_MODEL}-${TUNING_TOOL} | |||
| HF_ORG=CycloneDX | |||
| NUM_LAYERS=16 | |||
| NUM_LAYERS=32 | |||
There was a problem hiding this comment.
While working on cdx1-mini model, I noticed the bug with our lora fine-tuning. The performance for cdx1 and cdx1-pro can be improved slightly in a future iteration.
prabhu
commented
Aug 9, 2025
contrib/fine-tuning/fine-tune-mlx.sh
Outdated
| mlx_lm.lora --model ${BASE_MODEL} --train --data ${DATASET_PATH} --adapter-path ${ADAPTERS_PATH} --mask-prompt --fine-tune-type lora --batch-size 1 --num-layers ${NUM_LAYERS} --iters 2000 --grad-checkpoint --max-seq-length 16000 --learning-rate "3e-5" --optimizer adam | ||
| if [ "$TOOL_BASE_MODEL" = "cdx1-mini" ]; then | ||
| echo "Full fine-tune with cdx-docs dataset. This might take a while ..." | ||
| mlx_lm.lora --model ${BASE_MODEL} --train --data ${DATASET_PATH} --adapter-path ${ADAPTERS_PATH} --mask-prompt --fine-tune-type full --batch-size 2 --iters 1500 --grad-checkpoint --max-seq-length 8192 --learning-rate "1e-5" --optimizer adamw --num-layers 32 --seed 42 |
There was a problem hiding this comment.
For mini, we take a small 4B model and perform a full fine-tuning rather than lora/qlora.
b9de9e9 to
2c4f58c
Compare
94ad088 to
4a34e36
Compare
prabhu
commented
Aug 9, 2025
| mlx_lm.lora --model ${BASE_MODEL} --train --data ${DATASET_PATH} --adapter-path ${ADAPTERS_PATH} --mask-prompt --fine-tune-type full --batch-size 2 --num-layers ${NUM_LAYERS} --iters ${ITERS} --grad-checkpoint --max-seq-length ${MAX_SEQ} --learning-rate "1e-5" --optimizer adamw | ||
| elif [ "$TOOL_BASE_MODEL" = "cdx1" ]; then | ||
| echo "Low-Rank Adaptation (LoRA) fine-tuning ${BASE_MODEL} with cdx-docs dataset. This might take a while ..." | ||
| mlx_lm.lora --model ${BASE_MODEL} --train --data ${DATASET_PATH} --adapter-path ${ADAPTERS_PATH} --mask-prompt --fine-tune-type lora --batch-size 1 --num-layers ${NUM_LAYERS} --iters ${ITERS} --grad-checkpoint --max-seq-length ${MAX_SEQ} --learning-rate "1e-4" --optimizer adamw |
There was a problem hiding this comment.
I have started a new fine-tuning job for cdx1. Will create a separate PR tomorrow once it is ready for reevaluation.
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.
cdx1-mini (base: unsloth/Qwen3-4B-Instruct-2507) is now available for beta testing.
Logic Category Comparison
Beats o4-mini-high!
Spec Category Comparison
Matches
cdx1-proeven though this is just a 4B model 😎Other categories
Due to the small size, it has limited knowledge in these topics.
Known bugs