Repository navigation
Conversation
|
@tastelikefeet Could you please review this GS1 decision-training PR and approve the pending |
|
The training path this PR adds does not run end-to-end in the default configuration, and the PR's own test fails against the PR's own code. Below are the issues: |
Reuse the native GS1 model and compiler in ms-swift's decision-training path. Add request preprocessing, model/template registration, and a single-stage distribution NLL plus configurable expected Brier objective.
2da8ed5 to
92215f2
Compare
@z0o0ey @tastelikefeet I have manually updated the commit to improve by your comments, feel free to relaunch the CI test and provibe any feedback to us. Thank you so much! |
|
Thank you very much for the quick fix; I have one more question.The preprocessor adds "Choose the best matching option." to instructions that are otherwise missing—an approach that differs from the official stance, which rejects the inclusion of such default instructions. While this serves as a reasonable, permissive default for generic data, it deviates from official behavior; is this intentional? |
|
Additionally, there is a latent issue where labels and logits rows become misaligned within a mixed batch. In |
@z0o0ey Thank you for the detailed review. I’ve updated the local branch to address the issues I could verify: • GS1 preprocessing now keeps the model available during dataset encoding and disables multiprocessing for this model-dependent template. Static checks passed (git diff --check, compileall on changed Python files, and bash -n on the example script). I could not run pytest because the current environment does not have PyTorch installed. I have not yet verified inference behavior, the native compiler’s tokenizer state handling, or DDP behavior for calibrator.beta against the GS1 dependency and checkpoints. I’ll follow up on those after |
- Preserve GS1 model during dataset encoding and avoid worker serialization\n- Keep labels aligned for records without targets\n- Return zero Brier contribution for empty option rows\n- Reject questions with missing instructions
PR type
PR information
Refs #10598
Add native GS1 decision-model support to ms-swift. The integration reuses GS1's model and prompt compiler, supports the 0.8B, 2B, 4B, and 9B checkpoints, and trains with
distribution NLL + brier_weight * expected Brier. SetGS1_BRIER_WEIGHTto configure the Brier coefficient; training currently supports text states.AI assistance was used. The human submitter has reviewed the changes and is responsible for this contribution.
GS1 project: https://gs1.mybigai.ac.cn
Source: https://github.com/bigai-nlco/GS1
GS1 model family:
Experiment results
git diff --check: passedflake8,isort, andyapf: passed on changed Python filespytest tests/train/test_gs1.py tests/train/test_decision_loss.py -q: 16 passedbash -n examples/train/decision/gs1.sh: passedChecklist