suryadv/IncorrectTraceSFT-Qwen2.5-7B-AIME-SplitB-correct
IncorrectTraceSFT · AIME SplitB correct
A full fine-tune of Qwen2.5-7B from How Should Incorrect Traces Be Used in Supervised Fine-Tuning? This checkpoint is the held_out_correct_6000 condition in the aime splits experiment.
Code and reproduction instructions · Training data
Training
The student starts from `Qwen/Qwen2.5-7B`, revision d149729398750b98c0af14eb82c78cfe92750796. Training uses SLIME with Megatron-LM, BF16, global batch size 64, and AdamW with a peak learning rate of 5e-6. Full settings and exact data selections are in the code repository.
This is the final checkpoint after 188 optimizer updates. The original checkpoint identifier is 4166bc1cf70e6961561c8f814841d8cd9ebfdaaa2f613b62d294ab9497ae0c57.
Evaluate
From a checkout of the code repository with evaluation dependencies installed:
python -m incorrect_trace_sft.evaluate --experiment aime_splits --model suryadv/IncorrectTraceSFT-Qwen2.5-7B-AIME-SplitB-correct --output runs/AIME-SplitB-correctThe evaluation command preserves the experiment's prompts, context limit, sampling parameters, and answer scorer. It reports pass@1 and the Monte Carlo standard error across decoding repeats. See the repository's results table for all conditions and comparisons.
Files and intended use
The safetensors weights are unchanged from the evaluated checkpoint. provenance.json records their original SHA-256 hashes. The tokenizer configuration has been normalized for Transformers 4.57: legacy special-token metadata is represented using the current field format, and the existing chat template is embedded. Vocabulary, token IDs, and the chat template are unchanged.
This model is released for research on mathematical reasoning and SFT data selection. Correct final answers in the training data do not imply that intermediate reasoning steps were verified. ID/OOD in this study refers to problem identity relative to SFT; it says nothing about pretraining exposure. Coding-agent checkpoints will be released separately.