README.md

metadata

license: cc-by-nc-4.0
language:
  - en
pretty_name: FML project, multi-teacher distillation data (experiment 2)
tags:
  - knowledge-distillation
  - math
  - code
task_categories:
  - text-generation
size_categories:
  - 10K<n<100K

Multi-teacher distillation data (CS725 course project, experiment 2)

Training data for distilling Qwen2.5-Math-7B-Instruct (math) and Qwen2.5-Coder-7B-Instruct (code) into a small student (Qwen2.5-1.5B-Instruct). Files are in exp2/; exp2/README.md describes every file and how the data was built (exp2/gen_data.py).

Each row holds token ids (Qwen2.5 vocabulary) of prompt (the teacher's chat template and system prompt), response (the teacher's own answer, greedy decoding) and gold (the dataset's reference answer). The last 200 rows of each file are validation rows.

File	Rows	Sources
exp2/math.jsonl	14,559	GSM8K train + MATH train
exp2/code.jsonl	14,956	KodCode-Light-RL-10K + KodCode-V1
exp2/math_hard_val.jsonl	200	MATH train (held out)

Sources and licenses

• GSM8K (openai/gsm8k, train split) — MIT.
• MATH (EleutherAI/hendrycks_math, train split; Hendrycks et al., 2021) — MIT.
• KodCode (KodCode/KodCode-Light-RL-10K, KodCode/KodCode-V1; Xu et al., 2025) — CC BY-NC 4.0: non-commercial use only, with attribution. Because the code rows derive from KodCode, this dataset is released under CC BY-NC 4.0.
• Teacher answers generated with Qwen2.5-Math-7B-Instruct and Qwen2.5-Coder-7B-Instruct (Apache-2.0).

Teacher answers are not filtered for correctness (the math teacher is right on ~96% of GSM8K and ~83% of MATH rows). Overlap with the GSM8K test set, MATH-500, HumanEval(+) and MBPP(+) was removed (see exp2/README.md).