DistillDetect-gemma-3-4b-pt-from-o1-s1

Unofficial reproduction of a distilled student model from the paper Reference-Based Distillation Detection in LLMs (Rawat et al., arXiv:2607.09692), retrained with the authors' released code and teacher-generated data (github.com/RajatRawat-creator/DistillDetect, MIT). The original authors did not release student checkpoints; this repo is an independent reproduction and is not affiliated with the authors.

• Base (student) model: google/gemma-3-4b-pt
• Teacher: openai/o1 (API, summarized traces)
• Training data: s1 (1K prompts) answered by OpenAI o1 — 1000 teacher-generated responses, shipped verbatim in the authors' repo (data/training/o1__s1k__chat__openai_responses)
• Prompt template: plain Problem:\n{question}\n\nSolution:\n

Training

SFT with the authors' released training/ scripts (paper Appendix A recipe): 3 epochs, LR 1e-5, cosine schedule, 5% warmup, per-device batch 4 x grad-accum 4 (effective batch 16), block size 4096, bf16, gradient checkpointing, loss on response tokens only (prompt masked -100). Teacher responses were pre-truncated to 2,048 tokens in the released data for the controlled teachers; the o1 responses are not truncated (max ~4k tokens, capped by the 4096 block). Trained on 1x H100 (paper used 2x H200; hyperparameters identical), transformers 4.55.4 / trl 0.19.1, seed 42 (HF default; paper seed unknown). One compatibility patch: trl renamed max_seq_length to max_length, so the block size is passed explicitly (no behavioral change).

Evaluation (ours vs. paper Table 9)

Greedy decoding, template matched to training, scored with math_verify. GSM8K 4-shot; MATH500 zero-shot. Few-shot counts were calibrated so the base models reproduce their Table 9 baselines. The paper does not document its eval protocol, so treat cross-paper comparisons as approximate.

Benchmark	This reproduction	Paper Table 9	Gen. budget	Hit budget cap
GSM8K	46.78	n/a (config not retained in paper)	4096 tok	1.0%
MATH500	21.80	n/a (config not retained in paper)	16384 tok	14.4%

A high "hit budget cap" fraction means the model was still generating when the token budget ran out, so that accuracy is a lower bound — these students are SFT'd on teacher traces that were themselves truncated at 2,048 tokens, which makes some of them generate very long self-checking traces.

Base-model reference (our protocol / paper): GSM8K 27.37 / 38.13, MATH500 18.80 / 23.60.

License

The base model's license applies: gemma. Teacher-generated training data redistributed by the paper authors under their repo's MIT license; note the o1 responses originate from the OpenAI API and were released by the paper authors.