This student shows no measurable capability gain over its base model.

A later trajectory study (exp_student_trajectory_2026-08-20) compared this checkpoint with its own base model on the full GSM8K test set: 65.9 → 67.9 (+2.0 points), McNemar paired test over 1,319 questions p=0.12 — not significant at α=0.05. 307 of 1,319 answers changed (140 right → wrong, 167 wrong → right) for a net of +27.

Fine-tuning did change the model's behaviour; it did not change how often it is right. That may still be useful for studying style transfer independent of capability, but do not treat this as an improved model.

DistillDetect-Qwen2.5-1.5B-from-gpt-oss-120b-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: Qwen/Qwen2.5-1.5B
• Teacher: openai/gpt-oss-120b
• Training data: s1 (1K prompts) — 1000 teacher-generated responses, shipped verbatim in the authors' repo (data/training/Teacher=GPT-OSS-120B_Data=S1_Template=Chat.jsonl)
• 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. 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	68.92	69.67	8192 tok	3.4%
MATH500	31.40	40.60	16384 tok	58.8%

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 58.07 / 67.25, MATH500 35.40 / 32.60.

License

The base model's license applies: apache-2.0. Teacher-generated training data redistributed by the paper authors under their repo's MIT license.