RockToken/qwen3_30b_a3b_to_4b_onpolicy_code_5k

Qwen3-4B distilled from Qwen3-30B-A3B — On-policy 5k (code)

On-policy KD for a Qwen3-4B student toward the Qwen3-30B-A3B teacher, on 5,000 code prompts from OpenThoughts-3.

Domain-transfer companion to `qwen3_30b_a3b_to_4b_onpolicy_science_5k` and the math-domain onpolicy_5k_src* chain: same student starting point and recipe, different domain slice. Training exposure of this checkpoint:

1. 1.Off-policy KD on 20k math prompts → `qwen3_30b_a3b_to_4b_offpolicy_20k`
2. 2.On-policy KD (this run) on 5k code prompts → this checkpoint

Models

enable_thinking=False throughout.

Training data

• —Source: open-thoughts/OpenThoughts3-1.2M, domain == "code" slice
• —5,000 single-user-turn prompts; 70 exceeding prompt_max_len=2560 were filtered, leaving 4,930
• —Only the prompts are used; on-policy KD never reads the dataset's reference answers

Training setup

Framework: KDFlow — FSDP2 + SGLang rollout, Ray-orchestrated GPU co-location with sleep/wakeup. Same pipeline revision as the math-domain onpolicy_5k_src* runs.

Hardware: 1× node, 4× H100 (94 GB), 33 h 09 min wall-clock (~133 GPU-hours).

Key hyperparameters

prompt_max_len is 2560 instead of the math runs' 800: code prompts carry problem statements and starter code (p99 ≈ 2,356 tokens).

Note on the LR schedule: as in the earlier runs of this series, the scheduler's horizon is sized to 4× the actual number of updates, so the cosine decay is only partially traversed — the LR moves from 2.00e-6 to 1.79e-6 over the run rather than reaching zero. Kept as is for comparability within the series.

Training dynamics

Means over 500-step blocks (per-step values are noisy at batch 4):

The loss starts near 0.19 rather than the ~3 seen when distilling from a raw instruct model — the student has already been through off-policy KD against the same teacher, so on-policy training starts close and grinds out the remaining distribution gap.

⚠ Truncation caveat

Code responses run long, and a substantial share of rollouts hit the 8,000-token generation cap: per-step mean lengths sit at p50 = 7,069 against the cap, and in 20% of steps the whole batch was truncated (step-mean ≥ 7,990). Truncated rollouts contribute KD signal only for their first 8,000 tokens, and the student receives no supervision on how the teacher would have ended those responses. Treat comparisons against the science and math checkpoints (which trained essentially truncation-free) with this in mind.

Intended use

Research on distillation dynamics, in particular whether on-policy KD gains transfer across domains (math-trained off-policy base → code on-policy round). Domain: code (OpenThoughts-3 code split).

Limitations

• —The on-policy round is code-only; not tuned for chat or safety.
• —enable_thinking=False — this student does not emit thinking traces.
• —Heavy generation-length truncation during training; see the caveat above.
• —Single checkpoint at the end of one epoch; no intermediate checkpoints were kept.