Qwen3.5-9B-Coding-SoloRL350-Resume63

One training run, multiple full-weight checkpoints on branches. This revision contains u315. 默认 main 为 u315；每个 uN 分支保存对应 checkpoint 的完整权重、训练来源和评测记录。

Checkpoint branches and orch evaluations

Branch	Orch eval150 resolved	Evaluation resource profile
u195	87/150 (58.0%)	CPU-quota thread limits; gcp2-w32-cpuquota-20260920
u315 / main	89/150 (59.3%)	CPU-quota thread limits; gcp2-w32-cpuquota-20260920
u355	89/150 (59.3%)	CPU-quota thread limits; gcp1

main uses the strongest result within the current matched-control cohort (ties prefer the later checkpoint). Historical site/thread settings are labeled above; their scores are separate observations, not pooled comparisons. The primary evaluation is orch: MiniMax-M2.7 coordinates and this checkpoint is the small coding worker.

This branch: 89/150; matching actual starting-model control: 84/150, difference +5 tasks. Paired outcomes and control fingerprint are included in EVAL_RESULTS.json.

All listed primary evaluations completed 150 tasks, passed their evaluator audits, and had zero final infrastructure errors. Model/test failures and verifier timeouts count as unresolved. Every selected checkpoint's available completed historical repetitions are preserved in EVAL_RESULTS.json. Selection used eval150 results, so these are exploratory candidates, not independent confirmation of a gain or an official full SWE-bench leaderboard score.

Training run and provenance

• W&B run: 53b2b537
• Run name: solo350_from-local63_g8-b64_lr5e-7_kl0_retract_continuous-rl
• Campaign: solo350-restart63-g8-20260917
• Lineage: Original HF OPD step109 → earlier Solo350 local63 → resumed Solo GRPO on SWE-bench train350.
• Training algorithm / mode: Solo GRPO / solo.
• Training data: Historical SWE-bench Verified train350 subset.
• Local checkpoint: u315. uN is the recorded local checkpoint iteration label, not the original HF OPD step number. Solo350 continues numbering from the earlier local63 checkpoint.
• Optimizer: weights-only resumes use a fresh optimizer; this run includes restarts.

This Solo GRPO training phase has no teacher or orchestrator. Its evaluation still uses M2.7 orchestration.

Training setting	Value
Learning rate	5e-07
Samples per update	64
Group size	8
Prompt groups per update	8
Reference-policy KL coefficient	0.0
Maximum weight staleness	1
Save interval	4 updates
Worker thinking	disabled

Exact parent revisions, dataset hashes, archived training settings and native checkpoint identity are in CHECKPOINT_PROVENANCE.json and TRAINING_CONFIG.json.

Evaluation settings

150 held-out tasks, 32 episodes per job, 2 CPUs and 10 GiB per sandbox; worker thinking disabled. Worker/coordinator temperatures: 0.2; maximum worker turns: 64; response budgets: 8192/4096 tokens; verifier timeout: 2400 seconds. CPU-quota profiles additionally set Django/numeric thread defaults to 2. Exact per-branch settings, task outcomes, fingerprints, repetitions and resource profiles are in EVAL_RESULTS.json.

Download a checkpoint

from huggingface_hub import snapshot_download
path = snapshot_download("CharlieLLL/Qwen3.5-9B-Coding-SoloRL350-Resume63", revision="u315")

Use a serving stack with Qwen3.5 support and the included tokenizer/chat template. These are full BF16 weights plus tokenizer/processor assets; optimizer state is not included. Evaluation covers text-based coding tasks; inherited vision configuration does not establish evaluated multimodal behavior. License: Apache-2.0; see LICENSE.