license: apache-2.0
base_model: Qwen/Qwen3.6-27B
tags:
- debugging
- code
- reinforcement-learning
- gdpo
- precisecoder
PreciseCoder-27B
馃搫 Paper (arXiv:2610.18755) 路 馃寪 Project page 路 馃捇 Code
Qwen3.6-27B post-trained with reinforcement learning for precise debugging: given buggy Python code, produce a minimal, correct fix (PreciseCoder project). Best-validation checkpoint, global step 328 (2 epochs).
Recipe. GDPO from the base model (no SFT warm-start): two reward signals, unit-test pass (weight 2.0) and unit-gated F1 (weight 1.0), each normalized within the rollout group before being combined into the advantage. F1 is the edit-block precision/recall F1 against the ground-truth diff (PDB evaluator), counted only when the unit tests pass. Trained on the PDB train split (buggy inputs only) with verl on 1 node of 8脳H200 (n=6 rollouts, temperature 1.0, lr 1e-6, batch 32, max response 6144 tokens, KL loss 0.005, rollout TP=2, token-level rollout importance sampling). Minimal-edit prompt template.
Eval (PDB benchmark, BigCodeBench + LiveCodeBench debugging tasks):
| eval | result |
|---|---|
| pdb_test (buggy inputs, 4k think + 2k answer) | unit 0.793 路 precision 0.792 路 recall 0.867 路 F1 0.810 路 edit lines 2.6 |
| gt_holdout (1512 clean inputs, 16k think + 4k answer) | no-op 45.8% 路 damage 11.1% 路 unit pass 89.0% |
| PDB-Wild (228 SWE-smith repo-level bugs) | unit 0.522 路 precision 0.551 路 recall 0.629 路 F1 0.566 路 edit lines 9.2 |
no-op = share of clean (bug-free) inputs returned unchanged; damage = share broken by an edit.
Load with transformers / vLLM (>=0.17) as a standard Qwen3.6 model (thinking enabled by default).
Source: verl experiment qwen36-27b-GDPO_u2f1_f1ur_h200 (Delta, Oct 2026). The previous release
(GRPO, reward (unit + 0.5路F1)/1.5) is at revision 82cf7e4.
Citation
@article{precisecoder,
title = {{PreciseCoder}: Faithful and Generalizable Debugging Beyond Unit-Test Rewards},
author = {Zhan, Yunkai and Chai, Miaosen and Wang, Shangshang and Zhong, Jike and Yang, Yuqing and Fu, Deqing and Chang, Serina and Zhu, Wang Bill},
journal = {arXiv preprint arXiv:2610.18755},
year = {2026}
}