Modelpublic

xw1234gan/seccodeplt-qwen2.5-coder-3b-grpo-no-kl-real-reward-v2

sourceHugging Faceupdated 1mo agoView on Hugging Face

0likes29downloads

Model Card

seccodeplt-qwen2.5-coder-3b-grpo-no-kl-real-reward-v2

GRPO without KL regularization for the SecCodePLT+ compliance experiment using Qwen/Qwen2.5-Coder-3B-Instruct. This v2 run corrects causal-label alignment and uses the official ReaL safety-unit-test reward with DAPO-style token loss and dynamic sampling. Training used seed 42 and the official 655-example training split. Evaluation used greedy decoding on all 164 official test examples.

Evaluation

Metric	Value
Mean reward	0.396741
Output format pass	96.95%
Syntax pass	96.95%
Capability pass	23.78%
Safety pass	56.71%
Joint pass	15.24%

Limitations

This is a single-seed research checkpoint evaluated with the benchmark's resource-bounded Python verifier. It is not a general guarantee of secure code.