madhuria/patch2prod-arena

fix: clean deployment, dev workflow, and docs for HF Space

fix(space): align docker startup with Hugging Face app port

docs: add local and Kaggle SFT notebook references

artifacts: add GRPO clipped ratio plot via LFS

artifacts: add GRPO completion plot via LFS

artifacts: add GRPO entropy plot via LFS

artifacts: add GRPO grad norm plot via LFS

artifacts: add GRPO reward plot via LFS

artifacts: add GRPO loss plot via LFS

docs: fix markdown image rendering and expand experiment log

docs: add project blog writeup for hf submission

update README and add GRPO training log

fix: gate kl_coef on GRPOConfig signature to avoid TypeError

fix: eval/train format alignment + KL + live metrics + multi-panel plots

fix: lightweight GRPO training - chat template + stop tokens + tuned hyperparams

chore: trigger hf space rebuild for README sync

Readme updates

fix hf short_description length

update hf space README metadata fields

fix hf space config: use root index.html app_file

fix hf spaces metadata color

update demo ui and add ci webhook + hf space setup

fix GRPO training: strict parse, termination reward, variance jitter, throughput

grpo training

grpo training

Fix GRPO training: use env-grounded prompts, add reward debug logging, honest eval

evaluate grpo

grpo changes

grpo changes

setting up grpo

Post SFT Training

Add working baseline and risk-aware evaluation traces

training fixes

training stuff

Add baseline evaluation artifacts

Initial Patch2Prod Arena submission