serene-abyss/peer-review-env

fix: modernize pyproject.toml and cleanup parent folder for validation

fix: OpenEnv validation errors and cleanup redundant logs

UI: side-by-side agent and reward cards, prompt tuning for exact flaw matching

fix: resolve UI layout clipping inside HF iframes and fully implement real LLM execution in RL Training dashboard

feat: switch dashboard UI LLM backend to use ultra-stable Google Gemini API to prevent HF routing errors

fix: switch default model to zephyr-7b-beta since Qwen 1.5B throws model_not_supported in HF inference router

fix: map root path (/) to dashboard UI to fix 404 on HF Space default page

fix: change openenv-core version from >=0.3.0 to >=0.2.0 (0.3.0 doesn't exist)

fix: split pip install - torch from CPU index, rest from PyPI

fix: single-layer pip install to reduce build footprint

fix: remove SBERT pre-bake from Dockerfile (fails in HF build)

fix: CPU-only torch for HF Spaces build (avoids 2GB CUDA download)

peer_review_env: OpenEnv RL environment for automated scientific peer review