Spaces:
Running
Running
Commit History
-
feat: real continuous GRPO reward curve + 5-plot dashboard
- data/trainer_state.json: 30 log entries from the actual 300-step GRPO run on Colab A100 (manually retrieved from Drive) - regenerate plots from real training data: - reward_curve.png: continuous mean-reward + ±std band, 10-step granularity (replaces the previous discrete bypass-vs-step-count chart) - kl_loss_curve.png: KL divergence + policy loss on twin axes - completion_stats.png: mean length + clipped_ratio (the 100% cap signal) - Dashboard now shows 5 plots + a live-stats card with proper captions matching what each plot actually depicts Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>723a54a
-
Phase 7: replay traces for HF Space demo
colabe60c2d2 -
Phase 7: replay traces for HF Space demo
colaba4470c7 -
feat: HF Space replay backend — trace store, /api endpoints, Docker, Cell 9
- env/replay.py: TraceStore loads pre-recorded JSON traces from data/traces/ - env/server.py: /api/attack-types, /api/attack, /api/stream/{type}/{steps}, /api/highlight, /api/stats; static frontend mount; INJECTARENA_MODE switch - Dockerfile: switch CMD to uvicorn env.server:app, set INJECTARENA_MODE=replay - README.md: HF Space frontmatter (sdk: docker, app_port: 7860) - scripts/generate_traces.py: runs trained attacker against scenarios on Colab, captures stage-by-stage timeline and outcomes into JSON traces - notebooks: Cell 9 to invoke the trace generator and push results to GitHub Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>8c536e6