Spaces:
Sleeping
Sleeping
Commit History
-
v3: benchmark results, final report, agent/eval improvements, smoke test fixes
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>b3fc5ee -
v2: curriculum scheduling, SFT pipeline, reward redesign, agent guide
Phases 1-8 complete: - Reward system centralized in reward.py - Synthetic dataset generators with NoiseProfile (light/medium/heavy) - Evaluation harness with RandomAgent and HeuristicAgent baselines - CurriculumScheduler: easy→medium→hard progressive difficulty - LLMAgentAdapter + training harness (run_experiment.py) - SFT demonstration pipeline (demonstrations.py + training_sft.py) - GRPO warm-start support in training_colab.py - AGENT_GUIDE.md: full tool reference, reward formulas, onboarding - examples/minimal_agent.py: simplest runnable agent Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>16038fc -
Upload folder using huggingface_hub
Verified signature1779f34