Spaces:
Paused
Paused
Commit History
-
Fix dashboard 404 by shipping artifacts/ into the Space image
siddeshwar-kagatikarThe Space homepage exposed Post-Training and Pre-Training dashboard buttons but every click returned {"detail":"Post-training dashboard not found"} because the Dockerfile never copied the artifacts/ tree into the image and .dockerignore explicitly excluded it. The html files were on hf/main but absent from the runtime container fs. .dockerignore: keep ignoring artifacts/* by default but un-ignore the small evaluation/baseline/dashboard files that the Space UI loads: artifacts/baselines/**, artifacts/local_hf_eval_round_006/**, artifacts/latest_evaluation.json, artifacts/leaderboard.json. Dockerfile: COPY --chown=user artifacts $HOME/app/artifacts so the selectively-allowed files actually land inside the image. Heavy training outputs stay excluded by the artifacts/* rule above. Also rename the dashboard html titles to "OSINT Benchmark Dashboard post Training" / "OSINT Benchmark Dashboard pre Training" so the two views are distinguishable inside the iframe view. Made-with: Cursor7e4ee5e -
Add missing post_training_benchmark_dashboard.html to HF Space
siddeshwar-kagatikarMade-with: Cursor07286e6 -
Add pre/post benchmark dashboard options in Space UI
siddeshwar-kagatikarReplace the single default dashboard entry on the Space home page with explicit Post-Training and Pre-Training dashboard links, and wire those to dedicated routes. Update the compare-dashboard path mapping to use pre_training_benchmark_dashboard_original.html and keep compatibility aliases for existing /dashboard/finetuned and /dashboard/base URLs. Made-with: Cursor9f98669 -
added images (hf-space text-only subset)
siddeshwar-kagatikarMade-with: Cursor957b4b2 -
Add evaluation, minor updates to HF space
8ad6382
-
fix(rewards): never crash GRPO on malformed completions
siddeshwar-kagatikarTrainer crashed with ``ValueError: invalid literal for int() with base 10: 'none'`` because the model emitted ``"none"`` as a string for an orchestrator integer field. A single bad completion takes down the whole RL run, so harden parsing end-to-end: - Add ``_coerce_int`` that gracefully handles ``None``, booleans, floats (including NaN/inf), and arbitrary strings (``"none"``, ``"N/A"``, ``"2 agents"``). Use it for every orchestrator integer. - Wrap ``decode_completion_text`` and ``parse_generated_task_completion`` in defensive ``try/except`` inside ``GeneratorRewardFunction.__call__``; on any parsing/reward exception, fall through to a graded floor reward so GRPO keeps training and the failure is logged via ``[reward_debug][parse_error]`` / ``[reward_debug][reward_error]``. - Add regression tests for both behaviours. Made-with: Cursord814291