Spaces:
Sleeping
Sleeping
Commit History
-
Add BrowseComp+ tab: dataset explorer with question, evidence docs, gold answer
timchen06185cc2a94
-
Derive new_status from new_trajectory; fix sidebar check mark; fix question for incomplete
timchen0618- selected_tools backend: compute new_status from [Final Answer] presence in new_trajectory, instead of trusting the eval file is_completed which reflects the reference run, not new run - SelectedToolsApp: use new_status for incomplete detection and status badge (fixes query 49) - SelectedToolsApp: question row and incomplete badge now use new_status (correct for all runs) - patch_sel_tools_test300_questions.py: fills empty question fields from BrowseComp JSONL (11 selected-tools test300 repos, running on torch)d14bce3
-
Show incomplete runs as incorrect; fix missing questions via BrowseComp JSONL fallback
timchen0618- SelectedToolsApp: show ✗ Incorrect badge for incomplete runs (correct===null && status==="incomplete") - ScoutRunsApp: same for scout_status==="incomplete" - Both tabs: show question row for incomplete runs; fallback text if question is empty - build_scout_runs_with_eval.py: add --browsecomp-jsonl arg to fill missing questions - patch_orig_analysis_with_eval.py: same fallback - rebuild_all_scout_runs_with_eval.sh: pass BC_JSONL to all 12 variants1eb493c -
Add question/answer/accuracy to Scout Runs tab; fix selected-tools reload cache
timchen0618Scout Runs: accuracy %, ✓/✗ badges in sidebar, question+answer in header. Selected Tools: same pattern already deployed; reload now deletes HF cache first. Both backends load question/correct_answer/correct from HF datasets. Patch scripts for adding eval columns to existing scout-run and orig-analysis datasets.30fb9c5 -
Add question, correct answer, and accuracy to Selected Tools tab
timchen0618Backend loads new columns from HF datasets (question, correct_answer, correct). Frontend shows: question + answer in header, accuracy % in stats banner, correct/incorrect badges (✓/✗) on sidebar items and header. Rationale and selected steps now hidden when empty (new-format datasets).d0a0739
-
Fix Selected Tools default variant (was hardcoded to removed traj_summary_ext)
timchen061874c4b8c -
Fix Scout Runs default variant (was hardcoded to removed budget5)
timchen0618921df29
-
Add Scout Runs tab: new backend endpoint, ScoutRunsApp frontend, budget5 variant
timchen061852b0adf
-
Add traj_summary_orig_ext variant to Selected Tools tab with variant switcher
timchen0618fa6b40b
-
SFT Diff: add qwen template dropdown, 830 rows (both templates)
timchen0618a45ef56
-
Add SFT Diff tab: side-by-side original excerpt vs converted Axolotl messages
timchen061847d3d12
-
Upload folder using huggingface_hub
Verified signature15f4894 -
Upload folder using huggingface_hub
Verified signature4010073 -
traj_ext: add display name mapping for 6 run types
timchen06186307feb
-
traj_ext: fix 1.1GB response — lazy-load formatted_prompt, truncate tool results in list
timchen06183c34d20 -
traj_ext: add 'direct' badge when n_tool_calls == 0
timchen061855037ce -
fix: use correct index.html from fresh build (was stale from rebase conflict)
timchen061852302c1 -
traj_ext: add run selector, run_name badge, prompt-on-top in both view; load bcp-full-runs-v1
timchen06187ea306b
-
Show selected step indices in header
Verified signature2b7fb4d -
Add Selected Tools tab with side-by-side excerpt/trajectory viewer
Verified signaturebd51d10
-
Add Traj Ext viewer tab with trajectory block rendering and HF dataset
timchen0618c682eb7 -
Deploy research dashboard
timchen0618b03f016