Commit History
-
Fix: filter on is_new_dataset when showing dataset papers
confidence_score stores the max class probability (always >= 0.5 for binary classification), so filtering on confidence >= 0.5 returned all 212K papers instead of just the 49K new dataset papers. Now requires is_new_dataset = True when confidence threshold >= 0.5. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>fa3f012 -
Use dynamic percentile-based confidence filter options
Replaces hardcoded confidence thresholds (50/90/99%) with percentile-based options (Top 75%/50%/25%) computed from actual data distribution. This adapts automatically when the underlying model changes (e.g., v2 → v3). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>0aea3fd
-
Fix confidence score: truncate instead of round to prevent 99.95%+ showing as 100%
The previous %.1f formatting rounded 0.9999 → 100.0%. Added a custom Jinja2 filter that uses math.floor to truncate, so 0.9999 → 99.9% and only true 1.0 shows 100.0%. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>460689f -
Fix confidence score display for v2 model
v2 model has 75.8% of scores above 0.99, so %.0f formatting made everything show "100%". Switch to %.1f for meaningful differentiation. Update confidence filter thresholds from v1 range (0.5-0.72) to v2 range (0.5-0.99). Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>2740f73
-
Fix: Update dataset download path for compatibility with HF_HOME environment variable
a546a29
-
Upload folder using huggingface_hub
Verified signature4cf63e7
-
Update app.py
19042f6 -
Update app.py
91eea9e
-
Update app.py
30f0f50
-
add better retries
a38b615
-
Update app.py
e878844
-
reduce number
f2e5441 -
linting
608e523 -
small fixes
0bb28cf -
fixes
45edcec -
catch all changes to toggle
d46183f -
catch all changes to toggle
ab438d6 -
typo
e5cdc54 -
typo
4fd027c -
update
c9dab8a -
add hf_transfer
17b4878 -
add hf_transfer
284db10 -
improvements
976f652
-
draft app
ad38c8f