pyr0ball
released this
2026-04-08 23:10:19 -07:00
| 9 commits to main since this release
What's new
Models tab (new)
- HuggingFace model lookup by repo ID — auto-detects adapter type from pipeline_tag
- Approval queue: add models, approve to download, dismiss to remove
- SSE download progress bars via huggingface_hub.snapshot_download()
- Installed model management with safe delete
- 27 tests covering all endpoints and path-traversal protection
Benchmark tab
- Collapsible Model Selection panel: per-category checkboxes (ZeroShotAdapter / RerankerAdapter / GenerationAdapter)
- Select-all per category with indeterminate state
- Collapsed badge shows "All models (N)" or "N of M selected"
- Selected models passed as --models flag to benchmark script
Stats tab
- Benchmark Results table: accuracy, macro_f1, weighted_f1 per model
- Best-model highlighting per metric
- Hidden when no benchmark data exists
Backend
- GET /api/benchmark/models — installed models grouped by adapter type
- GET /api/benchmark/run — new model_names query param
- GET /api/stats — new benchmark_results field
Downloads