Spaces:
Paused
Paused
Commit History
-
Force Space rebuild to pick up new HF_TOKEN secret
Arjunvir Singhb4316e6 -
Enable Docling OCR + table-structure in the Space's default config
Arjunvir SinghDocling ships built-in OCR (EasyOCR) and a TableFormer model for table-structure detection. They were both off in configs/docling.yaml which meant the Space's default "Docling + PyMuPDF" mode never exercised them — scanned PDFs returned 0 elements / quality 0.0, and PDFs with positioned-text tables produced 0 table objects. Turning both on so the spec §5 vision-model expectations actually fire on the live deploy: - do_ocr: true (EasyOCR runs on pages with weak text layers) - do_table_structure: true (TableFormer detects table cells with bbox-aware extraction) Picture description / classification stay off — those tasks are routed through the project's own GPU repair pipeline (Qwen2.5-VL) in Live GPU repair mode, for consistent task accounting.be07287 -
TransformersClient stateless refactor + Smokes/Benchmark UI surfaces
Arjunvir Singh- TransformersClient: same ZeroGPU state-loss fix as EmbeddingRetriever. GPU work moves to a free `_gpu_run_pipeline(model_id, task, dtype, device, prompt, image_path) -> str` helper; execute_task stays in the main process and only returns the picklable extracted text. Removed the `cached_property` _pipeline (it was useless under @spaces.GPU since the worker process exits each call) and the `raw_output` field (raw pipeline output may contain torch tensors that don't pickle cleanly across the worker boundary). - configs/live_gpu_repair.yaml: third pipeline mode that flips repair.execute_gpu_escalations=true and points at Qwen2.5-VL-3B for table/OCR/figure repair tasks. - app.py: - "Live GPU repair" added as a third pipeline-mode option, mapped to configs/live_gpu_repair.yaml. - ZSGDP_SMOKE_EMBEDDING_MODEL_ID defaults to all-MiniLM-L6-v2 so the embedding smoke avoids jina-v3's transformers-compat issue out of the box; override via Space settings if you…8bd8e7a -
Initial commit: zeroshotGPU MVP with full eval surface
Arjunvir SinghProfiler, router, parser registry, schema, merger with conflict detection, verifier (coverage/reading-order/table/figure/formula/chunk-readiness plus GT-comparison: layout F1, table structure, formula CER, retrieval recall), iterative repair loop with optional GPU escalation, agentic chunker, benchmark suite (per-doc + per-parser + ablation + cross-dataset), Gradio Spaces UI with abuse guards + per-artifact downloads, structured JSON logging, preflight runner, regression-fixture format with perf floors, .env loading, pre-commit/pre-push hooks, CONTRIBUTING.md + docs/space_smoke.md, scripts/run_space_smoke.py runner. Test count: 240/240 passing.db06ffa