Spaces:
Paused
Paused
Commit History
-
smoke_embedding: drop eager probe (incompatible with ZeroGPU CUDA emulation)
Arjunvir SinghZeroGPU runs the main process in CUDA-emulation mode that intercepts operations only inside @spaces.GPU-decorated functions. Calling SentenceTransformer(...) in the main process — as the smoke probe did to give early load diagnostics — triggers 'Low-level CUDA init reached' and skips the smoke before the actual benchmark runs. The benchmark below already routes through _gpu_encode_batch (decorated, runs in worker), so the right thing is to let any model-load failure surface through that path as an status. Test updated: assert benchmark exceptions yield status=error rather than the old skip-with-install-hint path.31ea79c -
Fix ZeroGPU state-loss in EmbeddingRetriever; add einops + smoke surfaces
Arjunvir SinghThe @spaces.GPU decorator runs wrapped functions in a separate worker process; mutations to `self` inside the worker do not propagate back to the caller. Decorating EmbeddingRetriever.index() / .query() therefore silently set the indexed vectors in the worker only, leaving the calling instance with empty state and producing recall=0 even though the model loaded and encoded successfully on GPU. Refactor: GPU work moved to a free stateless helper `_gpu_encode_batch(model_id, task, texts) -> vectors`. EmbeddingRetriever methods stay in the main process and dispatch via a new `_encode` method that picks the explicit injected embedder (test path) or the GPU helper (production path). Same fix pattern applies to TransformersClient when we wire live GPU repair end-to-end (deferred — current decorator is single-shot bursty work where state loss doesn't matter for one call). Also lands: - einops>=0.7.0 in requirements.txt (jina-v3's xlm_roberta_flash custom modeling needs it; sentence-transf…52764bf -
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