Spaces:
Running
Running
Commit History
2 days ago
-
Grading: LLM search in cross-domain, one grader per search Running
Verified signature4693ca0 -
Grading page for the embedding model comparison; detail pages open from their address; asset version label10
Verified signature65a9937
3 days ago
-
Search defaults, LLM search count and grading, links panel clicks and card under the graph; asset version label9
Verified signatureSearch Model starts on BGE large v1.5 tuned; Number of Results 100 for embedding, LLM search 50 phenotypes / 1000 codes, and POST /api/label-search takes k; LLM search rows can be graded (annotations keep judge and label). Links panel: curved links, aligned labels, cumulative line weight, faint waiting branches, card under the graph, one click one result.243e328
4 days ago
-
Name embedding models by their base model, with tuned on the tuned ones; asset version label6
Verified signature8209680
5 days ago
-
Label search as a job with short polls; asset version label4
Verified signatureea4fedd -
Label search; built-in model from ENCODE_LLM_* settings; links wheel update
Verified signatureGroup by relevance: phenotype and code results in relevant, related and not relevant groups by the built-in model (GPT 6 Luna). The model endpoint, id and name now come from Space variables; the key stays the OPENAI_API_KEY secret. Links wheel update from GitHub ff8d3d1.86f07a9
10 days ago
-
Links panel replaces the code graphs: one ring graph per code, linked concept and phenotype, Source or Derived on every link (MED-RT statements on drug links), concept cards labelled AI generated; also carries the 2026-10-20 dead-code trim and PARSE source name
Verified signature68fc933
12 days ago
-
Serve four embedding models: BGE VA-tuned (default), BGE base, Harrier PARSE-tuned, Harrier base
Verified signatureModel registry with per-model code indexes over shared rows, shared phenotype state, Harrier base and PARSE LoRA on one fp32 model, per-model low-confidence floors, model-aware CSV export, peft requirement. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>5e5072d
-
Serve one planner model (GPT 5.6 Luna on OpenAI); remove the Agent model picker and bring-your-own-model dialog
Verified signature81c7b7b
-
Upload folder using huggingface_hub
Verified signaturec4bc1a0
-
Upload folder using huggingface_hub
Verified signatureecb6008
-
Upload folder using huggingface_hub
Verified signature6993a14 -
Upload folder using huggingface_hub
Verified signature21711e7
-
Upload folder using huggingface_hub
Verified signature65dff80
-
Upload folder using huggingface_hub
Verified signature9ce728e -
Upload folder using huggingface_hub
Verified signature0372560 -
Upload src/backend/llm.py with huggingface_hub
Verified signature4220e54 -
Upload src/backend/llm.py with huggingface_hub
Verified signaturebbf7584 -
Upload folder using huggingface_hub
Verified signature2b2a570 -
Upload folder using huggingface_hub
Verified signature0734562
-
Upload folder using huggingface_hub
Verified signature2351481 -
Upload folder using huggingface_hub
Verified signature6583c2e -
Upload folder using huggingface_hub
Verified signature16ff626
-
Upload folder using huggingface_hub
Verified signature6495d6b -
Upload folder using huggingface_hub
Verified signaturec7415fe
-
Show full graph node details
Verified signatured82df63 -
Simplify public copy and method labels
Verified signaturee3e0bee -
Remove development-only files from Space source
Verified signature2f659d8 -
Deploy reviewed LOINC graph and updated UI/backend
Verified signaturef7b909f
-
Remove model description text from the picker
Verified signatureDrop the blurb/tags fields from the registry and the copy they fed: the sidebar note under the dropdown and the About panel's embedding-model section. The picker shows model names only.a08364d -
Add embedding-model picker backed by a server-side model registry
Verified signatureNew /api/models registry publishes which models exist, what each can search, and whether this deployment serves them. Search endpoints accept &model=<id> and stamp responses with model_id/model_label; annotations record the model. Scopes the ENCODE_PARTA_MODEL query-encoder override to the default vector set so an alternate model cannot silently encode queries with mismatched weights.bc7e936 -
Deploy ENCODE with private artifact bucket
Verified signature9a3640a