vrr/SurfSense
2
0
Fork 0
mirror of https://github.com/MODSetter/SurfSense.git synced 2026-10-21 22:01:21 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 5
SurfSense/surfsense_local/backend/scripts
History
Exact
DESKTOP-RTLN3BA\$punk aefef6bdc0 perf(local): faster Word reads, cheapest-first uploads, and a still-reading reply 
- Docling's Word reader looked up each run's paragraph style by scanning the
  styles part, 41,671 times on a 292 KB contract. A per-conversion cache of
  DocumentPart.get_style cuts that file from 25.8 s to about 5.4 s with
  byte-identical output on 804 DOCX files. It is scoped to one conversion
  and one thread, and restored on failure.
- An upload's files are queued text first, then smallest first, then
  recordings, so the quickest files in a dropped batch are usable first.
  Priorities are unchanged.
- When a Basic turn finds nothing because its ticked sources are still being
  read, the model says so instead of saying the knowledge base does not
  cover the question. The chat eval gains two cases for it.
2026-10-16 00:32:11 -07:00
..
capability_list data(capability): commit the OpenRouter sweep's input and its two runs 2026-10-11 17:55:51 -07:00
chat_eval perf(local): faster Word reads, cheapest-first uploads, and a still-reading reply 2026-10-16 00:32:11 -07:00
local_manifest feat(local): add Canary 180M Flash as a second speech model 2026-10-16 03:14:00 +05:30
notices fix(local): list the Silero VAD pack in the notices, with its own licence 2026-10-15 02:31:28 +02:00
remote_manifest feat(llm): refresh the remote model manifest and list what a refresh reads differently 2026-10-04 09:46:55 -07:00
retrieval_eval feat(local): choose the embedding model in onboarding 2026-10-02 01:45:10 +05:30
audit_gate.py refactor(llm): serve the local catalog at /llm/catalog/local 2026-09-23 22:13:52 +05:30
build_binaries.py fix(local): fail Mac freezes that cannot load sqlite-vec 2026-09-18 19:19:49 +02:00
catalog_rows.py fix(capability): let a later sweep correct a variant's or an assumption's row 2026-10-07 16:12:49 -07:00
fetch_bundled_voice.py feat(local): run image models built from several files, only while Studio needs them 2026-09-25 04:16:08 +02:00
fetch_docling_models.py fix(packaging): cut the installer payload below the NSIS 2 GB limit 2026-09-15 19:20:40 -07:00
fetch_embedding_model.py feat(local): choose the embedding model in onboarding 2026-10-02 01:45:10 +05:30
fetch_limit_small.py feat(local): measure the lexical-match case with LIMIT-small 2026-09-24 19:49:15 -07:00
fetch_silero_vad.py feat(local): transcribe speech through one gate in the API 2026-10-14 05:52:54 -07:00
measure_figure_ingest.py feat(ingest): keep each source's figures beside its original 2026-10-04 05:25:10 -07:00
refresh_local_manifest.py feat(local): choose the embedding model in onboarding 2026-10-02 01:45:10 +05:30
refresh_remote_manifest.py feat(llm): refresh the remote model manifest and list what a refresh reads differently 2026-10-04 09:46:55 -07:00
run_chat_eval.py feat(local): review chat eval runs with a frontier judge 2026-09-24 18:47:10 -07:00
run_retrieval_eval.py feat(local): choose the embedding model in onboarding 2026-10-02 01:45:10 +05:30
write_capability_list.py feat(capability): rows for the OpenRouter screening sweep and assumed flagships 2026-10-07 14:30:52 -07:00
write_model_notices.py feat(local): generate third-party notices at build time 2026-10-06 18:10:43 +02:00
write_python_notices.py feat(local): generate third-party notices at build time 2026-10-06 18:10:43 +02:00