All pull requests

Pull requests

• fix(extraction): keep text inside infographic boxes when infographic extraction is off

  #2839·jioffe502 opened Nov 7, 2026Collaborator·

  ·
  1

  

• Add opt-in NRL_NVTX profiling ranges

  #2838·jioffe502 opened Nov 7, 2026Collaborator·

  ·
  1

  

• Nemotron 3.5 vLLM embed: AsyncLLM with two UDF threads per actor

  #2837·jioffe502 opened Nov 7, 2026Collaborator·

  ·
  1

  

• Nemotron 3.5 vLLM embed: per-image vision tower with flash attention

  #2836·jioffe502 opened Nov 7, 2026Collaborator·

  ·
  1

  

• feat(extraction): add Parse 3.5 postprocessing and chart routing

  #2835·jiaruic09 opened Nov 6, 2026Contributor·3 tasks done·

  ·
  1

• fix(embed): auto-select NVFP4 Nemotron embed checkpoint in single-node batch mode

  #2834·jioffe502 opened Nov 5, 2026Collaborator·3 tasks done·

  ·
  9

  

• Deduplicate HF package publications and report obsolete nightlies

  #2833·charlesbluca opened Nov 5, 2026Collaborator·3 tasks done·

  ·

• build: add opt-in Torch 2.13 OCR prereleases

  #2832·ChrisJar opened Nov 5, 2026Collaborator·

  ·
  1

• fix(vdb): keep LanceDB stream input errors and clean up files from failed worker commits

  #2830·jioffe502 opened Nov 4, 2026Collaborator·3 tasks done·

  ·
  18

  

• fix(ingest): treat existing files with glob characters in their names as literal paths

  #2829·pcbeingused333 opened Nov 4, 2026·2 of 3 tasks·

  ·
  1

  4

• Render each PDF page once in pdf_extraction

  #2828·jioffe502 opened Nov 4, 2026Collaborator·

  ·
  7

  

• Report Parse page completion and fix local vLLM progress failures

  #2825·charlesbluca opened Nov 2, 2026Collaborator·3 tasks done·

  ·
  1

• Pin LanceDB to 0.39.0 for retrieval accuracy

  #2823·charlesbluca opened Nov 2, 2026Collaborator·3 tasks done·

  ·

• Add opt-in page-elements CPU pre/post split (BatchTuningParams.page_elements_cpu_split)

  #2819·jioffe502 opened Oct 31, 2026Collaborator·3 tasks done·

  ·
  21

  

• Add opt-in NRL_CUDA_BLOCKING_SYNC so GPU actors sleep instead of spin while waiting on the GPU

  #2818·jioffe502 opened Oct 31, 2026Collaborator·3 tasks done·

  ·
  18

  

• feat(vdb): write-only DuckDB and Elasticsearch backends on the worker-write contract

  #2813·jioffe502 opened Oct 29, 2026Collaborator·3 tasks done·

  ·
  66

  

• refactor(vdb)!: carry the vector as a top-level record embedding, not metadata["embedding"]

  #2807·jioffe502 opened Oct 23, 2026Collaborator·

  ·
  1

  

• Add dense-search controls and paired exact diagnostics

  python
  #2806·charlesbluca opened Oct 23, 2026Collaborator·3 tasks done·

  ·
  4

• Add bounded embedding-input diagnostics

  python
  #2805·charlesbluca opened Oct 23, 2026Collaborator·3 tasks done·

  ·
  12

• fix(util): keep hostnames starting with v1 in remove_url_endpoints

  #2799·udsy19 opened Oct 21, 2026·

  ·
  5

• fix(service): reject non-ASCII credentials with 401 instead of 500

  #2790·v0ropaev opened Oct 16, 2026·

  ·
  8

• perf(vdb): convert float-list Arrow columns through NumPy when building rows

  #2778·jioffe502 opened Oct 12, 2026Collaborator·1/3·3 tasks done·

  ·
  13

  

• feat(batch): opt-in elastic actor pools from heuristic bounds

  #2775·jioffe502 opened Oct 7, 2026Collaborator·3 tasks done·

  ·
  9

  

• feat(ingest): stage batch ingest rows to Parquet from Ray workers (--stage-dir)

  #2771·jioffe502 opened Oct 5, 2026Collaborator·

  ·

• feat: add scalable Parquet vector indexing with LanceDB and Elasticsearch/cuVS

  #2766·jperez999 opened Oct 5, 2026Collaborator·

  ·
  17