All pull requests

Pull requests

• feat(v3): complete Nemotron 3 and 3.5 embedding integration

  #800·michaelfeil opened Oct 24, 2026Owner·

  ·
  5

• refactor(v3): share dense MLP ownership and precision execution

  #797·michaelfeil opened Oct 24, 2026Owner·

  ·
  5

• feat(libtorch): add optimized FP8 Gemma4 MoE inference

  #776·michaelfeil opened Oct 23, 2026Owner·

  ·
  14

• perf(v3): normalize GDN heads with warp-local reductions

  #772·michaelfeil opened Oct 23, 2026Owner·

  ·
  4

• perf(v3): scan single-sequence OneJev GDN once

  #771·michaelfeil opened Oct 23, 2026Owner·

  ·
  5

• research(v3): prototype opt-in NVFP4 FFN inference for Blackwell

  #770·michaelfeil opened Oct 23, 2026Owner·

  ·
  3

• fix(v3): avoid per-replica checkpoint shard copies

  #769·michaelfeil opened Oct 23, 2026Owner·

  ·
  3

• perf(v3): reuse Hopper GDN prefill for cached suffixes

  #768·michaelfeil opened Oct 23, 2026Owner·

  ·
  7

• perf(v3): accelerate long Qwen3.5 state prefill on Hopper

  #767·michaelfeil opened Oct 23, 2026Owner·

  ·
  9

• Enable text-only CUDA graph capture for multimodal Qwen families

  #687·infinity-collaborator-ci opened Oct 14, 2026Collaborator·

  ·
  4

• fix(args): broadcast environment options across models

  #682·linjiaye929 opened Oct 13, 2026·4 tasks done·

  ·
  5

• fix(api): preserve binary percent-encoded data URI payloads

  #681·linjiaye929 opened Oct 12, 2026·3 tasks done·

  ·
  2

• fix(embeddings): slice Matryoshka dimensions on final axis

  #680·linjiaye929 opened Oct 12, 2026·3 tasks done·

  ·
  2

• fix(engine): preserve single-pass engine iterables

  #679·linjiaye929 opened Oct 12, 2026·3 tasks done·

  ·
  2

• fix(cache): distinguish full input content in cache keys

  #678·linjiaye929 opened Oct 12, 2026·3 tasks done·

  ·
  7

• Scope Python CI and publishing workflows

  #675·michaelfeil opened Oct 10, 2026Owner·

  ·
  3

• Support Qwen3 models: bump transformers to >=4.51 and fix optimum/colpali compatibility

  #674·hassanraha opened Sep 9, 2026·3 tasks done·

  ·
  2

  4

• fix(classifier): adopt the tokenizer pad_token_id when the model config lacks one

  #670·olesxg opened Aug 24, 2026·

  ·
  1

  1

• fix(cache): a failed disk-cache write no longer kills the writer thread

  #669·olesxg opened Aug 24, 2026·

  ·
  1

• fix: prevent persistent 429 responses after model worker failures

  #668·cyb-fox opened Aug 10, 2026·3 tasks done·

  ·
  5

• feat(rerank): add per-model token limits for cross-encoder rerank

  #666·vlsi opened Jun 17, 2026·

  ·
  11

• chore: add publication for linux/arm64 for -cpu containers

  #665·vlsi opened Jun 17, 2026·1 of 3 tasks·

  ·
  1

  10

• fix(deps): bump typer to >=0.20.0 so the CLI works with click >=8.3 (#650)

  #663·Anai-Guo opened Jun 4, 2026·

  ·
  1

  4

• fix(classify): honor raw_scores flag (return logits, softmax only when raw_scores=False)

  #662·Anai-Guo opened Jun 4, 2026·

  ·
  9

• feat: Support Customizable Logging Format in Infinity

  #644·shalousun opened Sep 19, 2025·3 tasks done·

  ·
  1

  3