All pull requests

Pull requests

• fix(distributed): bind TPU ranks to device index 0, not LOCAL_RANK

  #49543·tengomucho opened Oct 12, 2026Member·

  ·
  2

• Add a ring-buffer layer for static caches' sliding windows

  #49542·IlyasMoutawwakil opened Oct 12, 2026Member·

  ·
  2

• [docs] Hardware section

  #49541·michaelbenayoun opened Oct 12, 2026Member·

  ·
  7

• Fix code_revision propagation and model_kwargs support in pipeline

  #49539·NabiBukhsh-AI opened Oct 12, 2026·5 tasks done·

  ·
  1

  1

• fix(ci): harden GitHub Actions workflows (#49357)

  #49538·hf-security-analysis[bot] opened Oct 12, 2026Bot·

  ·
  3

• Draft PR to fix thinking -> reasoning_content

  #49535·Rocketknight1 opened Oct 12, 2026Member·

  ·
  2

• add systemone unit tests

  #49534·SunMarc opened Oct 12, 2026Member·

  ·
  2

• fix(ci): harden GitHub Actions workflows (#49532)

  #49533·hf-security-analysis[bot] opened Oct 12, 2026Bot·

  ·
  3

• CI: Model Integration Contracts

  #49532·tarekziade opened Oct 12, 2026Collaborator·

  ·
  2

  

• feat: Add MTP support to Mellum2/2.1

  #49531·shadeMe opened Oct 12, 2026Contributor·3 of 5 tasks·

  ·
  1

• Subconfig creation in testers

  #49528·zucchini-nlp opened Oct 12, 2026Member·

  ·
  2

• build and expand once the tp/ep_plan

  #49527·3outeille opened Oct 12, 2026Member·

  ·
  2

• [Qwen2Audio] Fix dtype mismatch in legacy audio merge

  #49526·syc632 opened Oct 12, 2026·3 tasks done·

  ·
  1

• Gate enable_gqa on fused SDPA kernel support

  #49521·Jose-Ribeir opened Oct 12, 2026·5 of 6 tasks·

  ·
  1

  7

• Dispatch BarthezTokenizer on vocabulary shape for BPE checkpoints

  #49520·Irfanwani opened Oct 12, 2026·

  ·
  1

  1

• Declare AXK1 routing config fields so configs that omit them still expose them

  #49518·hmellor opened Oct 12, 2026Member·

  ·
  3

• Make T5-style compute_bias return a contiguous position_bias so SDPA can use fused kernels

  #49517·IvenHsu01 opened Oct 12, 2026·1 of 6 tasks·

  ·
  1

  3

• Preserve RoPE base parameters in the rope_scaling setter

  #49515·jinwukong opened Oct 12, 2026·3 of 6 tasks·

  ·
  1

  4

• [CB] Fix the CPU memory budget used to size the paged cache

  #49512·jiqing-feng opened Oct 12, 2026Contributor·

  ·
  2

• Preserve resolved remote backend references in AutoVideoProcessor

  #49511·DimensionSTP opened Oct 12, 2026Contributor·3 of 5 tasks·

  ·
  1

  6

• Complete PR #49285: remove unused device=None from rope __init__ in 7 remaining model families

  #49508·Ayoubhm07 opened Oct 12, 2026·

  ·
  1

• [serge] Fix 2 integration tests for model pegasus failing with other (other (2))

  #49507·sergereview[bot] opened Oct 11, 2026Bot·

  ·
  2

• [fix][Qwen3-Omni] Respect generation_mode in audio generation

  #49505·Sunt-ing opened Oct 11, 2026Contributor·2 of 5 tasks·

  ·
  1

• [Cache] Fix get_max_length on empty cache and add missing properties to EncoderDecoderCache

  #49504·Saauf opened Oct 11, 2026·3 of 5 tasks·

  ·
  4

• [fix][ASR] Fix Qwen3 ASR pipeline transcription

  #49503·Sunt-ing opened Oct 11, 2026Contributor·2 of 5 tasks·

  ·
  2