All pull requests

Pull requests

• Reject num_beam_groups=0 before dividing by it in GenerationConfig::validate

  category: GGUFcategory: sampling
  #4713·Somay-kousis opened Oct 24, 2026Contributor·4 tasks done·

  ·
  1

• [CB] Cache Omni intermediate hidden states for prefix caching

  category: continuous batchingcategory: GGUFcategory: omnino-match-files
  #4711·sgonorov opened Oct 23, 2026Contributor·4 tasks done·

  ·
  9

• Docs/wwb agent models

  category: WWB
  #4710·sswierze opened Oct 23, 2026Contributor·4 tasks·

  ·
  2

• [VLM] Support Qwen3.8-Flash-Next for SDPA

  category: LLMcategory: visual language
  #4707·yatarkan opened Oct 23, 2026Contributor·1 of 4 tasks·

  ·
  4

• [Docs] document SmolLM and SmolLM2 checkpoints

  category: GH Pages Docs
  #4706·PrashantTakale369 opened Oct 23, 2026·2 of 4 tasks·

  ·
  4

• Limit num_assistant_tokens

  bugcategory: CPP APIcategory: GGUFcategory: GH Pages Docscategory: sampling
  #4704·pavel-esir opened Oct 23, 2026Contributor·4 tasks done·

  ·

  · 2026.5
  3

• [CB] Drop queued requests when add_request throws in speculative decoding and prompt lookup

  category: continuous batchingcategory: GGUFcategory: LLMcategory: prompt lookupcategory: speculative decodingcategory: visual language
  #4703·sgonorov opened Oct 23, 2026Contributor·4 tasks done·

  ·
  3

• [VLM] Rename VisionRegistry to EmbeddingsRegistry

  category: continuous batchingcategory: GGUFcategory: GHAcategory: omnicategory: prompt lookupcategory: speculative decodingcategory: visual languageno-match-files
  #4702·sgonorov opened Oct 22, 2026Contributor·4 tasks done·

  ·
  1

  

• [ASR] Add batched inference for SenseVoiceSmall

  category: automatic speech recognitioncategory: CPP APIcategory: GGUFcategory: GH Pages Docscategory: Python API
  #4701·byekrang opened Oct 22, 2026Contributor·4 tasks done·

  ·
  4

• Fix LCMScheduler::threshold_sample returning an empty vector

  category: image generation
  #4700·Daksha1611 opened Oct 22, 2026·2 of 4 tasks·

  ·
  1

• test(js): enable perfMetrics coverage on macOS

  category: JS APIdo_not_mergedo_not_reviewno-match-files
  #4697·Retribution98 opened Oct 22, 2026Contributor·4 tasks·

  ·
  4

• Streaming input poc

  category: automatic speech recognitioncategory: CPP APIcategory: GH Pages Docscategory: Python APIno-match-files
  #4696·kbalka opened Oct 22, 2026·4 tasks·

  ·
  11

• Bump the npm-dependencies group across 1 directory with 3 updates

  category: JS APIdependenciesjavascript
  #4694·dependabot[bot] opened Oct 22, 2026Bot·

  ·

• [Docs] Fix usage of the text2video and taylorseer samples

  category: Video generation samples
  #4693·jayyyyqwq opened Oct 21, 2026·2 of 4 tasks·

  ·
  6

• LTX2.* RoPE cos, sin

  category: CPP APIcategory: Video generation samplescategory: video generation
  #4692·anatyrova opened Oct 21, 2026Contributor·4 tasks·

  ·
  11

• Fix byte count copied when growing u4 KV cache on CPU

  category: continuous batching
  #4691·Daksha1611 opened Oct 21, 2026·2 of 4 tasks·

  ·
  1

  2

  

• [VLM] Fix chat history handling and template/video processing

  category: GGUFcategory: tokenizerscategory: visual languageno-match-files
  #4689·mvafin opened Oct 21, 2026Contributor·4 tasks done·

  ·
  20

• Enable Qwen-Image 2.1 CI tests

  category: GGUFcategory: GHAcategory: image generation
  #4688·likholat opened Oct 21, 2026Contributor·2 of 4 tasks·

  ·
  4

• [VLM] Avoid duplicate BOS tokens in paged-attention chat

  category: continuous batchingcategory: GGUFcategory: visual language
  #4687·mvafin opened Oct 21, 2026Contributor·3 of 4 tasks·

  ·
  11

• Gemma4 mtp continuous batching

  category: continuous batchingcategory: CPP APIcategory: GGUFcategory: GH Pages Docscategory: GHAcategory: LLMcategory: speculative decodingcategory: visual languageno-match-files
  #4686·sbalandi opened Oct 21, 2026Contributor·4 tasks·

  ·
  14

• [GGUF] Reuse shared LLM feature tests for frontend models

  category: GGUFcategory: GHAcategory: LLM
  #4684·mvafin opened Oct 21, 2026Contributor·4 tasks done·

  ·
  13

• [GGUF] Load supported multimodal projectors through VLMPipeline

  category: cmake / buildcategory: continuous batchingcategory: CPP APIcategory: GGUFcategory: GH Pages Docscategory: LLMcategory: visual languageno-match-files
  #4683·mvafin opened Oct 21, 2026Contributor·3 of 4 tasks·

  ·
  43

• Add per-position logprobs for requested token ids

  category: continuous batchingcategory: CPP APIcategory: GGUFcategory: GH Pages Docscategory: LLMcategory: Python APIcategory: samplingcategory: speculative decodingcategory: visual languageno-match-files
  #4682·sergey-zinchenko opened Oct 21, 2026·3 of 4 tasks·

  ·
  35

• Draft pr for a public build, do not review

  category: cmake / buildcategory: continuous batchingcategory: CPP APIcategory: llm_benchcategory: Python APIcategory: speculative decodingcategory: VLM samplesdo_not_mergedo_not_reviewdo-not-mergeno-match-files
  #4680·zaixing-wang opened Oct 21, 2026·

  ·
  39

• [wwb] Add Whisper support to speech recognition evaluator

  category: GHAcategory: WWB
  #4673·mateusztabaka opened Oct 20, 2026·4 tasks·

  ·
  21