All pull requests

Pull requests

• [OpenVINO] Add int4 compression tuning for openbmb/MiniCPM-V-4.7-1B and openbmb/MiniCPM-V-4.7-35B-A3B

  #2056·avasenin-14 opened Oct 7, 2026·3 tasks·

  ·

• Fix save_pretrained() method of Qwen3-TTS.

  #2055·popovaan opened Oct 7, 2026Collaborator·

  ·
  3

• Support LTX2.5

  #2054·anatyrova opened Oct 7, 2026Collaborator·3 tasks·

  ·
  2

• raise before loading model if transformers not compatible

  #2051·echarlaix opened Oct 6, 2026Collaborator·

  ·
  3

• Add int2 and int3 to the --weight-format CLI parameter

  #2045·andrey-churkin opened Oct 5, 2026Collaborator·3 tasks·

  ·
  11

• Add support of video MiniCPM-V_4_5

  #2044·pavel-esir opened Oct 2, 2026Collaborator·3 tasks done·

  ·
  10

• [OpenVINO] Fix Gemma3n vision embedding overflow on GPU by setting activation scale factor

  #2043·andrew-k-park opened Oct 2, 2026Contributor·1 task done·

  ·
  3

• [OpenVINO] Fix Qwen3-Omni-MoE mRoPE positions for image and audio

  #2042·sgonorov opened Oct 1, 2026Contributor·1 task·

  ·
  2

• Add OVGuard for content moderation, support token classifiers and causal language models

  #2039·p-wysocki opened Oct 1, 2026·2 of 3 tasks·

  ·
  1

• [OpenVINO] Support SenseVoiceSmall

  #2038·popovaan opened Sep 30, 2026Collaborator·2 tasks done·

  ·
  36

• [PoC] Support continious batching for Dflash Qwen3.5

  #2028·rkazants opened Sep 25, 2026Collaborator·3 tasks·

  ·
  1

• Add LTX video 0.9.5 version support

  #2024·JohnLeFeng opened Sep 24, 2026·1 of 3 tasks·

  ·
  7

• [Qwen3-ASR] Ignore RoPE Pattern For Qwen3 ASR Model

  #2014·anzr299 opened Sep 21, 2026Contributor·3 tasks·

  ·
  5

• [OpenVINO] Support PaddleOCR-VL-1.5 with task image-text-to-text

  #2013·openvino-agent opened Sep 21, 2026Contributor·2 tasks done·

  ·

• [OpenVINO] Support LFM2.5-VL-3B with task image-text-to-text

  #2009·openvino-agent opened Sep 18, 2026Contributor·2 tasks done·

  ·

• Add OpenVINO export support for spark2_5 architecture (Spark-X2.5)

  #2008·bharqav opened Sep 18, 2026·2 of 3 tasks·

  ·
  1

  3

• [OpenVINO] Support LFM2.5-VL-3B with task image-text-to-text

  #2005·popovaan-bot opened Sep 17, 2026·2 tasks done·

  ·
  1

• Add slow T5 tokenizer support when transformers < 5

  do not merge
  #2002·JohnLeFeng opened Sep 17, 2026·1 of 3 tasks·

  ·
  3

• Make modeling_visual_language importable on newer transformers (VLLM 0.29 support)

  #1994·3QL opened Sep 12, 2026·

  ·
  2

• Improve OpenVINO export errors for unwritable output filesystems

  #1993·Gunjan00001 opened Sep 11, 2026·

  ·
  7

• Fix of MoE fusion for Granite-MoE.

  #1992·popovaan opened Sep 11, 2026Collaborator·3 tasks·

  ·
  1

• [OpenVINO] Support academic-ds-9B with task text-generation

  #1990·openvino-agent opened Sep 10, 2026Contributor·2 tasks done·

  ·

• [OpenVINO] Support Unlimited-OCR with task image-text-to-text

  #1987·openvino-agent opened Sep 10, 2026Contributor·2 tasks done·

  ·

• [OpenVINO] Qwen3-ASR implement vlm like audio_encoder + text_embedding + language model export

  #1985·as-suvorov opened Sep 10, 2026Collaborator·1 task done·

  ·
  20

• [OpenVINO] Add DFlash 2 two-model export

  #1984·imargulis opened Sep 10, 2026·5 tasks done·

  ·