Issues

• [Bug]: StructuredOutputConfig(json_schema=...) allows unbounded whitespace; a small model loops on it until max_new_tokens

  openvinotoolkit/openvino.genai#4885 · Filter by author ml-mind opened on Nov 10, 2026
  3

• [Bug] C API: set_eos_token_id() breaks validate() and generate(), set_stop_token_ids(NULL, n) segfaults

  openvinotoolkit/openvino.genai#4865 · Filter by author yasharth-0910 opened on Nov 7, 2026

• [Bug] Image generation crashes the process (integer division by zero) when num_inference_steps=0 or num_images_per_prompt=0

  openvinotoolkit/openvino.genai#4841 · Filter by author Aditya365x opened on Nov 5, 2026
  2

  

• Llama3PythonicToolParser merges parallel calls, drops non-string arguments and makes calls out of bracketed text

  openvinotoolkit/openvino.genai#4831 · Filter by author Somay-kousis opened on Nov 4, 2026

• generate(ChatHistory) after a plain generate() crashes in the sampler on a stateful pipeline

  openvinotoolkit/openvino.genai#4822 · Filter by author Somay-kousis opened on Nov 4, 2026
  2

• VLLMParserWrapper raises a cast error when vLLM's extract_reasoning returns None

  openvinotoolkit/openvino.genai#4820 · Filter by author Somay-kousis opened on Nov 4, 2026

• [wwb] Divergency crashes when one answer is empty and the other has one token

  openvinotoolkit/openvino.genai#4808 · Filter by author soujasK opened on Nov 3, 2026
  3

  

• [Bug][NPU] WhisperPipeline.generate() with return_timestamps=True returns empty or wrong text on NPU

  category: NPU
  openvinotoolkit/openvino.genai#4806 · Filter by author jcork-intel opened on Nov 3, 2026
  

• Python Parser subclass in GenerationConfig.parsers fails with 'parse method not implemented' once Python drops it

  openvinotoolkit/openvino.genai#4788 · Filter by author Somay-kousis opened on Nov 1, 2026

• [Bug] set_eos_token_id(-1) adds -1 to stop_token_ids, so validate() rejects every generate call when no EOS is defined

  openvinotoolkit/openvino.genai#4776 · Filter by author Daksha1611 opened on Oct 30, 2026

• [Feature][NPU] Support batch > 1 for LLM decoding - NPU scales linearly with batch (8.00x measured), decode is weight-bandwidth bound

  openvinotoolkit/openvino.genai#4764 · Filter by author Asahi-Prv opened on Oct 29, 2026
  3

  

• Feature Request: logprobs for a requested set of token ids (logprob_token_ids)

  openvinotoolkit/openvino.genai#4736 · Filter by author sergey-zinchenko opened on Oct 27, 2026

• [Feature] Continuous batching: store only the window for sliding-attention layers

  PSE
  openvinotoolkit/openvino.genai#4734 · Filter by author aweussom opened on Oct 26, 2026
  1

  

• Llama3JsonToolParser raises on text with braces that are not a JSON tool call

  openvinotoolkit/openvino.genai#4730 · Filter by author Somay-kousis opened on Oct 26, 2026
  5

  

• [Bug] NaN WWB similarity incorrectly passes accuracy check

  openvinotoolkit/openvino.genai#4729 · Filter by author PrashantTakale369 opened on Oct 26, 2026
  1

• [Bug] LCMScheduler::step reads empty buffers for prediction_type "sample" / "v_prediction"

  openvinotoolkit/openvino.genai#4717 · Filter by author Daksha1611 opened on Oct 26, 2026
  5

  

• [VLM][NPU] Qwen3-VL: a prompt without an image fails with "roi_end <= max_dim" when MAX_PROMPT_LEN is raised

  openvinotoolkit/openvino.genai#4709 · Filter by author Jcollier0120 opened on Oct 23, 2026
  2

  

• [VLM][NPU] After one generate() fails part-way, every later generate() fails with "prompt_ids.get_size() >= tokenized_history.size()"

  openvinotoolkit/openvino.genai#4708 · Filter by author Jcollier0120 opened on Oct 23, 2026

• [Bug] LCMScheduler::threshold_sample always returns an empty vector (thresholding=true)

  openvinotoolkit/openvino.genai#4699 · Filter by author Daksha1611 opened on Oct 22, 2026

• [Bug] KV cache growth with u4 precision on CPU memcpy's 2x the packed buffer size

  openvinotoolkit/openvino.genai#4690 · Filter by author Daksha1611 opened on Oct 21, 2026
  2

  

• VLMPipeline: CDPruner creates its own ov::Core and compiles on the pipeline device during generate(), even with pruning_ratio=0

  category: NPU
  openvinotoolkit/openvino.genai#4671 · Filter by author karanshukla opened on Oct 19, 2026
  3

  

• Qwen Image 2.1 clone loses reshaped VAE encoder dimensions

  category: image generation
  openvinotoolkit/openvino.genai#4662 · Filter by author YugShah1510 opened on Oct 17, 2026
  

• Llama3PythonicToolParser drops arguments written with single quotes

  openvinotoolkit/openvino.genai#4661 · Filter by author Somay-kousis opened on Oct 17, 2026
  2

  

• [Request ] add black-forest-labs/flux-3-image-klein distrill version 16 steps

  category: image generation
  openvinotoolkit/openvino.genai#4653 · Filter by author ak-sangal opened on Oct 16, 2026
  2

  

• Tokenizer.encode narrows max_length to int32: 2**32 + 2 truncates to 2 tokens, sys.maxsize crashes

  openvinotoolkit/openvino.genai#4650 · Filter by author Somay-kousis opened on Oct 15, 2026
  2