Issues

• Two copies of onnxruntime's Tensor class in the web build: wrapping a fromGpuBuffer tensor makes dispose() recurse forever

  huggingface/transformers.js#1840 · Filter by author daveycodez opened on Oct 30, 2026

• Bump onnxruntime-web to ≥ 1.31.1: sliding-window attention gets flash attention on WebGPU

  huggingface/transformers.js#1839 · Filter by author daveycodez opened on Oct 30, 2026

• Prefix caching across generate() calls: DynamicCache.crop and state snapshots for hybrid models

  huggingface/transformers.js#1838 · Filter by author daveycodez opened on Oct 30, 2026
  1

• TextStreamer decodes the whole line on every token (quadratic on long lines)

  huggingface/transformers.js#1837 · Filter by author daveycodez opened on Oct 30, 2026

• Feature request: chunked prefill in generate() (e.g. prefill_chunk_size) — reading long prompts in one pass crashes iOS Safari

  huggingface/transformers.js#1836 · Filter by author daveycodez opened on Oct 29, 2026
  1

• Use onnxruntime-web's JSPI build when the browser supports JSPI (much lower memory in Safari)

  huggingface/transformers.js#1835 · Filter by author daveycodez opened on Oct 29, 2026

• random.seed() still collapses large negative integer seeds to seed(0)

  huggingface/transformers.js#1833 · Filter by author michaelkillgta opened on Oct 29, 2026

• Proposal: a same-origin inference broker shared by browser tabs

  huggingface/transformers.js#1831 · Filter by author jabrailkhalil opened on Oct 29, 2026

• mean_pooling is slower than the model forward pass on GPU (stride-embedDim loop)

  huggingface/transformers.js#1830 · Filter by author tfriedel opened on Oct 29, 2026

• iOS Safari: tab is killed at the end of the first download of a large .onnx_data file (the file is held in memory about twice)

  huggingface/transformers.js#1827 · Filter by author daveycodez opened on Oct 29, 2026
  1

• Injecting a runtime via Symbol.for('onnxruntime') still loads the bundled onnxruntime-node, which crashes next to a different onnxruntime-node version

  huggingface/transformers.js#1813 · Filter by author iliev-blnk opened on Oct 22, 2026

• Qwen2VLImageProcessor merges several images into one (same size) or throws "offset is out of bounds" (different sizes)

  huggingface/transformers.js#1810 · Filter by author ranacseruet opened on Oct 21, 2026
  2

• Auxiliary operator sessions ignore the model device and cannot be configured (TensorOpRegistry.session_options has no executionProviders and is not exported)

  huggingface/transformers.js#1806 · Filter by author jakesalvesen opened on Oct 20, 2026
  1

• Qwen3.5 (Qwen VL family): text-only generate() with an external past_key_values calls the vision encoder

  huggingface/transformers.js#1802 · Filter by author toomore opened on Oct 17, 2026

• Add support for SAM 3 (Sam3Model): text-prompted concept segmentation

  new model
  huggingface/transformers.js#1800 · Filter by author AModSoul opened on Oct 15, 2026

• [Bug]VoxtralRealtime: generate() decodes one step after the audio runs out and streams a junk token

  bug
  huggingface/transformers.js#1794 · Filter by author jiangzhuo opened on Oct 8, 2026
  1

• Does Whisper Support MOV (.mov) Video Input?

  question
  huggingface/transformers.js#1786 · Filter by author limxinying0720 opened on Oct 6, 2026
  1

• Compiler Errors when using --skipLibcheck false

  bug
  huggingface/transformers.js#1784 · Filter by author dpfens opened on Sep 30, 2026

• createInferenceSession() memoises a rejected session promise, breaking every later session creation in the process

  huggingface/transformers.js#1767 · Filter by author angri450 opened on Sep 8, 2026
  2

• [Bug] model_file_name parameter ignored for encoder-decoder models; hardcoded _quantized suffix cannot be overridden

  bug
  huggingface/transformers.js#1754 · Filter by author Conlactor opened on Aug 22, 2026
  1

• WASM beat WebGPU by ~9.6x on a small embedding model — building an open benchmark to find out where else that's true

  enhancement
  huggingface/transformers.js#1750 · Filter by author Sallamhamza opened on Aug 17, 2026
  4

• onnx-community/OpenMed-PII-SuperClinical: q8 silently returns near-chance logits; fp16 and q4f16 fail to load

  huggingface/transformers.js#1749 · Filter by author matthewchung74 opened on Aug 17, 2026
  3

• Chrome 153 WebNN API: deviceType: 'webnn-npu' rejected, must use 'npu'

  huggingface/transformers.js#1748 · Filter by author yzhou51 opened on Aug 15, 2026
  4

• [WebGPU] WebGPU decode permanently degrades ~3x at S=2048/4096/6144

  huggingface/transformers.js#1741 · Filter by author kylo5aby opened on Aug 12, 2026

• Whisper ASR pipeline leaks ~650 MB of GPU memory per 30 s chunk on WebGPU (both return_timestamps modes)

  huggingface/transformers.js#1739 · Filter by author maboa opened on Aug 3, 2026
  1