Repository Issue Activity (beta)

#1704 RawAudio.toBlob() encodes the entire underlying ArrayBuffer, ignoring TypedArray byteOffset/length

#1734 Chatterbox KV cache leak

#1735 Moonshine ASR pipeline throws 'Unable to decode without a tokenizer' — _call_moonshine calls processor.batch_decode but processor has no tokenizer

#1749 onnx-community/OpenMed-PII-SuperClinical: q8 silently returns near-chance logits; fp16 and q4f16 fail to load

#1759 Importing Transformers.js causes warning in Rspack and Webpack

#1719 [Question] What is the release plan for the next version after 4.2.0(April 23)?

#1331 Embeddings generated different between js and python for some models

#1754 [Bug] model_file_name parameter ignored for encoder-decoder models; hardcoded _quantized suffix cannot be overridden

#1750 WASM beat WebGPU by ~9.6x on a small embedding model — building an open benchmark to find out where else that's true

#1752 [Discussion] AI Agent Collaboration for Browser-Based ML Testing

#1748 Chrome 153 WebNN API: deviceType: 'webnn-npu' rejected, must use 'npu'

#1310 Support for nvidia/parakeet-tdt

#1599 [Performance] Qwen3.5-4B (q4f16) is ~3x slower decode + ~20x slower TTFT vs Qwen3-4B on WebGPU (Transformers.js 4.0.0-next.7)

#1741 [WebGPU] WebGPU decode permanently degrades ~3x at S=2048/4096/6144

#1727 Audit flagged version of adm-zip via [email protected] via @huggingface/[email protected]

#1739 Whisper ASR pipeline leaks ~650 MB of GPU memory per 30 s chunk on WebGPU (both `return_timestamps` modes)

#1736 Unigram tokenizer's CharTrie costs ~200+ MB RSS (~166 MB JS heap) for XLM-R-sized vocabs

#1718 Bump onnxruntime-node to ^1.27 to drop deprecated boolean transitive dependency

#1729 sharp `^0.34.5` range cannot admit patched 0.35.x (libvips CVEs, GHSA-f88m-g3jw-g9cj)

#1728 [WebGPU] embeddinggemma-300m-ONNX q4/q8 produce silently wrong embeddings (cos ~0 vs WASM); fp32 correct

#1726 ParakeetForCTC outputs only blank tokens - empty transcription on all backends (v4.2.0)

#1634 @huggingface/transformers.js 4.0.1: onnxruntime-node 1.24.3 does not ship with x64 darwin (MacOS) bindings.

#1724 LFM uses {%- generation -%} found only in newer jinja not used by transformers.js

#1328 Output Generation Guided by JSON Schema

#1220 Support resuming model downloads from where they were interrupted.