Repository Issue Activity (beta)

#47116 Streaming Pipeline Regression in Transformers v5

#47085 Fast (Rust-backed) tokenizers aren't thread-safe anymore

#47067 [i18n-<languageCode>] Translating docs to <languageName>

#39842 Expected behavior of `compute_result` is hard to expect and inconsistent

#47086 Hybrid models: padding is not masked out of linear-attention/conv state on continued forwards (chunked prefill, cache continuation)

#47020 XLMRobertaTokenizer.__init__ passes dict to Unigram(vocab=...) expecting a Sequence

#47110 special_tokens, 'list' object has no attribute 'keys'

#47050 CanineModel fp16 ONNX export fails with dtype mismatch in attention (expected scalar type Float but found Half)

#47068 Potential host synchronization in deepspeed_sp_compute_loss due to Python conditional on CUDA tensor

#47070 ASR pipeline: unconditional torchcodec import crashes on broken torchcodec installs

#47065 DeformableDetrHungarianMatcher and LwDetrHungarianMatcher crash on non-finite cost matrices ("cost matrix is infeasible")

#47095 [Proposal] HeartFlow - Cognitive Engine for HuggingFace Transformers

#46439 Static cache `early_initialization` corrupts linear-attention layers (wrong shapes → `RuntimeError`)

#46441 Qwen3.5 crash with `cache_implementation="static"` — `KeyError: 'linear_attention'`

#46139 Discussion: optional RankSEG-style decoding for Transformers semantic segmentation post-processing

#46215 _sanitize_module_name doesnt handle path traversal chars

#39379 Handling of full_text_row_masked_out_mask in mllama is incorrect.

#47102 [serge] integration failure triage - 2026-07-06

#46962 [Qwen3] `create_masks_for_generate` return type change after #41251

#47091 [serge] integration failure triage - 2026-07-05

#46421 # Cache lazy initialization leads to spurious recompilation with chunked prefill

#46424 Cache config max_cache_length only honored for Quantized Cache

#46894 CohereASR training-loss double-shift bug (same pattern as Moonshine #46784)

#39224 transformers: FlaubertTokenizer: do_lowercase_and_remove_accent: make the logger warning actionable (don't only tell what's wrong, rather suggest what could be done about that)

#46216 is_torch_hpu_available permanently monkeypatches global torch functions