Search results

Add Talkie (talkie-1930-13b)

Status: Open.

#47618 In huggingface/transformers;

· ethanelasky opened on Jul 29, 2026

1 comment

BOS and EOS token ID not within vocabulary warning for siglip2-base-patch16-naflex

bug

Status: Open.

#47612 In huggingface/transformers;

· willxxy opened on Jul 29, 2026

22 linked PRs

1 comment

[serge] integration failure triage - 2026-07-28

Status: Open.

#47605 In huggingface/transformers;

· github-actions opened on Jul 28, 2026



Importing AutoImageProcessor fails when PyTorch is built without torch.distributed

Status: Open.

#47603 In huggingface/transformers;

· hlky opened on Jul 28, 2026

22 linked PRs

2 comments

CodeLlamaTokenizer FIM/infilling drops correct leading whitespace on decode

bug

Status: Open.

#47602 In huggingface/transformers;

· Yashash-Yallapragada opened on Jul 28, 2026

3 comments

Repetition penalty is gauge-dependent. Proposing opt-in normalized variant.

Feature requestRequest for a new featureRequest for a new feature

Status: Open.

#47595 In huggingface/transformers;

· captainpete opened on Jul 28, 2026

2 comments

NemotronH: no longer respected after #47452 (unconditionally fetches Hub kernel)

Status: Open.

#47577 In huggingface/transformers;

· albertvillanova opened on Jul 27, 2026

3 comments

Wav2Vec2 tokenizer rebuilds special-token lists per token in decode paths; phoneme variant carries a filter that never matches

Status: Open.

#47556 In huggingface/transformers;

· ishan-1010 opened on Jul 26, 2026

1

Mamba2-family cuda_kernels_forward decode step is broken since #47452: missing seq-dim squeeze

Status: Open.

#47532 In huggingface/transformers;

· qgallouedec opened on Jul 25, 2026

1

1 comment

Phi3/Phimoe: prepare_inputs_for_generation forwards logits_to_keep=None into forward() -> 4-dim logits -> torch.multinomial crash

bug

Status: Open.

#47530 In huggingface/transformers;

· pwasiewi opened on Jul 25, 2026

22 linked PRs

1 comment

Transformers 5.13 CPU memory growth loading Qwen3-235B-A22B with DeepSpeed ZeRO-3; 4.51.0 remains bounded

bug

Status: Open.

#47514 In huggingface/transformers;

· narankhuut opened on Jul 24, 2026

1

1 comment

MistralCommonBackend API documentation appears to render the dummy implementation instead of the actual class

Status: Open.

#47504 In huggingface/transformers;

· samyuktahegde opened on Jul 23, 2026

1

1 comment

DataCollatorWithFlattening uses 0-based position_ids, silently degrading RoBERTa-family models

bug

Status: Open.

#47496 In huggingface/transformers;

· tomaarsen opened on Jul 23, 2026

1

3 comments

[BUG] CANINE: dynamic ONNX export fixes sequence length to example input

bug

Status: Open.

#47491 In huggingface/transformers;

· sgpublic opened on Jul 23, 2026

1

2 comments

CodeLlama tokenizer drops leading whitespace on decode round-trip

bug

Status: Open.

#47487 In huggingface/transformers;

· SuryanshSS1011 opened on Jul 23, 2026

1

Regarding security issue i have reported

Status: Open.

#47477 In huggingface/transformers;

· liyander opened on Jul 22, 2026

SentencePieceBackend should decode byte-fallback tokens

bug

Status: Open.

#47473 In huggingface/transformers;

· toslunar opened on Jul 22, 2026

1

6 comments

ESMFold throws error when explicitly cast to fp16

bug

Status: Open.

#47470 In huggingface/transformers;

· crackalamoo opened on Jul 22, 2026

22 linked PRs

3 comments

Qwen3.5-MoE (Qwen3_5MoeExperts) data-dependent expert dispatch breaks torch.utils.checkpoint reproducibility — CheckpointError / silent NaN gradients

Status: Open.

#47467 In huggingface/transformers;

· sternryan opened on Jul 22, 2026

1

The kernel's version doesn't properly check.

bug

Status: Open.

#47455 In huggingface/transformers;

· Mohsen-Liaghat opened on Jul 21, 2026

1

2 comments

gemma4 12B arch is not recognized!

bug

Status: Open.

#47448 In huggingface/transformers;

· HuangChiEn opened on Jul 21, 2026

2 comments

Added-token encoder cache goes stale when tokenizers mutate _added_tokens_decoder directly, giving ids that do not round-trip

Status: Open.

#47439 In huggingface/transformers;

· ishan-1010 opened on Jul 21, 2026

1

1 comment

DynamicCache.crop(0) retains the full KV cache and oversized negative crops retain stale tokens

Status: Open.

#47433 In huggingface/transformers;

· HoneyTyagii opened on Jul 21, 2026

4 comments

tokenizers<=0.23.0 upper bound blocks the only existing 0.23.x release

bug

Status: Open.

#47429 In huggingface/transformers;

· EfiSiapiti opened on Jul 20, 2026

33 linked PRs

4 comments

_is_hf_initialized per-param check doesn't skip redundant init for built-in models on non-CUDA hardware

bug

Status: Open.

#47427 In huggingface/transformers;

· gygdh-001 opened on Jul 20, 2026