Repository navigation

Issues

New issue

Transformers' version 5 is out!
#43489 · LysandreJik opened on Jan 26, 2026
8
Tell Us: What Would Make Trainer Better?
#43595 · SunMarc opened on Jan 29, 2026
16

ProcessorMixin.apply_chat_template treats template kwargs unused by the template as processor kwargs
huggingface/transformers#49762 · opened on Oct 20, 2026
1
SmolVLMProcessor.apply_chat_template always warns "Kwargs passed to processor.__call__ have to be in processor_kwargs"
huggingface/transformers#49761 · opened on Oct 20, 2026
ProcessorMixin.apply_chat_template drops explicit processor_kwargs when an extra kwarg is passed
huggingface/transformers#49760 · opened on Oct 20, 2026
2
[serge] integration failure triage - 2026-10-19
huggingface/transformers#49751 · opened on Oct 19, 2026
Qwen4-Exp: cache.reset() leaves stale position_ids when reusing DynamicCache or StaticCache
huggingface/transformers#49745 · opened on Oct 19, 2026
1
[serve] Sampling not enabled for positive temperature
huggingface/transformers#49734 · opened on Oct 19, 2026
[Bug] Qwen3-MoE router auxiliary loss is multiplied by the DDP world size in Trainer
huggingface/transformers#49727 · opened on Oct 19, 2026
2
Torchvision backend ignores EXIF orientation for path, URL and base64 images
huggingface/transformers#49701 · opened on Oct 17, 2026
3
prefill_chunk_size drops image inputs, VLMs generate without the image
huggingface/transformers#49700 · opened on Oct 17, 2026
3
input_projections are not registered as PyTorch modules, causing missing parameters and CUDA device mismatch
huggingface/transformers#49698 · opened on Oct 17, 2026
3
Generation crashes with encoder_repetition_penalty when encoder_outputs are provided without input_ids
huggingface/transformers#49695 · opened on Oct 17, 2026
4
TextStreamer re-decodes the whole line on every token (quadratic on long lines)
huggingface/transformers#49675 · opened on Oct 16, 2026
1
Chat parsing: json parser's string_delims doesn't handle escaped delimiters
huggingface/transformers#49674 · opened on Oct 16, 2026
3
Quick survey: your experience as a PyTorch Ecosystem maintainer
huggingface/transformers#49672 · opened on Oct 16, 2026
GGUF tokenizer ignores tokenizer.ggml.add_bos_token, so the default encoding drops BOS
huggingface/transformers#49647 · opened on Oct 15, 2026
1
BatchEncoding.convert_to_tensors copies the whole batch to check if it's empty (~2x slower collation)
huggingface/transformers#49645 · opened on Oct 15, 2026
2
Gemma4-12B Training Failed
huggingface/transformers#49630 · opened on Oct 15, 2026
1
[i18n-ur] Translating docs to Urdu (اردو)
huggingface/transformers#49626 · opened on Oct 15, 2026
1
Janus / DeepSeek-VL image processors ignore background_color; deepseek-vl-7b-chat's low-res branch pads with 127 instead of its saved (122, 116, 104)
huggingface/transformers#49625 · opened on Oct 15, 2026
4
_patch_mistral_regex treats a missing transformers_version as "Mistral": false positive on local DeBERTa-v3, and following its advice breaks the tokenizer
huggingface/transformers#49623 · opened on Oct 15, 2026
4
DebertaV2Tokenizer drops SentencePiece NFKC normalization in v5: ligatures, fullwidth and circled characters encode to [UNK]
huggingface/transformers#49622 · opened on Oct 15, 2026
4
SAM 3: text-only prompts drop the geometry CLS token that Meta's implementation always adds
huggingface/transformers#49592 · opened on Oct 14, 2026
3
import transformers spends 1-3 s in an eager importlib.metadata.packages_distributions() when torch >= 2.14 is installed (Python 3.12)
huggingface/transformers#49591 · opened on Oct 14, 2026
5
Tokenizer loses its normalizer from tokenizer.json in 5.18
huggingface/transformers#49578 · opened on Oct 13, 2026
SDPA: enable_gqa=True silently falls back to the math kernel at batch 1 on torch < 2.14 without Flash Attention
huggingface/transformers#49564 · opened on Oct 13, 2026
2

Navigation Menu

Repository navigation

Issues

Issues

ProcessorMixin.apply_chat_template treats template kwargs unused by the template as processor kwargs

SmolVLMProcessor.apply_chat_template always warns "Kwargs passed to processor.call have to be in processor_kwargs"

ProcessorMixin.apply_chat_template drops explicit processor_kwargs when an extra kwarg is passed

[serge] integration failure triage - 2026-10-19

Qwen4-Exp: cache.reset() leaves stale position_ids when reusing DynamicCache or StaticCache

[serve] Sampling not enabled for positive temperature

[Bug] Qwen3-MoE router auxiliary loss is multiplied by the DDP world size in Trainer

Torchvision backend ignores EXIF orientation for path, URL and base64 images

prefill_chunk_size drops image inputs, VLMs generate without the image

input_projections are not registered as PyTorch modules, causing missing parameters and CUDA device mismatch

Generation crashes with encoder_repetition_penalty when encoder_outputs are provided without input_ids

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

Chat parsing: `json` parser's `string_delims` doesn't handle escaped delimiters

Quick survey: your experience as a PyTorch Ecosystem maintainer

GGUF tokenizer ignores tokenizer.ggml.add_bos_token, so the default encoding drops BOS

BatchEncoding.convert_to_tensors copies the whole batch to check if it's empty (~2x slower collation)

Gemma4-12B Training Failed

[i18n-ur] Translating docs to Urdu (اردو)

Janus / DeepSeek-VL image processors ignore `background_color`; deepseek-vl-7b-chat's low-res branch pads with 127 instead of its saved (122, 116, 104)

_patch_mistral_regex treats a missing transformers_version as "Mistral": false positive on local DeBERTa-v3, and following its advice breaks the tokenizer

DebertaV2Tokenizer drops SentencePiece NFKC normalization in v5: ligatures, fullwidth and circled characters encode to [UNK]

SAM 3: text-only prompts drop the geometry CLS token that Meta's implementation always adds

import transformers spends 1-3 s in an eager importlib.metadata.packages_distributions() when torch >= 2.14 is installed (Python 3.12)

Tokenizer loses its normalizer from tokenizer.json in 5.18

SDPA: enable_gqa=True silently falls back to the math kernel at batch 1 on torch < 2.14 without Flash Attention