Issues · huggingface/transformers · GitHub

Transformers' version 5 is out!
#43489 · LysandreJik opened on Jan 26, 2026
7
Tell Us: What Would Make Trainer Better?
#43595 · SunMarc opened on Jan 29, 2026
16

Labels Milestones

_prefill: cache.get_seq_length() returns device tensor, breaks slice arithmetic for static cache + torch.compile on non-CUDA backends

#46858

· silicon-dispatch-100 opened

on Jun 24, 2026

Qwen3.5 Does not work with variable length inputs using position_ids

#46851

· kolehma8 opened

on Jun 23, 2026

[Qwen3.5] Missing `linear_attn` entries in `base_model_tp_plan` causes OOM and shape error at TP>1

#46846

· tengomucho opened

on Jun 23, 2026

[RT-DETR] TypeError in RTDetrModel.forward when num_feature_levels > len(backbone outputs)

#46832

· c1prk opened

on Jun 23, 2026

4-bit load_in_4bit + device_map='auto' + explicit max_memory still dispatches modules to CPU on a model that fits -> ValueError

#46823

· trevorgordon981 opened

on Jun 22, 2026

ZeRO-3 zero.Init does not partition composite minimax_m3_vl language submodule -> OOM on multi-GPU load

#46822

· trevorgordon981 opened

on Jun 22, 2026

[TimesFM 2.5]:`window_size` argument raises `AttributeError`

#46821

· kanenorman opened

on Jun 22, 2026

Is transformers 5.12.1 requires tokenizers<=0.23.0,>=0.22.0 really right?

#46785

· jrp2014 opened

on Jun 20, 2026

LoRA fine-tuning of an FP8 checkpoint is blocked: `get_peft_model` does not clear `validate_quantization_for_training` (QuantizationMethod.FP8)

#46736

· brunopistone opened

on Jun 17, 2026

LabelSmoother ignores num_items_in_batch, inflating gradients by ~gradient_accumulation_steps with label smoothing

#46731

· Incheonkirin opened

on Jun 17, 2026

Several models hardcode `torch.float64` in their forward pass, crashing on MPS (Apple Silicon)

#46723

· qflen opened

on Jun 17, 2026

Clarify Gemma4 vision bidirectional mask behavior for full vs sliding attention

#46711

· wnma3mz opened

on Jun 17, 2026