All issues

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

[serge] integration failure triage - 2026-09-19

#48954

· github-actions opened

on Sep 19, 2026

DynamicCache(offloading=True) returns wrong logits after first cache reuse on hybrid Gated-DeltaNet + attention models (e.g. Qwen3.8-27B)

#48947

· com-junkawasaki opened

on Sep 19, 2026

T5 sentencepiece tokenizer without precompiled_charsmap cannot be loaded

#48942

· DRKV333 opened

on Sep 18, 2026

4D mask created under torch.compile when padding_mask is None

#48925

· msnliu opened

on Sep 18, 2026

OPT training with torch.compile decomposes SDPA into Triton kernels due to attention_mask

#48924

· msnliu opened

on Sep 18, 2026

Image processors silently mutate class defaults when instantiated with min_pixels

#48917

· Gracy769 opened

on Sep 18, 2026

MtpModel/MtpLayer topology is too rigid to represent non-uniform MTP architectures

#48913

· soyr-redhat opened

on Sep 17, 2026

MoE models that balance with a router bias never expose their router logits

#48911

· qgallouedec opened

on Sep 17, 2026

Non-persistent buffers materialized uninitialized by from_pretrained (_move_missing_keys_from_meta_to_device)

#48892

· taoshaoyu opened

on Sep 17, 2026

Llama 4 declares `output_router_logits` and `router_aux_loss_coef` but never reads them

#48889

· qgallouedec opened

on Sep 17, 2026

gemma 4 can not load audio file and produce Audio features and audio tokens do not match, tokens: 31, features: 468480

#48887

· chaowenguo opened

on Sep 17, 2026

Four MoE models expose `aux_loss` in their output type and never set it

#48886

· qgallouedec opened

on Sep 17, 2026