All issues

Search results

FP8 quantizer: update_tp_plan misses composite configs, expert scales stay replicated under EP

Status: Open.

#48757 In huggingface/transformers;

· malaiwah opened on Sep 13, 2026

get_balanced_memory ignores _no_placement_params, inflating every device's budget (device_map="auto" OOM)

Status: Open.

#48756 In huggingface/transformers;

· malaiwah opened on Sep 13, 2026

Add DeepSeek-V4.1-Flash (deepseek_v41)

Status: Open.

#48755 In huggingface/transformers;

· malaiwah opened on Sep 13, 2026

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

WIPLabel your PR/Issue with WIP for some long outstanding Issues/PRs that are work in progressLabel your PR/Issue with WIP for some long outstanding Issues/PRs that are work in progress

Status: Open.

#48752 In huggingface/transformers;

· efijeniagordonia-create opened on Sep 13, 2026

[serge] integration failure triage - 2026-09-12

Status: Open.

#48749 In huggingface/transformers;

· github-actions opened on Sep 12, 2026

DogeForCausalLM attends to future tokens with the default sdpa backend — SmallDoge/Doge-20M scores ppl 5 instead of 16, and cached and uncached generate() disagree

bug

Status: Open.

#48748 In huggingface/transformers;

· blipbyte opened on Sep 12, 2026

RoFormerForCausalLM, BigBirdForCausalLM, RemBertForCausalLM, MegatronBertForCausalLM attend to future tokens since v5.9.0 — CLM training loss collapses

bug

Status: Open.

#48745 In huggingface/transformers;

· blipbyte opened on Sep 12, 2026

Image processor backends return different keys and dtypes for the same input

Status: Open.

#48738 In huggingface/transformers;

· yupengtang opened on Sep 12, 2026

Follow-up: clean up video_duration handling in hyperclovax_vision_v2 (#44314)

Status: Open.

#48736 In huggingface/transformers;

· jp1924 opened on Sep 12, 2026

Text-to-audio pipeline failed in 5.17.0

bug

Status: Open.

#48732 In huggingface/transformers;

· zhangwei217245 opened on Sep 12, 2026

Loading RTDetrModel from a detection checkpoint (or SEWDForCTC from a SEW-D checkpoint) gives a model with every weight randomly initialized

bug

Status: Open.

#48722 In huggingface/transformers;

· blipbyte opened on Sep 11, 2026

Qwen3.5-9B fine-tuning with run_clm.py on 8x B200 ran the gated DeltaNet layers on the fp32 torch chunk loop because flash-linear-attention was not installed

Status: Open.

#48718 In huggingface/transformers;

· TarzanZhao opened on Sep 11, 2026