Skip to content

Issues: huggingface/transformers

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

1,006 Open 15,373 Closed
1,006 Open 15,373 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Assignee
Filter by who’s assigned
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Issues list

Missing weights are not properly initialized when using model.from_pretrained() bug
#35437 opened Dec 27, 2024 by YifanXu74
4 tasks done
Memory leak on python 3.10.* bug
#35434 opened Dec 27, 2024 by KhoiTrant68
2 of 4 tasks
1
tokenizers.apply_chat_template with continue_final_message=True with trailing spaces in input bug
#35433 opened Dec 27, 2024 by chuyishang
1 of 4 tasks
apply class transformers.SequenceBiasLogitsProcessor on Qwen model Feature request Request for a new feature
#35432 opened Dec 27, 2024 by buptspig
GPT2Attention() class with _attn() method when add_cross_attention=True and therefore is_cross_attention=True. Feature request Request for a new feature
#35430 opened Dec 27, 2024 by CHLEE-Leo
cannot custom warmup_min_lr of deepspeed lr scheduler bug
#35428 opened Dec 27, 2024 by SeunghyunSEO
2 of 4 tasks
Can't load model from state_dict + config when quantized bug
#35427 opened Dec 27, 2024 by KareemMusleh
1 of 4 tasks
5
model.config.to_diff_dict() delivers different result to model.save_pretrained() bug
#35426 opened Dec 27, 2024 by umarbutler
2 of 4 tasks
DeepSeek V3 Support New model
#35425 opened Dec 26, 2024 by casper-hansen
2 tasks done
4
LLaVa 1.5 and 1.6 not working with text-only inputs bug
#35424 opened Dec 26, 2024 by giobin
2 of 4 tasks
Text Only input using LlaVa Next bug
#35421 opened Dec 26, 2024 by sinngam-khaidem
1 of 4 tasks
modular_model_converter can not handle objects import via try - except bug
#35414 opened Dec 25, 2024 by HIT-cwh
2 of 4 tasks
Qwen2VLProcessor cannot handle odd number of video frames bug
#35412 opened Dec 25, 2024 by DarkLight1337
2 of 4 tasks
1
VisionEncoderDecoderModel ONNX Conversion - Swinv2-Xlm-roberta-base bug
#35409 opened Dec 24, 2024 by Billybeast2003
2 of 4 tasks
Training issues latest version bug
#35407 opened Dec 24, 2024 by WoutDeRijck
2 of 4 tasks
1
AttributeError: 'SegformerFeatureExtractor' object has no attribute 'reduce_labels' still has no clear guide around bug
#35402 opened Dec 23, 2024 by deanAirre
2 of 4 tasks
1
Set output_attentions=True for model.geneate bug
#35393 opened Dec 23, 2024 by yiyexy
2 of 4 tasks
2
IdeficsImageProcessor raises unexpected ValueError bug
#35391 opened Dec 22, 2024 by Kamichanw
2 of 4 tasks
ModernBERT inference fails on CPU: ValueError: Pointer argument (at 0) cannot be accessed from Triton (cpu tensor?) bug
#35388 opened Dec 21, 2024 by umarbutler
4 tasks
3
Potentially incorrect calculation of total_updates on >=4.46.0 since #34198 affecting multi gpu training bug
#35387 opened Dec 21, 2024 by chiragjn
2 of 4 tasks
1
1
modernbert logits do not have gradient bug
#35386 opened Dec 21, 2024 by andersonbcdefg
3 of 4 tasks
1
Support modernBERT for encoder-decoder models Feature request Request for a new feature
#35385 opened Dec 21, 2024 by Bachstelze
1
MultiModalityCausalLM does not support Flash Attention 2.0 yet
#35383 opened Dec 21, 2024 by AlanPonnachan
RuntimeError: self and mat2 must have the same dtype, but got Float and BFloat16 when training with torch_compile bug
#35382 opened Dec 21, 2024 by umarbutler
2 of 4 tasks
1
is_causal arg appears twice in FAttention call from GPT2Attention.forward() bug
#35380 opened Dec 21, 2024 by poedator
2 of 4 tasks
1
ProTip! no:milestone will show everything without a milestone.