repo string | github_id int64 | github_node_id string | number int64 | html_url string | api_url string | title string | body string | state string | state_reason string | locked bool | comments_count int64 | labels list | assignees list | created_at string | updated_at string | closed_at string | author_association string | milestone_title string | snapshot_id string | extracted_at string | author_login string | author_id int64 | author_node_id string | author_type string | author_site_admin bool |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
huggingface/transformers | 3,518,780,612 | I_kwDOCUB6oc7RvFTE | 41,628 | https://github.com/huggingface/transformers/issues/41628 | https://api.github.com/repos/huggingface/transformers/issues/41628 | Cannot import name 'AutoImageProcessor' from 'transformers' | ### System Info
Intel CPU
Nvidia 3090
ubuntu 22.04
python 3.10.12
transformers=5.0.0.dev0 (installed from the official git repo)
### PS:
It's also tested with transformers=4.57.1, which is installed using "pip install", the same error persisted while executing "from transformers import AutoImageProcessor, AutoModel".... | closed | completed | false | 6 | [
"bug"
] | [] | 2025-10-15T16:29:20Z | 2026-02-26T18:36:13Z | 2025-10-16T12:37:07Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | Pittmann-XIE | 103,981,664 | U_kgDOBjKiYA | User | false |
huggingface/transformers | 3,522,707,706 | I_kwDOCUB6oc7R-ED6 | 41,669 | https://github.com/huggingface/transformers/issues/41669 | https://api.github.com/repos/huggingface/transformers/issues/41669 | Remove import * usage from models, adds 10 seconds and 38k files | ### System Info
any
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details below)
### Reproduction
The import * is... | closed | completed | false | 10 | [
"bug"
] | [] | 2025-10-16T16:56:39Z | 2026-04-07T04:33:22Z | 2025-11-24T08:02:53Z | CONTRIBUTOR | null | 20260407T090028Z | 2026-04-07T09:00:28Z | bschnurr | 1,946,977 | MDQ6VXNlcjE5NDY5Nzc= | User | false |
huggingface/transformers | 3,528,715,552 | I_kwDOCUB6oc7SU-0g | 41,720 | https://github.com/huggingface/transformers/issues/41720 | https://api.github.com/repos/huggingface/transformers/issues/41720 | Qwen3 with auto device mapping fails due to cudaErrorAssert on A800 | ### System Info
- `transformers` version: 4.57.1
- Platform: Linux-4.19.90-2107.6.0.0192.8.oe1.bclinux.x86_64-x86_64-with-glibc2.35
- Python version: 3.12.12
- Huggingface_hub version: 0.35.3
- Safetensors version: 0.6.2
- Accelerate version: 1.10.1
- Accelerate config: not found
- DeepSpeed version: not installed
... | closed | completed | false | 7 | [
"bug"
] | [] | 2025-10-18T11:50:43Z | 2026-03-12T05:43:23Z | 2026-01-05T08:03:26Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | guosyjlu | 69,756,483 | MDQ6VXNlcjY5NzU2NDgz | User | false |
huggingface/transformers | 3,532,707,392 | I_kwDOCUB6oc7SkNZA | 41,749 | https://github.com/huggingface/transformers/issues/41749 | https://api.github.com/repos/huggingface/transformers/issues/41749 | `_get_num_multimodal_tokens` is not implemented for model `mllama` | vLLM 0.11’s Transformers-backend expects the HF processor to implement a method called `_get_num_multimodal_tokens` which is [not implemented for mllama](https://github.com/huggingface/transformers/blob/main/src/transformers/models/mllama/processing_mllama.py) in `transformers 4.57.1`.
Because of this, `vllm serve met... | closed | completed | false | 4 | [
"bug"
] | [] | 2025-10-20T14:38:22Z | 2026-01-26T10:05:20Z | 2025-10-21T09:58:49Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | mrtpk | 8,076,245 | MDQ6VXNlcjgwNzYyNDU= | User | false |
huggingface/transformers | 3,535,832,788 | I_kwDOCUB6oc7SwIbU | 41,762 | https://github.com/huggingface/transformers/issues/41762 | https://api.github.com/repos/huggingface/transformers/issues/41762 | `IndexError: index 0 is out of bounds for dimension 0 with size 0` when loading Gemma3ForConditionalGeneration with DeepSpeed ZeRO-3 | ### System Info
transformers=4.57.1
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details below)
### Reproduction... | closed | completed | false | 8 | [
"bug"
] | [] | 2025-10-21T09:58:58Z | 2026-02-20T15:36:18Z | 2025-10-22T15:10:46Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | Asunatan | 105,210,894 | U_kgDOBkVkDg | User | false |
huggingface/transformers | 3,548,058,215 | I_kwDOCUB6oc7TexJn | 41,842 | https://github.com/huggingface/transformers/issues/41842 | https://api.github.com/repos/huggingface/transformers/issues/41842 | Incorrect usage of `num_items_in_batch`? | It seems that `num_items_in_batch` is computed for all items in the batch [here](https://github.com/huggingface/transformers/blob/9c20660138830ca362533551ca978c27b48283a1/src/transformers/trainer.py#L2430).
However, when loss is computed in the `training_step`, it is computed for each input in the batch one by one. Do... | closed | completed | false | 3 | [] | [] | 2025-10-24T07:36:00Z | 2026-03-09T14:02:44Z | 2025-12-01T08:02:48Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | gohar94 | 6,470,801 | MDQ6VXNlcjY0NzA4MDE= | User | false |
huggingface/transformers | 3,570,611,821 | I_kwDOCUB6oc7U0zZt | 41,950 | https://github.com/huggingface/transformers/issues/41950 | https://api.github.com/repos/huggingface/transformers/issues/41950 | video-classification pipeline looks for image processors | ### System Info
4.57.1
### Who can help?
@zucchini-nlp I can take a stab at this sometime
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details bel... | open | null | false | 6 | [
"WIP",
"bug"
] | [] | 2025-10-30T12:45:06Z | 2026-02-19T10:56:02Z | null | MEMBER | null | 20260325T173244Z | 2026-03-25T17:32:44Z | merveenoyan | 53,175,384 | MDQ6VXNlcjUzMTc1Mzg0 | User | false |
huggingface/transformers | 3,590,608,152 | I_kwDOCUB6oc7WBFUY | 42,032 | https://github.com/huggingface/transformers/issues/42032 | https://api.github.com/repos/huggingface/transformers/issues/42032 | ValueError: Unrecognized configuration class <class 'transformers.models.qwen3_omni_moe.configuration_qwen3_omni_moe.Qwen3OmniMoeConfig'> for this kind of AutoModel: AutoModel. | ### System Info
I have started testing the Qwen3-Omni model and at that time there was transformers version 4.56.0 available which had the issues to the model. With the commits and bugs fixation for transformers version 4.57.0 it got fixed but that commit was available on git. Since there is transformer update on the ... | closed | completed | false | 5 | [
"bug"
] | [] | 2025-11-05T11:39:39Z | 2026-02-11T23:54:10Z | 2025-12-27T08:03:07Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | Tortoise17 | 36,593,708 | MDQ6VXNlcjM2NTkzNzA4 | User | false |
huggingface/transformers | 3,604,732,641 | I_kwDOCUB6oc7W29rh | 42,111 | https://github.com/huggingface/transformers/issues/42111 | https://api.github.com/repos/huggingface/transformers/issues/42111 | Add thinking-budget support (max_thinking_tokens) for reasoning-capable chat models | ### Feature request
A built-in way to cap how many tokens a reasoning model spends inside its ``<think> … </think>`` block. Today, we can only control the total response length via ``max_new_tokens``. No parameter limits the internal reasoning segment when ``enable_thinking=True``.
### Motivation
- Reasoning models ... | open | null | false | 1 | [
"Feature request"
] | [] | 2025-11-09T10:09:11Z | 2026-02-14T05:37:15Z | null | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | AndresAlgaba | 35,764,158 | MDQ6VXNlcjM1NzY0MTU4 | User | false |
huggingface/transformers | 3,607,099,901 | I_kwDOCUB6oc7W__n9 | 42,116 | https://github.com/huggingface/transformers/issues/42116 | https://api.github.com/repos/huggingface/transformers/issues/42116 | Integration of the SINQ quantization strategy | ### Feature request
Adding support for **SINQ** quantization for Hugging Face compatible models, enabling users to apply it directly through the configuration settings. The **SINQ** quantization method, recently introduced in the paper [SINQ: Sinkhorn-Normalized Quantization for Calibration-Free Low-Precision LLM Weig... | closed | completed | false | 8 | [
"Feature request"
] | [] | 2025-11-10T09:44:32Z | 2026-02-16T15:08:43Z | 2026-02-16T15:08:43Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | ChiaraBoretti | 83,216,540 | MDQ6VXNlcjgzMjE2NTQw | User | false |
huggingface/transformers | 3,619,868,194 | I_kwDOCUB6oc7Xws4i | 42,175 | https://github.com/huggingface/transformers/issues/42175 | https://api.github.com/repos/huggingface/transformers/issues/42175 | Tensorflow not include in the backend when using pip install '.[torch]' | ### System Info
I install the program successfully when using `pip install -e .[torch]`.
However, I encounter the below issue when using ``pip install '.[torch]'``:
```
(omni) pqyin@proj54:/data2/pqyin/transformers$ python
Python 3.13.9 | packaged by Anaconda, Inc. | (main, Oct 21 2025, 19:16:10) [GCC 11.2.0] on linu... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-11-13T07:21:50Z | 2026-02-13T22:47:40Z | 2025-11-18T14:49:34Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | yinpeiqi | 60,515,999 | MDQ6VXNlcjYwNTE1OTk5 | User | false |
huggingface/transformers | 3,623,280,797 | I_kwDOCUB6oc7X9uCd | 42,199 | https://github.com/huggingface/transformers/issues/42199 | https://api.github.com/repos/huggingface/transformers/issues/42199 | Cardinality error is incorrect for models derived from DETR that do not have an explicit background class | ## Issue
For DETR variants, the cardinality errors that are reported during training are incorrect. This was reported in the DeformableDETR repository, and was acknowledged but not resolved:
https://github.com/fundamentalvision/Deformable-DETR/issues/24
Since all the derived models no longer include an explicit bac... | closed | completed | false | 10 | [
"bug"
] | [] | 2025-11-13T23:46:02Z | 2026-02-09T17:30:44Z | 2026-02-09T17:30:44Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | jveitchmichaelis | 3,159,591 | MDQ6VXNlcjMxNTk1OTE= | User | false |
huggingface/transformers | 3,623,324,953 | I_kwDOCUB6oc7X940Z | 42,200 | https://github.com/huggingface/transformers/issues/42200 | https://api.github.com/repos/huggingface/transformers/issues/42200 | Request of rewriting implementation of prediction_step in trainer.py | ### System Info
Any system. Because it's a problem coming from source code.
### Who can help?
@SunMarc
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [x] My own task or dataset (gi... | open | null | false | 4 | [
"Good Second Issue",
"bug"
] | [] | 2025-11-14T00:13:40Z | 2026-02-24T22:09:56Z | null | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | Yacklin | 139,425,274 | U_kgDOCE91-g | User | false |
huggingface/transformers | 3,624,126,333 | I_kwDOCUB6oc7YA8d9 | 42,202 | https://github.com/huggingface/transformers/issues/42202 | https://api.github.com/repos/huggingface/transformers/issues/42202 | Deformable DETR Finetuning breaks for any dataset | ### System Info
- GPU: V100
- torch2.6.0+cu126
- transformers 4.57.1
### Who can help?
Hi @yonigozlan @molbap @NielsRogge
Thanks for the awesome work on vision models!
I've been trying to finetune the Deformable DETR models (SenseTime/deformable-detr-with-box-refine-two-stage) for the past few days on a custom ... | closed | completed | false | 6 | [
"bug"
] | [] | 2025-11-14T06:29:52Z | 2026-02-08T16:56:36Z | 2026-02-08T16:56:36Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | iamsashank09 | 26,921,144 | MDQ6VXNlcjI2OTIxMTQ0 | User | false |
huggingface/transformers | 3,628,809,478 | I_kwDOCUB6oc7YSz0G | 42,222 | https://github.com/huggingface/transformers/issues/42222 | https://api.github.com/repos/huggingface/transformers/issues/42222 | All vitpose model were brokentransformers/models/vitpose_ | ### System Info
transformers/models/vitpose_backbone/modeling_vitpose_backbone.py", line 304, in forward
raise ValueError(transformers/models/vitpose_backbone/modeling_vitpose_backbone.py", line 304, in forward
raise ValueError(
ValueError: dataset_index must be provided when using multiple experts (num_expert... | closed | completed | false | 11 | [
"bug"
] | [] | 2025-11-15T14:56:04Z | 2026-02-09T08:11:37Z | 2026-02-09T08:11:37Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | lucasjinreal | 21,303,438 | MDQ6VXNlcjIxMzAzNDM4 | User | false |
huggingface/transformers | 3,634,466,348 | I_kwDOCUB6oc7YoY4s | 42,249 | https://github.com/huggingface/transformers/issues/42249 | https://api.github.com/repos/huggingface/transformers/issues/42249 | `parse_response` should drop EOS | When using `parse_response`, I noticed it includes the EOS token in the `content`. However, the EOS token should be excluded, as it adds an unwanted EOS before tool calls during subsequent formatting.
```python
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-0.6B")
# Why ... | closed | completed | false | 7 | [] | [] | 2025-11-17T18:14:56Z | 2026-02-15T08:04:47Z | 2026-02-15T08:04:47Z | MEMBER | null | 20260325T173244Z | 2026-03-25T17:32:44Z | qgallouedec | 45,557,362 | MDQ6VXNlcjQ1NTU3MzYy | User | false |
huggingface/transformers | 3,660,616,450 | I_kwDOCUB6oc7aMJMC | 42,371 | https://github.com/huggingface/transformers/issues/42371 | https://api.github.com/repos/huggingface/transformers/issues/42371 | Please use the new API settings to control TF32 behavior, ... | ### System Info
> UserWarning: Please use the new API settings to control TF32 behavior, such as torch.backends.cudnn.conv.fp32_precision = 'tf32' or torch.backends.cuda.matmul.fp32_precision = 'ieee'. Old settings, e.g, torch.backends.cuda.matmul.allow_tf32 = True, torch.backends.cudnn.allow_tf32 = True, allowTF32CuD... | closed | completed | false | 18 | [
"Good First Issue",
"bug"
] | [] | 2025-11-24T21:38:12Z | 2026-02-05T06:12:27Z | 2025-12-01T09:58:32Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | wasertech | 79,070,834 | MDQ6VXNlcjc5MDcwODM0 | User | false |
huggingface/transformers | 3,661,658,315 | I_kwDOCUB6oc7aQHjL | 42,375 | https://github.com/huggingface/transformers/issues/42375 | https://api.github.com/repos/huggingface/transformers/issues/42375 | SAM3 single image inference with multiple text prompt | Hi
I'm trying to run inference on a single image, aiming to get the bbox of objects from several different categories (e.g. "a person" and "a car").
the only example i found for prompting with multiple categories is in the "Batched Inference with Text Prompts" example, but then i need to unnecessarily duplicate my imag... | closed | completed | false | 10 | [] | [] | 2025-11-25T06:20:09Z | 2026-02-08T08:04:26Z | 2026-02-08T08:04:26Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | iariav | 28,890,865 | MDQ6VXNlcjI4ODkwODY1 | User | false |
huggingface/transformers | 3,663,973,751 | I_kwDOCUB6oc7aY813 | 42,405 | https://github.com/huggingface/transformers/issues/42405 | https://api.github.com/repos/huggingface/transformers/issues/42405 | Integrate FA4 (Flash Attention for Blackwell) into HF Transformers | ### Feature request
Transformers currently supports Flash Attention 2 and 3, but not Flash Attention 4. Users with compatible hardware and the latest flash-attn package cannot leverage FA4's improvements . Lets add that as well .
### Motivation
Flash Attention 4 is now available in the flash-attn package, bringing ... | closed | completed | false | 0 | [
"Feature request"
] | [] | 2025-11-25T17:33:50Z | 2026-03-13T19:32:03Z | 2026-03-13T19:32:03Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | sambhavnoobcoder | 94,298,612 | U_kgDOBZ7h9A | User | false |
huggingface/transformers | 3,676,316,784 | I_kwDOCUB6oc7bICRw | 42,491 | https://github.com/huggingface/transformers/issues/42491 | https://api.github.com/repos/huggingface/transformers/issues/42491 | The LoRA model trained with qwen3_moe on hf4.x cannot be used on the current main branch (hf5.x). | ### System Info
- `transformers` version: 5.0.0.dev0
- Platform: Linux-5.15.0-122-generic-x86_64-with-glibc2.31
- Python version: 3.10.15
- Huggingface_hub version: 1.1.5
- Safetensors version: 0.4.5
- Accelerate version: 1.9.0
- Accelerate config: not found
- DeepSpeed version: 0.16.2
- PyTorch version (accelerato... | closed | completed | false | 6 | [
"bug",
"PEFT"
] | [] | 2025-11-29T03:57:15Z | 2026-01-24T10:06:54Z | 2026-01-24T10:06:54Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | linitra24 | 116,639,249 | U_kgDOBvPGEQ | User | false |
huggingface/transformers | 3,679,872,670 | I_kwDOCUB6oc7bVmae | 42,503 | https://github.com/huggingface/transformers/issues/42503 | https://api.github.com/repos/huggingface/transformers/issues/42503 | Add ModernVBERT models | ### Model description
Add the models from [ModernVBERT: Towards Smaller Visual Document Retrievers](https://arxiv.org/abs/2510.01149).
- `ModernVBertModel`
- `ColModernVBertForRetrieval`
### Open source status
- [ ] The model implementation is available
- [ ] The model weights are available
### Provide useful link... | closed | completed | false | 0 | [
"New model"
] | [] | 2025-12-01T08:33:46Z | 2026-02-23T12:55:43Z | 2026-02-23T12:55:43Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | paultltc | 73,120,933 | MDQ6VXNlcjczMTIwOTMz | User | false |
huggingface/transformers | 3,684,864,768 | I_kwDOCUB6oc7bopMA | 42,548 | https://github.com/huggingface/transformers/issues/42548 | https://api.github.com/repos/huggingface/transformers/issues/42548 | cannot import name 'PreTrainedModel' from 'transformers' | Installed `transformers-4.57.0` but encountered an error. However, it contains PreTrainedModel, yet the error still persists:
Traceback (most recent call last):
File "/checkpoint/binary/train_package/playground/benchmarks/discrete_vla_pretrain.py", line 39, in <module>
from dexbotic.model.discrete_vla.discrete_vla_arc... | closed | completed | false | 14 | [] | [] | 2025-12-02T09:10:32Z | 2026-03-08T08:03:46Z | 2026-03-08T08:03:46Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | missTL | 98,095,885 | U_kgDOBdjTDQ | User | false |
huggingface/transformers | 3,685,706,215 | I_kwDOCUB6oc7br2nn | 42,556 | https://github.com/huggingface/transformers/issues/42556 | https://api.github.com/repos/huggingface/transformers/issues/42556 | [v5] Remove `safe_serialization` parameter | ### Feature request
Mentioned in https://github.com/huggingface/transformers/pull/42391.
Original comment from @Wauplin :
> Do we still want to allow saving with `safe_serialization=False`? The v5 release feels a very good opportunity to definitely get rid of non-safetensors serialization. WDYT ?
At this point, un... | closed | completed | false | 5 | [
"Feature request",
"for_v5?"
] | [] | 2025-12-02T12:54:40Z | 2026-02-12T08:18:25Z | 2025-12-16T14:16:45Z | MEMBER | null | 20260325T173244Z | 2026-03-25T17:32:44Z | Wauplin | 11,801,849 | MDQ6VXNlcjExODAxODQ5 | User | false |
huggingface/transformers | 3,693,864,351 | I_kwDOCUB6oc7cK-Wf | 42,617 | https://github.com/huggingface/transformers/issues/42617 | https://api.github.com/repos/huggingface/transformers/issues/42617 | Not able to run 3d_parallel.py | ### System Info
- `transformers` version: 4.57.1
- Platform: Linux-5.4.0-218-generic-x86_64-with-glibc2.31
- Python version: 3.10.18
- Huggingface_hub version: 0.36.0
- Safetensors version: 0.6.2
- Accelerate version: 1.11.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accele... | closed | completed | false | 3 | [
"bug"
] | [] | 2025-12-04T10:14:17Z | 2026-02-06T08:08:59Z | 2026-02-06T08:08:59Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | quic-meetkuma | 200,747,495 | U_kgDOC_cp5w | User | false |
huggingface/transformers | 3,697,005,235 | I_kwDOCUB6oc7cW9Kz | 42,638 | https://github.com/huggingface/transformers/issues/42638 | https://api.github.com/repos/huggingface/transformers/issues/42638 | Routing Replay for MoEs | ### Feature request
RecentRL approaches for training MoE models increasingly rely on **Routing Replay**, as described in the following papers:
- https://huggingface.co/papers/2507.18071
- https://huggingface.co/papers/2510.11370
- https://huggingface.co/papers/2512.01374
Without going into the training details, Rout... | closed | completed | false | 2 | [
"Feature request"
] | [] | 2025-12-04T23:58:14Z | 2026-04-14T08:09:16Z | 2026-04-14T08:09:16Z | MEMBER | null | 20260414T122001Z | 2026-04-14T12:20:01Z | qgallouedec | 45,557,362 | MDQ6VXNlcjQ1NTU3MzYy | User | false |
huggingface/transformers | 3,701,324,455 | I_kwDOCUB6oc7cnbqn | 42,673 | https://github.com/huggingface/transformers/issues/42673 | https://api.github.com/repos/huggingface/transformers/issues/42673 | Qwen3ForCausalLM leaks VRAM if used in multiple dataloader threads | ### System Info
torch 2.8.0
transformers==4.56.2 or ransformers==4.57.3, both tested
### Who can help?
@ArthurZucker @Cyrilvallez
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [x] ... | closed | completed | false | 10 | [
"bug"
] | [] | 2025-12-06T09:01:40Z | 2026-03-02T11:19:28Z | 2026-03-02T11:19:28Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | dxqb | 183,307,934 | U_kgDOCu0Ong | User | false |
huggingface/transformers | 3,707,232,318 | I_kwDOCUB6oc7c9-A- | 42,710 | https://github.com/huggingface/transformers/issues/42710 | https://api.github.com/repos/huggingface/transformers/issues/42710 | Outstanding issues / PR before we can release v5 | - #42697 actual perf upgrade
- #34919 default dtype
- #42513
- #42563 final form and break for tokenization
- #42558 gradient checkpointing refactor
- #42555 shard size default
- #42491 PEFT + MOE refactor
- #41388 default to fast image processors
- #42418
- #42894
- #42564 | closed | completed | false | 4 | [] | [] | 2025-12-08T16:57:54Z | 2026-01-26T08:29:54Z | 2026-01-26T08:29:54Z | MEMBER | null | 20260325T173244Z | 2026-03-25T17:32:44Z | ArthurZucker | 48,595,927 | MDQ6VXNlcjQ4NTk1OTI3 | User | false |
huggingface/transformers | 3,710,462,747 | I_kwDOCUB6oc7dKSsb | 42,738 | https://github.com/huggingface/transformers/issues/42738 | https://api.github.com/repos/huggingface/transformers/issues/42738 | BERT-like models with RoPE | ### Model description
Some BERT-like models:
- [ ] NomicBert
- [ ] GTE
- [ ] Snowflake GTE
- [ ] Jinaai embeddings
Are well used custom models which implement BERT with RoPE.
This is a tracking issue to add support for them in Transformers.
---
One motivation for this is so that the Transformers modeling backend ... | open | reopened | false | 3 | [
"New model"
] | [] | 2025-12-09T11:30:40Z | 2026-04-02T14:35:17Z | null | MEMBER | null | 20260407T090028Z | 2026-04-07T09:00:28Z | hmellor | 19,981,378 | MDQ6VXNlcjE5OTgxMzc4 | User | false |
huggingface/transformers | 3,711,097,149 | I_kwDOCUB6oc7dMtk9 | 42,740 | https://github.com/huggingface/transformers/issues/42740 | https://api.github.com/repos/huggingface/transformers/issues/42740 | how to train trocr with transformers 4.57+? | i train trocr with tranfomers 4.15, the results is right,but train with 4.57.1,the acc is always 0 , i did't find the reason,did t can train succ with latest transofrmers? | closed | completed | false | 3 | [] | [] | 2025-12-09T14:07:50Z | 2026-02-18T08:10:11Z | 2026-02-18T08:10:11Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | cqray1990 | 32,585,434 | MDQ6VXNlcjMyNTg1NDM0 | User | false |
huggingface/transformers | 3,713,115,899 | I_kwDOCUB6oc7dUab7 | 42,754 | https://github.com/huggingface/transformers/issues/42754 | https://api.github.com/repos/huggingface/transformers/issues/42754 | Excluding weight decay not working properly on most LMs | The problem is consistent across all three files:
https://github.com/huggingface/transformers/blob/471d7ce9abbb3bc1b3bab673367378f9dbc3caac/examples/pytorch/language-modeling/run_clm_no_trainer.py#L518-L528
https://github.com/huggingface/transformers/blob/471d7ce9abbb3bc1b3bab673367378f9dbc3caac/examples/pytorch/langua... | closed | completed | false | 5 | [] | [] | 2025-12-10T00:36:55Z | 2026-02-13T10:46:14Z | 2026-01-18T08:01:57Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | wwt17 | 10,792,281 | MDQ6VXNlcjEwNzkyMjgx | User | false |
huggingface/transformers | 3,722,388,341 | I_kwDOCUB6oc7d3yN1 | 42,831 | https://github.com/huggingface/transformers/issues/42831 | https://api.github.com/repos/huggingface/transformers/issues/42831 | Accuracy issue associated with FineGrainedFP8 | ### System Info
Hello,
I am writing to report an issue I observed while evaluating the accuracy of a model quantized with FineGrainedFP8 through lm-eval. I observed significant accuracy discrepancies when deploying the quantized model with the HF backend versus the vLLM backend.
<img width="1408" height="1250" alt="... | closed | completed | false | 6 | [
"bug"
] | [] | 2025-12-12T07:47:42Z | 2026-02-08T08:04:07Z | 2026-02-08T08:04:07Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | sunghyuckhong | 131,639,753 | U_kgDOB9ipyQ | User | false |
huggingface/transformers | 3,722,400,306 | I_kwDOCUB6oc7d31Iy | 42,832 | https://github.com/huggingface/transformers/issues/42832 | https://api.github.com/repos/huggingface/transformers/issues/42832 | Question about tie_weights | Hi,
I noticed that the logic of the tie_weights function has changed in the transformers 5.0.0rc.
In v4.x, when tie_word_embeddings=True, weights between embed_tokens.weight and lm_head.weight were always tied, regardless of whether both tensors were present in the checkpoint.
However, in v5.0.0rc, if both embed_tok... | closed | completed | false | 12 | [] | [] | 2025-12-12T07:52:43Z | 2026-04-01T08:24:28Z | 2026-04-01T08:24:28Z | NONE | null | 20260407T090028Z | 2026-04-07T09:00:28Z | cjw-d | 73,046,570 | MDQ6VXNlcjczMDQ2NTcw | User | false |
huggingface/transformers | 3,732,475,843 | I_kwDOCUB6oc7eeQ_D | 42,886 | https://github.com/huggingface/transformers/issues/42886 | https://api.github.com/repos/huggingface/transformers/issues/42886 | Tokenizer fails to load from cache when HF_HUB_OFFLINE=1 on 4.57.3 | ### System Info
transformers==4.57.3
### Who can help?
@vasqu @ArthurZucker
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details below)
### Repr... | closed | completed | false | 7 | [
"bug"
] | [] | 2025-12-15T23:17:06Z | 2026-02-09T11:12:23Z | 2026-02-09T11:12:23Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | chtruong814 | 195,193,376 | U_kgDOC6JqIA | User | false |
huggingface/transformers | 3,733,715,750 | I_kwDOCUB6oc7ei_sm | 42,890 | https://github.com/huggingface/transformers/issues/42890 | https://api.github.com/repos/huggingface/transformers/issues/42890 | tests/models/sam_hq/test_modeling_sam_hq.py::SamHQModelIntegrationTest may fail since a lot of cases are lack of set_seed() | ### System Info
transformers built from latest main.
### Who can help?
@ydshieh
similar issue may occur in tests/models/sam/test_modeling_sam.py
see https://github.com/sywangyi/transformers/blob/main/src/transformers/models/sam_hq/modeling_sam_hq.py#L1077
randn() is used in positional_embedding, so you need to ... | closed | completed | false | 5 | [
"bug"
] | [] | 2025-12-16T08:22:52Z | 2026-01-26T16:22:26Z | 2026-01-26T16:22:26Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | sywangyi | 36,058,628 | MDQ6VXNlcjM2MDU4NjI4 | User | false |
huggingface/transformers | 3,734,408,824 | I_kwDOCUB6oc7elo54 | 42,898 | https://github.com/huggingface/transformers/issues/42898 | https://api.github.com/repos/huggingface/transformers/issues/42898 | `clean_up_tokenization_spaces` behavior changes in v5 | ### System Info
- `transformers` version: 4.57.3/5.0.0rc1
- Platform: Linux-5.15.0-56-generic-x86_64-with-glibc2.35
- Python version: 3.10.12
### Who can help?
@ArthurZucker and @itazap
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported task i... | closed | completed | false | 3 | [
"bug"
] | [] | 2025-12-16T11:34:46Z | 2026-02-11T10:01:39Z | 2026-02-11T10:01:39Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | apaniukov | 51,917,466 | MDQ6VXNlcjUxOTE3NDY2 | User | false |
huggingface/transformers | 3,734,648,644 | I_kwDOCUB6oc7emjdE | 42,902 | https://github.com/huggingface/transformers/issues/42902 | https://api.github.com/repos/huggingface/transformers/issues/42902 | Improve GPT OSS Conversion Script | ### Feature request
The GPT OSS Conversion Script exposes parameters that are not needed or used, has incorrect documentation, and crashes due to a tiktoken bug.
I improved the script and validated it works with GPT-OSS-20B.
### Motivation
The original motivation is from https://github.com/huggingface/accelerate/is... | closed | completed | false | 0 | [
"Feature request"
] | [] | 2025-12-16T12:46:20Z | 2026-02-02T09:25:02Z | 2026-02-02T09:25:02Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | KyleMylonakisProtopia | 122,286,752 | U_kgDOB0nyoA | User | false |
huggingface/transformers | 3,735,475,105 | I_kwDOCUB6oc7eptOh | 42,907 | https://github.com/huggingface/transformers/issues/42907 | https://api.github.com/repos/huggingface/transformers/issues/42907 | Failing to Save Dequantized Ministrals/Devstrals | ### System Info
When running in a clean google colab environment with transformers installed from source `!pip install -U git+https://github.com/huggingface/transformers.git` with `transformers===5.0.0rc1`.
It can be reproduced with the free T4 GPU.
### Who can help?
@ArthurZucker @Cyrilvallez
### Information
- ... | closed | completed | false | 10 | [
"bug"
] | [] | 2025-12-16T16:21:17Z | 2026-03-08T08:03:34Z | 2026-03-08T08:03:34Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | pandora-s-git | 128,635,000 | U_kgDOB6rQeA | User | false |
huggingface/transformers | 3,736,518,828 | I_kwDOCUB6oc7etsCs | 42,913 | https://github.com/huggingface/transformers/issues/42913 | https://api.github.com/repos/huggingface/transformers/issues/42913 | Unexpected tokenizer behavior difference from v4 to v5 | ```python
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained('mlx-community/Llama-3.2-1B-Instruct-4bit')
text = tokenizer.decode([128000, 64, 1174, 65])
print(text)
```
For 4.57.3 you get `<|begin_of_text|>a,b`
For 5.0.0rc1 you get `<|begin_of_text|>a ,b`
Is it expected the behavior ch... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-12-16T22:14:01Z | 2026-01-25T08:02:25Z | 2026-01-25T08:02:25Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | awni | 1,542,805 | MDQ6VXNlcjE1NDI4MDU= | User | false |
huggingface/transformers | 3,737,719,708 | I_kwDOCUB6oc7eyROc | 42,915 | https://github.com/huggingface/transformers/issues/42915 | https://api.github.com/repos/huggingface/transformers/issues/42915 | Qwen3Moe failed with FineGrainedFP8Config | ### System Info
- `transformers` version: 4.57.3
- Platform: Linux-5.15.0-136-generic-x86_64-with-glibc2.35
- Python version: 3.12.10
- Huggingface_hub version: 0.36.0
- Safetensors version: 0.5.3
- Accelerate version: 1.12.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accel... | closed | completed | false | 4 | [
"bug"
] | [] | 2025-12-17T07:35:47Z | 2026-01-25T08:02:24Z | 2026-01-25T08:02:24Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | jessiewiswjc | 70,051,089 | MDQ6VXNlcjcwMDUxMDg5 | User | false |
huggingface/transformers | 3,740,752,067 | I_kwDOCUB6oc7e91jD | 42,936 | https://github.com/huggingface/transformers/issues/42936 | https://api.github.com/repos/huggingface/transformers/issues/42936 | Mask2former model's ignore_value not used after definition | https://github.com/huggingface/transformers/blob/47b0e478f324b54f177ea7998a0791870fdd0324/src/transformers/models/mask2former/configuration_mask2former.py#L83
`ignore_value` does not seem to be used any where despite its definition above | closed | completed | false | 6 | [] | [] | 2025-12-17T22:44:36Z | 2026-03-11T08:08:16Z | 2026-03-11T08:08:16Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | mhmd-j | 56,149,395 | MDQ6VXNlcjU2MTQ5Mzk1 | User | false |
huggingface/transformers | 3,742,730,049 | I_kwDOCUB6oc7fFYdB | 42,943 | https://github.com/huggingface/transformers/issues/42943 | https://api.github.com/repos/huggingface/transformers/issues/42943 | Continuous batching: output queue requeue starvation and request-scoped iterator does not terminate on completion |
### **Description**
There are two related correctness issues in the continuous batching result consumption logic that can lead to unfairness and non-terminating iterators under concurrent workloads.
#### 1. Starvation and incorrect timeout handling in `get_result`
`ContinuousBatchingManager.get_result` currently r... | closed | completed | false | 1 | [] | [
"remi-or"
] | 2025-12-18T11:35:58Z | 2026-02-08T08:04:03Z | 2026-02-08T08:04:03Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | pythongiant | 13,624,560 | MDQ6VXNlcjEzNjI0NTYw | User | false |
huggingface/transformers | 3,744,470,098 | I_kwDOCUB6oc7fMBRS | 42,947 | https://github.com/huggingface/transformers/issues/42947 | https://api.github.com/repos/huggingface/transformers/issues/42947 | Gradient Checkpointing Ineffective with PEFT LoRA Despite Proper Configuration | ### System Info
- `transformers` version: 4.57.1
- Platform: Linux-5.15.0-161-generic-x86_64-with-glibc2.35
- Python version: 3.12.12
- Huggingface_hub version: 0.36.0
- Safetensors version: 0.7.0
- Accelerate version: 1.11.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (acceler... | closed | completed | false | 14 | [
"bug"
] | [] | 2025-12-18T19:10:03Z | 2026-02-18T08:10:01Z | 2026-02-18T08:10:01Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | yurkoff-mv | 82,467,993 | MDQ6VXNlcjgyNDY3OTkz | User | false |
huggingface/transformers | 3,750,144,911 | I_kwDOCUB6oc7fhquP | 42,971 | https://github.com/huggingface/transformers/issues/42971 | https://api.github.com/repos/huggingface/transformers/issues/42971 | Please create a Huggingface Transformers SKILL for Claude | ### Feature request
Please create a Huggingface Transformers SKILL for Claude and a PLUGIN for Claude Code.
### Motivation
It is very hard to navigate all the features of the Transformers library. Letting Claude guide us would make things much faster.
### Your contribution
I'll help testing the SKILL on my mac M4. | closed | completed | false | 17 | [
"Good First Issue",
"Feature request"
] | [] | 2025-12-20T15:17:03Z | 2026-02-03T14:20:04Z | 2026-02-03T14:20:04Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | Emasoft | 713,559 | MDQ6VXNlcjcxMzU1OQ== | User | false |
huggingface/transformers | 3,750,762,299 | I_kwDOCUB6oc7fkBc7 | 42,977 | https://github.com/huggingface/transformers/issues/42977 | https://api.github.com/repos/huggingface/transformers/issues/42977 | Add ViT NEPA | ### Model description
Summary : Next-Embedding Prediction: The Simple Secret to Strong Vision Learners
NEPA is a self-supervised method. It trains Vision Transformers to predict future patch embeddings. No complex loss functions or extra heads. Achieves 85.3% top-1 accuracy on ImageNet-1K with ViT-L.
The NEPA model ... | open | null | false | 1 | [
"New model"
] | [] | 2025-12-21T05:33:08Z | 2026-02-07T21:42:17Z | null | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | sbucaille | 24,275,548 | MDQ6VXNlcjI0Mjc1NTQ4 | User | false |
huggingface/transformers | 3,751,484,378 | I_kwDOCUB6oc7fmxva | 42,981 | https://github.com/huggingface/transformers/issues/42981 | https://api.github.com/repos/huggingface/transformers/issues/42981 | Change from v4 to v5c1: DeprecationWarning: builtin type SwigPyPacked/Object has no __module__ attribute | When I run a purest, the tests work as expected, after updating transformers to 5rc1 but I get a deprecation warning that is new. This seems harmless, for now, but may mean something to someone!
```
<frozen importlib._bootstrap>:488
<frozen importlib._bootstrap>:488: DeprecationWarning: builtin type SwigPyPacked h... | closed | completed | false | 2 | [] | [] | 2025-12-21T19:04:58Z | 2026-01-29T08:06:26Z | 2026-01-29T08:06:26Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | jrp2014 | 8,142,876 | MDQ6VXNlcjgxNDI4NzY= | User | false |
huggingface/transformers | 3,752,659,910 | I_kwDOCUB6oc7frQvG | 42,994 | https://github.com/huggingface/transformers/issues/42994 | https://api.github.com/repos/huggingface/transformers/issues/42994 | quantized model saving failed | ### System Info
regression PR #42734
### Who can help?
@Cyrilvallez
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details below)
### Reproductio... | closed | completed | false | 8 | [
"bug"
] | [] | 2025-12-22T07:35:08Z | 2026-03-02T08:09:13Z | 2026-03-02T08:09:13Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | jiqing-feng | 107,918,818 | U_kgDOBm614g | User | false |
huggingface/transformers | 3,755,864,802 | I_kwDOCUB6oc7f3fLi | 43,010 | https://github.com/huggingface/transformers/issues/43010 | https://api.github.com/repos/huggingface/transformers/issues/43010 | Cache's (and Layer's) `update(...)` method to be decorated with `@torch.no_grad` | ### System Info
5.0.0rc1, 2.9.1
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details below)
### Reproduction
It ... | closed | completed | false | 7 | [
"bug"
] | [] | 2025-12-23T03:08:33Z | 2026-03-16T08:18:54Z | 2026-03-16T08:18:54Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | vadimkantorov | 1,041,752 | MDQ6VXNlcjEwNDE3NTI= | User | false |
huggingface/transformers | 3,755,872,144 | I_kwDOCUB6oc7f3g-Q | 43,011 | https://github.com/huggingface/transformers/issues/43011 | https://api.github.com/repos/huggingface/transformers/issues/43011 | `StaticLayer` cache layer to implement `.crop(seq_len)` to match API of `DynamicLayer` | ### Feature request
5.0.0rc1, 2.91
### Motivation
It then is possible to crop the cache to a prefix sequence and reuse the prefix cache
### Your contribution
I've implemented it as follows:
```python
@torch.no_grad
def crop(self, seq_len):
self.keys[0, 0, seq_len:] = 0
```
btw it feels quite fragile for `get_s... | open | null | false | 8 | [
"Feature request"
] | [] | 2025-12-23T03:13:30Z | 2026-02-17T09:17:37Z | null | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | vadimkantorov | 1,041,752 | MDQ6VXNlcjEwNDE3NTI= | User | false |
huggingface/transformers | 3,755,941,571 | I_kwDOCUB6oc7f3x7D | 43,012 | https://github.com/huggingface/transformers/issues/43012 | https://api.github.com/repos/huggingface/transformers/issues/43012 | Compiling a bfloat16 model triggers float32 precision PyTorch warning | ### System Info
5.0.0rc1, 2.9.1
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details below)
### Reproduction
```... | closed | completed | false | 5 | [
"bug"
] | [] | 2025-12-23T03:59:45Z | 2026-03-05T08:07:35Z | 2026-03-05T08:07:35Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | vadimkantorov | 1,041,752 | MDQ6VXNlcjEwNDE3NTI= | User | false |
huggingface/transformers | 3,757,619,336 | I_kwDOCUB6oc7f-LiI | 43,023 | https://github.com/huggingface/transformers/issues/43023 | https://api.github.com/repos/huggingface/transformers/issues/43023 | How to investigate "CAS service error" during model downloading? | ### System Info
(nm) PS C:\Users\myuser\AppData\Local\anaconda3\envs\nm\Lib\site-packages\transformers\commands> python .\transformers_cli.py env
```
Copy-and-paste the text below in your GitHub issue and FILL OUT the two last points.
- `transformers` version: 4.57.3
- Platform: Windows-10-10.0.19045-SP0
- Python v... | closed | completed | false | 3 | [
"bug"
] | [] | 2025-12-23T14:48:51Z | 2026-01-31T08:02:33Z | 2026-01-31T08:02:33Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | satyrmipt | 113,777,913 | U_kgDOBsgc-Q | User | false |
huggingface/transformers | 3,760,414,050 | I_kwDOCUB6oc7gI11i | 43,032 | https://github.com/huggingface/transformers/issues/43032 | https://api.github.com/repos/huggingface/transformers/issues/43032 | [v5 Regression] BitsAndBytes 4-bit quantization OOM - core_model_loading bypasses quantizer device placement | I probably missed a new setting somewhere, because the load is very slow and it was fast before, but just in case i did not, here is a possible issue. just close it if i missed something.
**System Info**
transformers: 5.0.0.dev0 (installed from main)
bitsandbytes: 0.49.0
torch: 2.9.1+cu128
Python: 3.12
GPU: NVIDIA G... | closed | completed | false | 7 | [] | [] | 2025-12-24T14:11:26Z | 2026-03-18T15:42:39Z | 2026-03-18T15:42:39Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | jwm1969 | 35,038,618 | MDQ6VXNlcjM1MDM4NjE4 | User | false |
huggingface/transformers | 3,761,496,273 | I_kwDOCUB6oc7gM-DR | 43,037 | https://github.com/huggingface/transformers/issues/43037 | https://api.github.com/repos/huggingface/transformers/issues/43037 | DeepSeek v3.2 support | ### Feature request
When will Transformers officially support DeepSeek v3.2?
https://huggingface.co/deepseek-ai/DeepSeek-V3.2
### Motivation
None
### Your contribution
None | open | null | false | 6 | [
"Feature request"
] | [] | 2025-12-25T06:38:29Z | 2026-04-13T09:20:25Z | null | NONE | null | 20260413T094825Z | 2026-04-13T09:48:25Z | freedom-cui | 81,297,486 | MDQ6VXNlcjgxMjk3NDg2 | User | false |
huggingface/transformers | 3,761,801,871 | I_kwDOCUB6oc7gOIqP | 43,039 | https://github.com/huggingface/transformers/issues/43039 | https://api.github.com/repos/huggingface/transformers/issues/43039 | When using the Liger Kernel, torch.nn.functional.cross_entropy is called | ### System Info
```
accelerate 1.11.0
liger-kernel 0.0.3
numpy 2.3.3
peft 0.18.0
tokenizers 0.22.1
torch 2.9.0+cu126
torch-tb-profiler 0.4.3
tor... | open | null | false | 7 | [
"Good Second Issue",
"bug"
] | [] | 2025-12-25T10:55:29Z | 2026-03-25T03:50:08Z | null | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | yurkoff-mv | 82,467,993 | MDQ6VXNlcjgyNDY3OTkz | User | false |
huggingface/transformers | 3,763,183,796 | I_kwDOCUB6oc7gTaC0 | 43,048 | https://github.com/huggingface/transformers/issues/43048 | https://api.github.com/repos/huggingface/transformers/issues/43048 | Need to understand difference between TP support via transformers code v/s Pytorch's native parallelize_module API. | Based on the existing code base of transformers, below sequence of operations are performed on model object to make it TP compatible.
- TP Plan for Llama: https://github.com/huggingface/transformers/blob/a7f29523361b2cc12e51c1f5133d95f122f6f45c/src/transformers/models/llama/configuration_llama.py#L113
- self._tp_plan ... | closed | completed | false | 3 | [] | [] | 2025-12-26T10:05:38Z | 2026-03-08T08:03:22Z | 2026-03-08T08:03:22Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | quic-meetkuma | 200,747,495 | U_kgDOC_cp5w | User | false |
huggingface/transformers | 3,764,583,354 | I_kwDOCUB6oc7gYvu6 | 43,054 | https://github.com/huggingface/transformers/issues/43054 | https://api.github.com/repos/huggingface/transformers/issues/43054 | text embedding of siglip2 is much worse than siglip | ### System Info
- `transformers` version: 4.57.3
- Platform: Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.39
- Python version: 3.12.12
- Huggingface_hub version: 0.36.0
- Safetensors version: 0.7.0
- Accelerate version: 1.12.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorc... | closed | completed | false | 8 | [
"bug"
] | [] | 2025-12-27T08:46:30Z | 2026-01-28T14:52:48Z | 2026-01-28T14:52:48Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | fancyerii | 5,372,812 | MDQ6VXNlcjUzNzI4MTI= | User | false |
huggingface/transformers | 3,767,588,609 | I_kwDOCUB6oc7gkNcB | 43,064 | https://github.com/huggingface/transformers/issues/43064 | https://api.github.com/repos/huggingface/transformers/issues/43064 | Trainer.train() using v5 + FSDP2 + PEFT + cpu_ram_efficient_loading=True results in wrong optimizer states/params on all but 0th rank | ### System Info
v5 + FSDP2 + cpu_ram_efficient_loading=True + PEFT + "reshard_after_forward": True
### Who can help?
@S1ro1
@SunMarc
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
-... | closed | completed | false | 3 | [
"bug"
] | [] | 2025-12-29T14:44:46Z | 2026-02-06T08:08:32Z | 2026-02-06T08:08:32Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | d5031 | 46,198,216 | MDQ6VXNlcjQ2MTk4MjE2 | User | false |
huggingface/transformers | 3,767,686,605 | I_kwDOCUB6oc7gklXN | 43,065 | https://github.com/huggingface/transformers/issues/43065 | https://api.github.com/repos/huggingface/transformers/issues/43065 | Dummy `nn.Conv2d` in `Sam3PixelDecoder` | ### System Info
transformers==5.0.0rc1
### Who can help?
I’ve noticed that `Sam3Model.from_pretrained("facebook/sam3")` sets `num_upsampling_stages=3`. However, `backbone_features` passed to `Sam3PixelDecoder` is a tuple of size 3. As a result, `self.conv_layers[2]` and `self.norms[2]` are bypassed and do nothing.
... | closed | completed | false | 3 | [
"bug"
] | [] | 2025-12-29T15:28:42Z | 2026-02-08T08:03:51Z | 2026-02-08T08:03:51Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | james77777778 | 20,734,616 | MDQ6VXNlcjIwNzM0NjE2 | User | false |
huggingface/transformers | 3,767,691,311 | I_kwDOCUB6oc7gkmgv | 43,066 | https://github.com/huggingface/transformers/issues/43066 | https://api.github.com/repos/huggingface/transformers/issues/43066 | Wrong tokenizer decoder type in Transformers v5 | ### System Info
Wrong decoder type with `5.0.0rc1`.
### Information
- [x] The official example scripts
- [ ] My own modified scripts
### Tasks
- [x] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details below)
### Reproduction
Run this:
```pyth... | closed | completed | false | 6 | [
"bug"
] | [] | 2025-12-29T15:31:04Z | 2026-01-26T08:27:35Z | 2026-01-26T08:27:35Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | awni | 1,542,805 | MDQ6VXNlcjE1NDI4MDU= | User | false |
huggingface/transformers | 3,777,909,876 | I_kwDOCUB6oc7hLlR0 | 43,086 | https://github.com/huggingface/transformers/issues/43086 | https://api.github.com/repos/huggingface/transformers/issues/43086 | Add async_stopping_criteria flag to reduce GPU-CPU synchronization overhead | ### Feature request
Add an `async_stopping_criteria` flag to `GenerationConfig` that performs stopping criteria checks asynchronously on a separate CUDA stream. This reduces GPU-CPU synchronization overhead during autoregressive text generation by allowing the model to continue generating tokens while stopping criteri... | closed | completed | false | 5 | [] | [] | 2026-01-03T09:54:01Z | 2026-03-08T08:03:20Z | 2026-03-08T08:03:20Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | AmitMY | 5,757,359 | MDQ6VXNlcjU3NTczNTk= | User | false |
huggingface/transformers | 3,777,969,619 | I_kwDOCUB6oc7hLz3T | 43,089 | https://github.com/huggingface/transformers/issues/43089 | https://api.github.com/repos/huggingface/transformers/issues/43089 | Generation overhead: many GPU syncs per token + PyTorch dispatch overhead | # Generation overhead: 3.25 GPU syncs per token + PyTorch dispatch overhead
## System Info
- `transformers` version: 5.0.0.dev0 (main branch)
- Platform: Linux
- Python version: 3.12
- PyTorch version: 2.x with CUDA
- GPU: NVIDIA (tested)
## Who can help?
@gante @zucchini-nlp
## Information
- [x] My own modified ... | closed | completed | false | 2 | [] | [] | 2026-01-03T11:14:18Z | 2026-02-25T08:11:03Z | 2026-02-25T08:11:03Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | AmitMY | 5,757,359 | MDQ6VXNlcjU3NTczNTk= | User | false |
huggingface/transformers | 3,779,141,757 | I_kwDOCUB6oc7hQSB9 | 43,097 | https://github.com/huggingface/transformers/issues/43097 | https://api.github.com/repos/huggingface/transformers/issues/43097 | 5.0.0 `tie_embeddings_and_encoder_decoder` removed without indication | ### System Info
- `transformers` version: 5.0.0.dev0 (main branch)
- Platform: Linux
- Python version: 3.12
- PyTorch version: 2.x with CUDA
- GPU: NVIDIA (tested)
### Who can help?
@stevhliu
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported... | closed | completed | false | 6 | [
"bug"
] | [] | 2026-01-04T11:23:50Z | 2026-02-24T08:10:39Z | 2026-02-24T08:10:39Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | AmitMY | 5,757,359 | MDQ6VXNlcjU3NTczNTk= | User | false |
huggingface/transformers | 3,779,206,925 | I_kwDOCUB6oc7hQh8N | 43,099 | https://github.com/huggingface/transformers/issues/43099 | https://api.github.com/repos/huggingface/transformers/issues/43099 | Question about .T usage when loading video frames from file paths | Thank you very much for the excellent and high-quality work on this project.
While reading the code and experimenting with loading video frames via a list of image paths, I had a question regarding a specific implementation detail and would appreciate some clarification.
In processing_utils.py, for the case where vid... | closed | completed | false | 2 | [] | [] | 2026-01-04T12:41:09Z | 2026-02-12T08:09:45Z | 2026-02-12T08:09:45Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | c1ircle | 106,308,364 | U_kgDOBlYjDA | User | false |
huggingface/transformers | 3,782,023,965 | I_kwDOCUB6oc7hbRsd | 43,116 | https://github.com/huggingface/transformers/issues/43116 | https://api.github.com/repos/huggingface/transformers/issues/43116 | Multi-label classification always returns empty results in run_classification.py example script | ### System Info
Dell workstation with NVIDIA Titan Xp (12 GB RAM), driver version 535.261.03, CUDA 12.2.
Ubuntu Linux 24.04, Python 3.12.
### Who can help?
I'm using `run_classification.py` example script (in `pytorch/text-classification` folder), but when running with multi-labelled data it always returns empty val... | closed | completed | false | 9 | [
"bug"
] | [] | 2026-01-05T16:03:51Z | 2026-02-18T08:09:48Z | 2026-02-18T08:09:48Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | ziorufus | 3,517,832 | MDQ6VXNlcjM1MTc4MzI= | User | false |
huggingface/transformers | 3,782,800,686 | I_kwDOCUB6oc7hePUu | 43,122 | https://github.com/huggingface/transformers/issues/43122 | https://api.github.com/repos/huggingface/transformers/issues/43122 | Different tokenization with same tokenizer from 4.57.3 to 5.0 | ### System Info
Moving from transformers 4.57.3 to 5.0+ introduces a different and seemingly incorrect tokenization when using the same tokenizer.
I believe the new version is incorrect because when using it, we get bad results (the model starts to introduce unexpected artifacts in the response).
### Who can help?
... | closed | completed | false | 3 | [
"Fast Tokenizers",
"bug"
] | [] | 2026-01-05T20:40:19Z | 2026-01-26T08:27:15Z | 2026-01-26T08:27:15Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | awni | 1,542,805 | MDQ6VXNlcjE1NDI4MDU= | User | false |
huggingface/transformers | 3,784,637,664 | I_kwDOCUB6oc7hlPzg | 43,125 | https://github.com/huggingface/transformers/issues/43125 | https://api.github.com/repos/huggingface/transformers/issues/43125 | Saving bug using FSDP 2.0 + parallelism_config set (while works fine without providing parallelism_config in TrainingArguments) | The reason is the order of elifs here
https://github.com/huggingface/transformers/blob/main/src/transformers/trainer.py#L4024
elif getattr(self.accelerator, "parallelism_config", None) is not None:
# DeepSpeed SP already handles checkpoint saving below, so skip manual save in that case
... | closed | completed | false | 5 | [] | [] | 2026-01-06T10:28:26Z | 2026-01-24T17:30:07Z | 2026-01-24T17:30:07Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | d5031 | 46,198,216 | MDQ6VXNlcjQ2MTk4MjE2 | User | false |
huggingface/transformers | 3,799,228,707 | I_kwDOCUB6oc7ic6Ej | 43,208 | https://github.com/huggingface/transformers/issues/43208 | https://api.github.com/repos/huggingface/transformers/issues/43208 | [xLSTM] Three bugs preventing training models smaller than 7B | ### System Info
- transformers version: 5.0.0.dev0
- Platform: Linux/macOS
- Python: 3.12
### Who can help?
@ArthurZucker @vasqu
### Information
- The official example scripts
- My own modified scripts
### Reproduction
```python
from transformers import xLSTMConfig, xLSTMForCausalLM
import torch
# Config for ~125M... | closed | completed | false | 1 | [] | [] | 2026-01-10T06:52:23Z | 2026-02-09T18:05:45Z | 2026-02-09T18:05:45Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | Anri-Lombard | 76,818,211 | MDQ6VXNlcjc2ODE4MjEx | User | false |
huggingface/transformers | 3,804,003,927 | I_kwDOCUB6oc7ivH5X | 43,232 | https://github.com/huggingface/transformers/issues/43232 | https://api.github.com/repos/huggingface/transformers/issues/43232 | _update_model_kwargs_for_generation after sync_gpus when generation | ### System Info
- `transformers` version: main (latest)
- Platform: Linux
- Python version: 3.11
- PyTorch version: 2.9
### Who can help?
@gante
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQ... | closed | completed | false | 5 | [
"bug"
] | [] | 2026-01-12T11:55:52Z | 2026-03-05T08:07:26Z | 2026-03-05T08:07:26Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | xin-w8023 | 43,900,898 | MDQ6VXNlcjQzOTAwODk4 | User | false |
huggingface/transformers | 3,806,595,287 | I_kwDOCUB6oc7i5AjX | 43,240 | https://github.com/huggingface/transformers/issues/43240 | https://api.github.com/repos/huggingface/transformers/issues/43240 | kwargs are not passed to loss calculation function. | https://github.com/huggingface/transformers/blob/3aa89c07f210df18865daee9df81fe2766d13884/src/transformers/loss/loss_utils.py#L36
If we want to use label_smoothing here. The loss_kwargs can be passed into the fixed_cross_entropy but not used. | closed | completed | false | 2 | [] | [] | 2026-01-13T00:51:17Z | 2026-02-21T08:03:15Z | 2026-02-21T08:03:15Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | XinhaoMei | 58,569,453 | MDQ6VXNlcjU4NTY5NDUz | User | false |
huggingface/transformers | 3,809,161,904 | I_kwDOCUB6oc7jCzKw | 43,257 | https://github.com/huggingface/transformers/issues/43257 | https://api.github.com/repos/huggingface/transformers/issues/43257 | Qwen3 MOE weights not converted when loading with accelerate + deepspeed | ### System Info
```
- `transformers` version: 5.0.0.dev0
- Platform: Linux-5.15.0-1048-aws-x86_64-with-glibc2.31
- Python version: 3.12.11
- Huggingface_hub version: 1.3.1
- Safetensors version: 0.7.0
- Accelerate version: 1.12.0
- Accelerate config: not found
- DeepSpeed version: 0.18.3
- PyTorch version (accelerat... | closed | completed | false | 11 | [
"bug"
] | [
"kashif"
] | 2026-01-13T14:39:25Z | 2026-02-11T19:46:19Z | 2026-01-24T17:16:33Z | MEMBER | null | 20260325T173244Z | 2026-03-25T17:32:44Z | edbeeching | 7,275,864 | MDQ6VXNlcjcyNzU4NjQ= | User | false |
huggingface/transformers | 3,809,998,924 | I_kwDOCUB6oc7jF_hM | 43,262 | https://github.com/huggingface/transformers/issues/43262 | https://api.github.com/repos/huggingface/transformers/issues/43262 | Audio processors: `apply_chat_template()` defaults to 16kHz sampling rate, even if the processor config sets a different value | Firstly, thanks for a fantastic library!
### System Info
- `transformers` version: 4.57.5
- Platform: macOS-26.2-arm64-arm-64bit-Mach-O
- Python version: 3.14.2
- Huggingface_hub version: 0.36.0
- Safetensors version: 0.7.0
- Accelerate version: not installed
- Accelerate config: not found
- DeepSpeed version: not in... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-01-13T18:18:01Z | 2026-02-02T11:32:38Z | 2026-02-02T11:32:38Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | rmclarke | 25,023,912 | MDQ6VXNlcjI1MDIzOTEy | User | false |
huggingface/transformers | 3,812,129,522 | I_kwDOCUB6oc7jOHry | 43,278 | https://github.com/huggingface/transformers/issues/43278 | https://api.github.com/repos/huggingface/transformers/issues/43278 | Embedding layer dtype changed from BF16 in training to FP32 in evaluate. | ### System Info
- `transformers` version: 4.57.1
- Platform: Linux-5.15.0-1071-azure-x86_64-with-glibc2.35
- Python version: 3.12.12
- Huggingface_hub version: 0.36.0
- Safetensors version: 0.7.0
- Accelerate version: 1.12.0
- Accelerate config: not found
- DeepSpeed version: 0.18.4
- PyTorch version (accelerator?)... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-01-14T08:23:51Z | 2026-02-23T08:10:44Z | 2026-02-23T08:10:44Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | TriLoo | 16,267,477 | MDQ6VXNlcjE2MjY3NDc3 | User | false |
huggingface/transformers | 3,813,016,307 | I_kwDOCUB6oc7jRgLz | 43,284 | https://github.com/huggingface/transformers/issues/43284 | https://api.github.com/repos/huggingface/transformers/issues/43284 | Customize Quantization-Friendly Backward Compatibility | ### Feature request
Hi guys,
It’s great to see Transformers moving to V5 with a modular design and improved performance! We’ve started adapting it with the RC branch, but noticed that some of the changes are not very friendly for quantization tools.
Here’s the context:
Given a BF16 model, quantization tools typically ... | closed | completed | false | 20 | [
"Feature request"
] | [] | 2026-01-14T12:32:00Z | 2026-03-02T10:10:49Z | 2026-03-02T10:10:49Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | yiliu30 | 106,061,964 | U_kgDOBlJgjA | User | false |
huggingface/transformers | 3,814,786,615 | I_kwDOCUB6oc7jYQY3 | 43,295 | https://github.com/huggingface/transformers/issues/43295 | https://api.github.com/repos/huggingface/transformers/issues/43295 | [Regression] v4.57.5 breaks custom model code accessing processor.tokenizer and passing images to tokenizer | ### System Info
- `transformers` version: 4.57.5
- Platform: Linux-5.15.0-161-generic-x86_64-with-glibc2.35
- Python version: 3.12.12
- Huggingface_hub version: 0.36.0
- Safetensors version: 0.7.0
- Accelerate version: 1.12.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accel... | closed | completed | false | 3 | [
"bug"
] | [] | 2026-01-14T20:37:49Z | 2026-02-23T08:10:42Z | 2026-02-23T08:10:42Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | AndreasKaratzas | 42,451,412 | MDQ6VXNlcjQyNDUxNDEy | User | false |
huggingface/transformers | 3,815,747,215 | I_kwDOCUB6oc7jb66P | 43,296 | https://github.com/huggingface/transformers/issues/43296 | https://api.github.com/repos/huggingface/transformers/issues/43296 | Failed to load PaddleOCR-VL model with transformers 4.53.0 in vLLM 0.11.0 | ### System Info
transformers 4.53.0
platform Linux
python version 3.11.4
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or datas... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-01-15T02:52:00Z | 2026-02-10T02:58:27Z | 2026-02-10T02:58:27Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | jhlddz | 51,847,122 | MDQ6VXNlcjUxODQ3MTIy | User | false |
huggingface/transformers | 3,816,324,669 | I_kwDOCUB6oc7jeH49 | 43,298 | https://github.com/huggingface/transformers/issues/43298 | https://api.github.com/repos/huggingface/transformers/issues/43298 | continous batching not support audio Model | ### Feature request
<img width="719" height="457" alt="Image" src="https://github.com/user-attachments/assets/ac0ec577-6568-4e21-a3c9-78e1d3ab846d" />
continous batching not support audio Model input_ids，audio_model input_features(is a tensor([1, 80, 1280]))，please support~
### Motivation
please support continous ba... | open | null | false | 5 | [
"Feature request"
] | [
"remi-or"
] | 2026-01-15T07:23:05Z | 2026-03-13T22:42:27Z | null | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | MengLeebin | 13,679,493 | MDQ6VXNlcjEzNjc5NDkz | User | false |
huggingface/transformers | 3,816,484,709 | I_kwDOCUB6oc7jeu9l | 43,299 | https://github.com/huggingface/transformers/issues/43299 | https://api.github.com/repos/huggingface/transformers/issues/43299 | Transformer version: 5.0.0.dev0 breaks Qwen3VL Moe models loading | ### System Info
I am trying to run inference on Qwen3VL Moe models, both 30B-A3B and 235B-A22B; however, there is a size mismatch between the HF checkpoint and the model weights. Inference done on H100 GPU with standard inference script from Qwen repo. Error message shown below:
Note: downgrading transformer version... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-01-15T08:18:25Z | 2026-02-11T12:43:58Z | 2026-01-16T11:27:21Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | daulettoibazar | 113,344,861 | U_kgDOBsGBXQ | User | false |
huggingface/transformers | 3,821,478,506 | I_kwDOCUB6oc7jxyJq | 43,316 | https://github.com/huggingface/transformers/issues/43316 | https://api.github.com/repos/huggingface/transformers/issues/43316 | API discrepancy between `Gemma3TextConfig` and others | ### System Info
transformers==v5.0.0rc3
### Who can help?
@zucchini-nlp
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details below)
### Reproduc... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-01-16T10:47:00Z | 2026-01-26T09:12:56Z | 2026-01-26T09:12:56Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | Tcc0403 | 76,503,978 | MDQ6VXNlcjc2NTAzOTc4 | User | false |
huggingface/transformers | 3,821,662,877 | I_kwDOCUB6oc7jyfKd | 43,317 | https://github.com/huggingface/transformers/issues/43317 | https://api.github.com/repos/huggingface/transformers/issues/43317 | device_map=auto fails to load the dequantized model on gpu+cpu offload | ### System Info
On a A100 using transformers from main and latest torch
### Who can help?
@Cyrilvallez
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (gi... | closed | completed | false | 0 | [
"bug"
] | [] | 2026-01-16T11:35:02Z | 2026-01-26T12:30:24Z | 2026-01-26T12:30:24Z | MEMBER | null | 20260325T173244Z | 2026-03-25T17:32:44Z | IlyasMoutawwakil | 57,442,720 | MDQ6VXNlcjU3NDQyNzIw | User | false |
huggingface/transformers | 3,822,275,080 | I_kwDOCUB6oc7j00oI | 43,322 | https://github.com/huggingface/transformers/issues/43322 | https://api.github.com/repos/huggingface/transformers/issues/43322 | Segmentation Fault when loading Llava Next Models | ### Running in Segmentation Fault when trying to run LLaVA-NeXT-Video-34B/7B
Running inference with LLaVA-NeXT-Video-34B (and the 7B) consistently results in a segmentation fault during generation on CUDA. This occurs on an NVIDIA H200 (140GB) GPU with sufficient memory available.
I tested on other models like Qwen2... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-01-16T14:28:51Z | 2026-02-24T08:10:31Z | 2026-02-24T08:10:31Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | omrastogi | 43,903,014 | MDQ6VXNlcjQzOTAzMDE0 | User | false |
huggingface/transformers | 3,824,075,616 | I_kwDOCUB6oc7j7sNg | 43,329 | https://github.com/huggingface/transformers/issues/43329 | https://api.github.com/repos/huggingface/transformers/issues/43329 | [BUG] _get_num_multimodal_tokens: video branch uses undefined a) get_number_of_video_patches, b)merge_size. Tests never hit video route (multiple VLM processors) | ### System Info
- `transformers` version: 4.57.3
- Platform: Linux-6.8.0-52-generic-x86_64-with-glibc2.35
- Python version: 3.12.9
- Huggingface_hub version: 0.34.3
- Safetensors version: 0.5.3
- Accelerate version: 1.9.0
- Accelerate config: - compute_environment: LOCAL_MACHINE
- distributed_type: MULTI_GP... | closed | completed | false | 2 | [
"bug"
] | [
"zucchini-nlp"
] | 2026-01-17T00:18:39Z | 2026-02-24T08:10:28Z | 2026-02-24T08:10:28Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | stefgina | 24,375,599 | MDQ6VXNlcjI0Mzc1NTk5 | User | false |
huggingface/transformers | 3,824,859,369 | I_kwDOCUB6oc7j-rjp | 43,334 | https://github.com/huggingface/transformers/issues/43334 | https://api.github.com/repos/huggingface/transformers/issues/43334 | Qwen3-VL can't be loaded in transformers dev: AttributeError: 'Qwen3VLTextConfig' object has no attribute 'pad_token_id' | Qwen3-VL can't be loaded in transformers dev branch (5.0.0.dev0 at commit 24807bfcf4a21286fa2a7e728f381ddaaca7bbc7):
> AttributeError: 'Qwen3VLTextConfig' object has no attribute 'pad_token_id'
After investigation, the issue appeared after the merge of this PR:
- https://github.com/huggingface/transformers/pull/41541... | closed | completed | false | 5 | [
"bug"
] | [] | 2026-01-17T10:42:01Z | 2026-01-27T09:19:51Z | 2026-01-22T18:25:38Z | MEMBER | null | 20260325T173244Z | 2026-03-25T17:32:44Z | albertvillanova | 8,515,462 | MDQ6VXNlcjg1MTU0NjI= | User | false |
huggingface/transformers | 3,824,943,220 | I_kwDOCUB6oc7j_AB0 | 43,335 | https://github.com/huggingface/transformers/issues/43335 | https://api.github.com/repos/huggingface/transformers/issues/43335 | [BUG] SwitchTransformersConfig creates sparse layer when num_sparse_encoder_layers=0 with single layer model | ### System Info
* `transformers` version: `5.0.0.dev0`
* Platform: `Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.39`
* Python version: `3.12.3`
* `huggingface_hub` version: `1.3.2`
* `safetensors` version: `0.7.0`
* `accelerate` version: `1.12.0`
* Accelerate config: `not installed`
* DeepSpeed version:... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-01-17T11:28:35Z | 2026-02-09T17:33:46Z | 2026-02-09T17:33:46Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | harshaljanjani | 75,426,551 | MDQ6VXNlcjc1NDI2NTUx | User | false |
huggingface/transformers | 3,828,221,249 | I_kwDOCUB6oc7kLgVB | 43,344 | https://github.com/huggingface/transformers/issues/43344 | https://api.github.com/repos/huggingface/transformers/issues/43344 | invalid test cases for glm_image model | ### System Info
A100 w/ 80GB memory
For test cases:
```
tests/models/glm_image/test_modeling_glm_image.py::GlmImageIntegrationTest::test_small_model_integration_test_batch_flashatt2
tests/models/glm_image/test_modeling_glm_image.py::GlmImageIntegrationTest::test_small_model_integration_test_batch
```
it will throw erro... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-01-19T06:33:52Z | 2026-01-26T15:31:00Z | 2026-01-26T15:31:00Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | kaixuanliu | 13,268,042 | MDQ6VXNlcjEzMjY4MDQy | User | false |
huggingface/transformers | 3,830,059,142 | I_kwDOCUB6oc7kShCG | 43,352 | https://github.com/huggingface/transformers/issues/43352 | https://api.github.com/repos/huggingface/transformers/issues/43352 | Error: NemotronHForCausalLM does not support Flash Attention 2.0 yet | ### System Info
Error: NemotronHForCausalLM does not support Flash Attention 2.0 yet. Please request to add support where the model is hosted, on its model hub page: https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/discussions/new or in the Transformers GitHub repo: https://github.com/huggingface/tran... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-01-19T14:50:02Z | 2026-02-28T08:02:53Z | 2026-02-28T08:02:53Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | OrlandoWhite88 | 119,964,986 | U_kgDOByaFOg | User | false |
huggingface/transformers | 3,832,557,479 | I_kwDOCUB6oc7kcC-n | 43,366 | https://github.com/huggingface/transformers/issues/43366 | https://api.github.com/repos/huggingface/transformers/issues/43366 | GGUF model with architecture gpt-oss support | ### System Info
- `transformers` version: 4.57.5
- Platform: Linux-5.4.0-193-generic-x86_64-with-glibc2.35
- Python version: 3.11.7
- Huggingface_hub version: 0.36.0
- Safetensors version: 0.7.0
- Accelerate version: not installed
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (acc... | open | null | false | 5 | [
"Good Second Issue",
"Feature request",
"bug"
] | [] | 2026-01-20T08:03:36Z | 2026-02-12T21:10:20Z | null | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | akayunov | 13,577,138 | MDQ6VXNlcjEzNTc3MTM4 | User | false |
huggingface/transformers | 3,834,922,327 | I_kwDOCUB6oc7klEVX | 43,377 | https://github.com/huggingface/transformers/issues/43377 | https://api.github.com/repos/huggingface/transformers/issues/43377 | [BUG] MIMI Encoder produces different outputs for batched vs single inputs due to missing padding mask support | ### System Info
* `transformers` version: `5.0.0.dev0`
* Platform: `Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.39`
* Python version: `3.12.3`
* `huggingface_hub` version: `1.3.2`
* `safetensors` version: `0.7.0`
* `accelerate` version: `1.12.0`
* Accelerate config: `not installed`
* DeepSpeed version:... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-01-20T18:29:14Z | 2026-03-15T08:06:43Z | 2026-03-15T08:06:43Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | harshaljanjani | 75,426,551 | MDQ6VXNlcjc1NDI2NTUx | User | false |
huggingface/transformers | 3,835,812,087 | I_kwDOCUB6oc7kodj3 | 43,381 | https://github.com/huggingface/transformers/issues/43381 | https://api.github.com/repos/huggingface/transformers/issues/43381 | Gradient checkpointing cannot be used in eval mode | ### System Info
- `transformers` version: 4.57.1
- Platform: Linux-6.17.0-8-generic-x86_64-with-glibc2.42
- Python version: 3.12.9
- Huggingface_hub version: 0.34.3
- Safetensors version: 0.5.3
- Accelerate version: 1.9.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerat... | closed | completed | false | 5 | [
"bug"
] | [] | 2026-01-20T22:58:34Z | 2026-03-02T06:53:05Z | 2026-03-01T08:02:58Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | smarter | 63,430 | MDQ6VXNlcjYzNDMw | User | false |
huggingface/transformers | 3,836,016,486 | I_kwDOCUB6oc7kpPdm | 43,383 | https://github.com/huggingface/transformers/issues/43383 | https://api.github.com/repos/huggingface/transformers/issues/43383 | Parakeet model limited to ~8 minutes audio and yet NeMo supports hours-long audio | I noticed that the current Parakeet model appears to only support ~8 minutes of audio input before running into length limitations (likely tied to max positional encoding constraints which is 5000).
However, in NeMo, same ASR models (`nvidia/parakeet-ctc-0.6b` and `nvidia/parakeet-ctc-1.1b`) can process hours-long aud... | closed | completed | false | 3 | [] | [] | 2026-01-21T00:22:36Z | 2026-03-04T08:06:00Z | 2026-03-04T08:06:00Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | penguinwang96825 | 28,087,825 | MDQ6VXNlcjI4MDg3ODI1 | User | false |
huggingface/transformers | 3,836,753,591 | I_kwDOCUB6oc7ksDa3 | 43,386 | https://github.com/huggingface/transformers/issues/43386 | https://api.github.com/repos/huggingface/transformers/issues/43386 | Support other types of model inputs for continuous batching | ### Feature request
Unless I'm mistaken, the continuous batching api currently does not seem to support any other input modality other than token ids via `input_ids`. VLM models require inputs such as `pixel_values`, which are not accepted in `add_request()`.
https://github.com/huggingface/transformers/blob/v5.0.0rc2... | open | null | false | 5 | [
"Feature request"
] | [] | 2026-01-21T06:13:13Z | 2026-03-04T15:36:33Z | null | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | yhshin11 | 5,031,800 | MDQ6VXNlcjUwMzE4MDA= | User | false |
huggingface/transformers | 3,837,507,900 | I_kwDOCUB6oc7ku7k8 | 43,388 | https://github.com/huggingface/transformers/issues/43388 | https://api.github.com/repos/huggingface/transformers/issues/43388 | gather_for_metrics incorrectly drops label elements in the last batch when labels is a tuple with several label types e.g. used by mask2former | ### System Info
accelerate==1.7.0 (but code is the same also in current 1.12.0)
transformers==4.53.0.dev0
torch==2.6.0
python3.10
### Who can help?
@yonigozlan @molbap @SunMarc
### Information
- [x] The official example scripts
- [x] My own modified scripts
### Tasks
- [x] An officially supported task in the `e... | closed | completed | false | 4 | [
"Examples",
"bug"
] | [] | 2026-01-21T10:05:53Z | 2026-03-02T08:08:56Z | 2026-03-02T08:08:56Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | J-Bracke | 74,211,542 | MDQ6VXNlcjc0MjExNTQy | User | false |
huggingface/transformers | 3,841,563,178 | I_kwDOCUB6oc7k-Zoq | 43,404 | https://github.com/huggingface/transformers/issues/43404 | https://api.github.com/repos/huggingface/transformers/issues/43404 | Bug: lm_head weight not tied in Mistral3ForConditionalGeneration (affects AutoModelForImageTextToText) | ### System Info
- `transformers` version: 5.0.0.dev0
- Platform: Linux-5.15.133+-x86_64-with-glibc2.35
- Python version: 3.12.0
- PyTorch version: 2.9.0+cu126
- CUDA/cuDNN version: 12.6
- GPU: Tesla T4 (compute capability 7.5)
### Who can help?
@zucchini-nlp
@ArthurZucker
@amyeroberts
### Information
- [x] The o... | closed | completed | false | 6 | [
"bug"
] | [] | 2026-01-22T07:06:48Z | 2026-01-26T09:30:44Z | 2026-01-26T09:30:44Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | aswin00000 | 91,382,951 | MDQ6VXNlcjkxMzgyOTUx | User | false |
huggingface/transformers | 3,842,899,772 | I_kwDOCUB6oc7lDf88 | 43,408 | https://github.com/huggingface/transformers/issues/43408 | https://api.github.com/repos/huggingface/transformers/issues/43408 | Warning: You are using a model of type sam3_video to instantiate a model of type sam3_tracker | ### System Info
When using the sample code to create a SAM3 Tracker Model like:
```python
from transformers import Sam3TrackerProcessor, Sam3TrackerModel
model = Sam3TrackerModel.from_pretrained("facebook/sam3")
processor = Sam3TrackerProcessor.from_pretrained("facebook/sam3")
```
The following warning will be displ... | closed | completed | false | 5 | [
"bug"
] | [] | 2026-01-22T13:12:57Z | 2026-02-05T13:03:34Z | 2026-02-05T13:03:34Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | gboeer | 1,067,159 | MDQ6VXNlcjEwNjcxNTk= | User | false |
huggingface/transformers | 3,843,302,597 | I_kwDOCUB6oc7lFCTF | 43,412 | https://github.com/huggingface/transformers/issues/43412 | https://api.github.com/repos/huggingface/transformers/issues/43412 | gemma3n executorch export fails missing self training guard and erfinv not supported | system info
- transformers 5.0.0rc1
- torch 2.9.1
- executorch 1.0.1
- python 3.11
- ubuntu 24.04
who can help
@ArthurZucker @younesbelkada
information
- the official example scripts
tasks
- an officially supported task in the examples folder
problem description
when exporting gemma3n models to executorch pt... | closed | completed | false | 5 | [] | [] | 2026-01-22T14:51:37Z | 2026-03-29T08:08:17Z | 2026-03-29T08:08:17Z | NONE | null | 20260407T090028Z | 2026-04-07T09:00:28Z | maceip | 804,368 | MDQ6VXNlcjgwNDM2OA== | User | false |
huggingface/transformers | 3,844,198,299 | I_kwDOCUB6oc7lIc-b | 43,421 | https://github.com/huggingface/transformers/issues/43421 | https://api.github.com/repos/huggingface/transformers/issues/43421 | [FEATURE] TokenizersBackend does not update post-processor when special tokens are modified at runtime | ### System Info
* `transformers` version: `5.0.0.dev0`
* Platform: `Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.39`
* Python version: `3.12.3`
* `huggingface_hub` version: `1.3.2`
* `safetensors` version: `0.7.0`
* `accelerate` version: `1.12.0`
* Accelerate config: `not installed`
* DeepSpeed version:... | closed | completed | false | 3 | [
"bug"
] | [] | 2026-01-22T18:29:02Z | 2026-03-09T08:09:27Z | 2026-03-09T08:09:27Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | harshaljanjani | 75,426,551 | MDQ6VXNlcjc1NDI2NTUx | User | false |
huggingface/transformers | 3,844,895,799 | I_kwDOCUB6oc7lLHQ3 | 43,425 | https://github.com/huggingface/transformers/issues/43425 | https://api.github.com/repos/huggingface/transformers/issues/43425 | Torch 2.10 incompatible | ### System Info
transformers 4.57.6
torch 2.10
accelerate 1.12.0
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give det... | closed | not_planned | false | 9 | [
"bug"
] | [] | 2026-01-22T21:51:30Z | 2026-03-06T22:09:41Z | 2026-02-06T22:24:23Z | MEMBER | null | 20260325T173244Z | 2026-03-25T17:32:44Z | qgallouedec | 45,557,362 | MDQ6VXNlcjQ1NTU3MzYy | User | false |
huggingface/transformers | 3,846,965,862 | I_kwDOCUB6oc7lTApm | 43,441 | https://github.com/huggingface/transformers/issues/43441 | https://api.github.com/repos/huggingface/transformers/issues/43441 | [BUG] Ministral-3 fails with FlashAttention in Transformers v5 RC | ### System Info
**Description**
In Transformers `5.0.0rc3`, running `mistralai/Ministral-3-8B-Instruct-2512` with FlashAttention enabled results in an `IndexError` during the attention forward pass.
This issue does not occur when FlashAttention is disabled.
---
**Model**
* `mistralai/Ministral-3-8B-Instruct-2512`... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-01-23T11:00:47Z | 2026-01-26T08:18:44Z | 2026-01-26T08:18:44Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | dysby | 28,685,434 | MDQ6VXNlcjI4Njg1NDM0 | User | false |
huggingface/transformers | 3,847,750,539 | I_kwDOCUB6oc7lWAOL | 43,450 | https://github.com/huggingface/transformers/issues/43450 | https://api.github.com/repos/huggingface/transformers/issues/43450 | Video processors return incorrect shape when input is batched | ### System Info
- `transformers` version: 4.57.6
- Platform: Linux-5.10.0-37-cloud-amd64-x86_64-with-glibc2.31
- Python version: 3.10.19
- Huggingface_hub version: 0.36.0
- Safetensors version: 0.7.0
- Accelerate version: not installed
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version... | closed | completed | false | 0 | [
"bug"
] | [] | 2026-01-23T14:42:11Z | 2026-01-30T10:27:51Z | 2026-01-30T10:27:51Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | shaform | 367,172 | MDQ6VXNlcjM2NzE3Mg== | User | false |
huggingface/transformers | 3,848,223,339 | I_kwDOCUB6oc7lXzpr | 43,452 | https://github.com/huggingface/transformers/issues/43452 | https://api.github.com/repos/huggingface/transformers/issues/43452 | gguf_file breaks for AutoTokenizer.from_pretrained and AutoModelForCausalLM.from_pretrained | ### System Info
- `transformers` version: 5.0.0.dev0
- Platform: macOS-15.6-arm64-arm-64bit-Mach-O
- Python version: 3.13.2
- Huggingface_hub version: 1.3.1
- Safetensors version: 0.5.3
- Accelerate version: 1.12.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?): 2... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-01-23T16:34:19Z | 2026-01-24T18:07:17Z | 2026-01-24T18:07:17Z | MEMBER | null | 20260325T173244Z | 2026-03-25T17:32:44Z | xenova | 26,504,141 | MDQ6VXNlcjI2NTA0MTQx | User | false |
huggingface/transformers | 3,848,651,246 | I_kwDOCUB6oc7lZcHu | 43,454 | https://github.com/huggingface/transformers/issues/43454 | https://api.github.com/repos/huggingface/transformers/issues/43454 | [BUG] AyaVisionConfig fails to tie lm_head weights causing garbage text generation | ### System Info
* `transformers` version: `5.0.0.dev0`
* Platform: `Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.39`
* Python version: `3.12.3`
* `huggingface_hub` version: `1.3.2`
* `safetensors` version: `0.7.0`
* `accelerate` version: `1.12.0`
* Accelerate config: `not installed`
* DeepSpeed version:... | closed | completed | false | 0 | [
"bug"
] | [] | 2026-01-23T18:28:13Z | 2026-01-24T17:12:22Z | 2026-01-24T17:12:22Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | harshaljanjani | 75,426,551 | MDQ6VXNlcjc1NDI2NTUx | User | false |
huggingface/transformers | 3,852,393,114 | I_kwDOCUB6oc7lntqa | 43,472 | https://github.com/huggingface/transformers/issues/43472 | https://api.github.com/repos/huggingface/transformers/issues/43472 | Introduce standardized BatchLinear module for MoE architectures to facilitate PEFT and Quantization | ### Feature request
Starting from transformer v5, Linear module of Moe experts are fused into one single module. While this improve speed of Moe, it is making downstream library difficult to adapt to this change.
I propose introducing a standardized `BatchLinear` (or `MoELinear`) module within transformers. This modu... | open | null | false | 3 | [
"Feature request"
] | [] | 2026-01-25T01:50:15Z | 2026-01-25T16:15:17Z | null | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | ITcarrot | 41,422,161 | MDQ6VXNlcjQxNDIyMTYx | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.