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 | 4,256,151,805 | I_kwDOCUB6oc79r7j9 | 45,412 | https://github.com/huggingface/transformers/issues/45412 | https://api.github.com/repos/huggingface/transformers/issues/45412 | RT-DETR models do not release memory when deleted / garbage-collected | ### System Info
Transfomers: 5.5.3
PyTorch: 2.8.0+cu126
TorchVision: 0.23.0+cu126
System: Debian 13 (trixie)
Python: 3.13.5
### Who can help?
@yonigozlan @molbap
### Information
- [x] The official example scripts
- [x] My own modified scripts
### Tasks
- [x] An officially supported task in the `examples` folder ... | closed | completed | false | 8 | [
"bug"
] | [] | 2026-04-13T15:46:43Z | 2026-05-15T20:21:31Z | 2026-05-15T20:21:31Z | NONE | null | 20260516T000042Z | 2026-05-16T00:00:42Z | dhdaines | 3,325,008 | MDQ6VXNlcjMzMjUwMDg= | User | false |
huggingface/transformers | 4,258,468,813 | I_kwDOCUB6oc790xPN | 45,419 | https://github.com/huggingface/transformers/issues/45419 | https://api.github.com/repos/huggingface/transformers/issues/45419 | Chat template inconsistencies in tool-calling support | Chat templates across model families handle tool-calling messages inconsistently. This creates fragility for any library (like TRL) that needs to construct tool-calling conversations programmatically, since there's no single "safe" way to build an assistant message with `tool_calls`.
I ran a systematic check across al... | open | null | false | 2 | [] | [] | 2026-04-13T23:27:06Z | 2026-04-30T17:11:09Z | null | MEMBER | null | 20260501T113108Z | 2026-05-01T11:31:08Z | qgallouedec | 45,557,362 | MDQ6VXNlcjQ1NTU3MzYy | User | false |
huggingface/transformers | 4,262,515,718 | I_kwDOCUB6oc7-ENQG | 45,431 | https://github.com/huggingface/transformers/issues/45431 | https://api.github.com/repos/huggingface/transformers/issues/45431 | Wrong checkpoint path in Dinov2 model_docs | Wrong checkpoint path in Dinov2 model_docs.
The current checkpoint "google/dinov2-base-patch16-224" does not exist. The correct one should be "facebook/dinov-base".
This issue is fixed in PR #45430
### Who can help?
@yonigozlan @molbap @stevhliu
### Notes
This is a minor issue, but it might help new users. Than... | closed | completed | false | 0 | [] | [] | 2026-04-14T13:53:21Z | 2026-04-15T09:02:16Z | 2026-04-15T09:02:16Z | CONTRIBUTOR | null | 20260415T224019Z | 2026-04-15T22:40:19Z | ambroiseodt | 64,415,312 | MDQ6VXNlcjY0NDE1MzEy | User | false |
huggingface/transformers | 4,263,833,996 | I_kwDOCUB6oc7-JPGM | 45,440 | https://github.com/huggingface/transformers/issues/45440 | https://api.github.com/repos/huggingface/transformers/issues/45440 | Native `DeepseekV3MoE` diverges from the remote DeepSeekV3 implementation | ### System Info
Hello, the `DeepseekV3MoE` class in transformers (native) differs from the official remote DeepSeekV3
implementation (which was updated for a bug but not in `transformers`, hence the difference).
<details>
<summary> See trf DeepSeekV3 MoE code </summary>
https://github.com/huggingface/transformers/bl... | closed | completed | false | 0 | [
"bug"
] | [] | 2026-04-14T17:50:49Z | 2026-04-22T04:53:27Z | 2026-04-22T04:53:26Z | CONTRIBUTOR | null | 20260422T060051Z | 2026-04-22T06:00:51Z | casinca | 47,400,729 | MDQ6VXNlcjQ3NDAwNzI5 | User | false |
huggingface/transformers | 4,266,475,063 | I_kwDOCUB6oc7-TT43 | 45,446 | https://github.com/huggingface/transformers/issues/45446 | https://api.github.com/repos/huggingface/transformers/issues/45446 | Incorrect PyTorch version check for AuxRequest import in flex_attention | ### System Info
In src/transformers/integrations/flex_attention.py, the code currently checks for PyTorch version >= 2.9.0 to import AuxRequest from torch.nn.attention.flex_attention. However, AuxRequest was actually introduced in PyTorch 2.9.1.
According to the official PyTorch documentation, AuxRequest is available ... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-04-15T05:26:22Z | 2026-04-15T11:36:37Z | 2026-04-15T11:35:34Z | NONE | null | 20260415T224019Z | 2026-04-15T22:40:19Z | ZSLsherly | 142,322,697 | U_kgDOCHusCQ | User | false |
huggingface/transformers | 4,266,677,435 | I_kwDOCUB6oc7-UFS7 | 45,447 | https://github.com/huggingface/transformers/issues/45447 | https://api.github.com/repos/huggingface/transformers/issues/45447 | granitemoehybrid: HybridMambaAttentionDynamicCache missing from modeling_granitemoehybrid — breaks ibm-granite/granite-4.0-3b-vision remote code | ## Summary
The `ibm-granite/granite-4.0-3b-vision` model's remote `modeling.py` imports `HybridMambaAttentionDynamicCache` from `transformers.models.granitemoehybrid.modeling_granitemoehybrid`. This class does not exist in transformers 5.5.4 (latest) or on the current `main` branch, causing an `ImportError` whenever a... | open | null | false | 4 | [] | [] | 2026-04-15T06:13:12Z | 2026-05-15T08:58:11Z | null | NONE | null | 20260515T120027Z | 2026-05-15T12:00:27Z | Steve-Allison | 3,996,420 | MDQ6VXNlcjM5OTY0MjA= | User | false |
huggingface/transformers | 4,268,948,257 | I_kwDOCUB6oc7-cvsh | 45,458 | https://github.com/huggingface/transformers/issues/45458 | https://api.github.com/repos/huggingface/transformers/issues/45458 | Add typing support incrementally (meta issue) | We’re progressively adding typing support to the codebase using `ty`. This issue tracks the overall progress as we extend coverage directory by directory.
# Current status
The tooling is already in place. Type checking is enabled for a subset of directories
You can run it locally with:
```
make typing
```
# How to... | closed | completed | false | 0 | [] | [
"tarekziade"
] | 2026-04-15T12:36:24Z | 2026-04-30T07:55:39Z | 2026-04-30T07:55:39Z | MEMBER | null | 20260430T120024Z | 2026-04-30T12:00:24Z | tarekziade | 250,019 | MDQ6VXNlcjI1MDAxOQ== | User | false |
huggingface/transformers | 4,269,023,860 | I_kwDOCUB6oc7-dCJ0 | 45,459 | https://github.com/huggingface/transformers/issues/45459 | https://api.github.com/repos/huggingface/transformers/issues/45459 | `except import_protobuf_decode_error()` hides real tokenizer errors when protobuf isn't installed | ### System Info
transformers 5.5.4 (latest release) and 5.6.0.dev0 (main).
`PreTrainedTokenizerBase._from_pretrained` has `except import_protobuf_decode_error():` at `src/transformers/tokenization_utils_base.py:1919` (line 1933 on main). The helper raises `ImportError` when protobuf isn't installed. The except-class ... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-04-15T12:48:42Z | 2026-04-20T12:33:17Z | 2026-04-20T12:33:17Z | CONTRIBUTOR | null | 20260420T180040Z | 2026-04-20T18:00:40Z | jw9603 | 70,795,645 | MDQ6VXNlcjcwNzk1NjQ1 | User | false |
huggingface/transformers | 4,270,313,753 | I_kwDOCUB6oc7-h9EZ | 45,464 | https://github.com/huggingface/transformers/issues/45464 | https://api.github.com/repos/huggingface/transformers/issues/45464 | chat/completions API fail on Qwen3.5-0.8B for streaming inference | ### System Info
- `transformers` version: 5.5.0 - 5.5.4
- Platform: macOS-26.4.1-arm64-arm-64bit-Mach-O
- Python version: 3.14.3
- Huggingface_hub version: 1.10.2
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerat... | open | null | false | 3 | [
"bug"
] | [] | 2026-04-15T16:36:31Z | 2026-05-16T08:31:39Z | null | NONE | null | 20260516T120032Z | 2026-05-16T12:00:32Z | zhangwei217245 | 346,451 | MDQ6VXNlcjM0NjQ1MQ== | User | false |
huggingface/transformers | 4,273,783,730 | I_kwDOCUB6oc7-vMOy | 45,468 | https://github.com/huggingface/transformers/issues/45468 | https://api.github.com/repos/huggingface/transformers/issues/45468 | [BUG] Gemma-4 Gemma4AudioRelPositionalEncoding | ### System Info
N/A.
### Who can help?
The hard coded numbers **12** and **-1** seem to be related to `attention_context_left` and `attention_context_right`.
https://github.com/huggingface/transformers/blob/8426e7e63d49d9c3b5f0c09d43e792a59c75c62c/src/transformers/models/gemma4/modular_gemma4.py#L160
@eustlb @ebez... | open | null | false | 5 | [
"bug"
] | [] | 2026-04-16T06:52:43Z | 2026-04-23T14:43:10Z | null | CONTRIBUTOR | null | 20260423T180030Z | 2026-04-23T18:00:30Z | foldl | 4,046,440 | MDQ6VXNlcjQwNDY0NDA= | User | false |
huggingface/transformers | 4,276,519,756 | I_kwDOCUB6oc7-5oNM | 45,478 | https://github.com/huggingface/transformers/issues/45478 | https://api.github.com/repos/huggingface/transformers/issues/45478 | [BUG] transformers>=5.4.0, Qwen3.5 Moe from_pretrained error | ### System Info
https://github.com/huggingface/transformers/issues/45310
This issue has not been fixed in the main branch.
```
import os
os.environ['CUDA_VISIBLE_DEVICS'] = '0'
from transformers import Qwen3_5ForConditionalGeneration, AutoTokenizer
model = Qwen3_5ForConditionalGeneration.from_pretrained('Qwen/Qwe... | closed | completed | false | 4 | [
"Should Fix",
"bug"
] | [] | 2026-04-16T14:48:15Z | 2026-04-20T01:35:12Z | 2026-04-20T01:31:00Z | CONTRIBUTOR | null | 20260420T060046Z | 2026-04-20T06:00:46Z | Jintao-Huang | 45,290,347 | MDQ6VXNlcjQ1MjkwMzQ3 | User | false |
huggingface/transformers | 4,276,582,143 | I_kwDOCUB6oc7-53b_ | 45,479 | https://github.com/huggingface/transformers/issues/45479 | https://api.github.com/repos/huggingface/transformers/issues/45479 | `problem_type="single_label_classification"` with `num_labels=1` leads to degenerate zero loss across multiple sequence-classification models | ### System Info
Hi, I found what looks like a library-wide issue in `transformers` affecting multiple `ForSequenceClassification` models, not just ModernBERT.
If a model is initialized with:
```python
num_labels=1
problem_type="single_label_classification"
```
the forward pass uses `CrossEntropyLoss()` with only one ... | closed | completed | false | 3 | [
"bug"
] | [] | 2026-04-16T14:58:54Z | 2026-04-24T16:51:31Z | 2026-04-24T16:51:31Z | NONE | null | 20260424T180025Z | 2026-04-24T18:00:25Z | BohdanBabii | 73,220,903 | MDQ6VXNlcjczMjIwOTAz | User | false |
huggingface/transformers | 4,276,916,345 | I_kwDOCUB6oc7-7JB5 | 45,482 | https://github.com/huggingface/transformers/issues/45482 | https://api.github.com/repos/huggingface/transformers/issues/45482 | Gemma4 26B-A4B: cross-device errors with CPU offload (RoPE, inputs, layer_scalar, SDPA mask, mm_token_type_ids) | # Bug: Gemma4 cross-device tensor errors with accelerate CPU offload
## Environment
- transformers latest (Gemma4 support, `modeling_gemma4.py`)
- Gemma4 26B-A4B-it (MoE, 4B active params)
- `accelerate` device_map with CPU offload (layers overflow to RAM)
- BnB INT8 + PEFT LoRA + Gradient Checkpointing
- RTX 4090 (2... | open | null | false | 3 | [] | [] | 2026-04-16T15:57:28Z | 2026-04-20T12:19:31Z | null | NONE | null | 20260420T180040Z | 2026-04-20T18:00:40Z | sirfyyn | 31,549,942 | MDQ6VXNlcjMxNTQ5OTQy | User | false |
huggingface/transformers | 647,983,215 | MDU6SXNzdWU2NDc5ODMyMTU= | 5,391 | https://github.com/huggingface/transformers/issues/5391 | https://api.github.com/repos/huggingface/transformers/issues/5391 | Training a GPT-2 from scratch in Greek-text, results in a low perplexity score of 7 after 15 epochs. Is it normal that score? | # ❓ Questions & Help
<!-- The GitHub issue tracker is primarly intended for bugs, feature requests,
new models and benchmarks, and migration questions. For all other questions,
we direct you to Stack Overflow (SO) where a whole community of PyTorch and
Tensorflow enthusiast can help you out. Make s... | closed | completed | false | 3 | [
"wontfix"
] | [] | 2020-06-30T08:37:47Z | 2026-02-09T16:33:04Z | 2020-09-13T17:12:37Z | NONE | null | 20260325T173244Z | 2026-03-25T17:32:44Z | Nkonstan | 35,643,708 | MDQ6VXNlcjM1NjQzNzA4 | User | false |
huggingface/transformers | 665,862,946 | MDU6SXNzdWU2NjU4NjI5NDY= | 6,045 | https://github.com/huggingface/transformers/issues/6045 | https://api.github.com/repos/huggingface/transformers/issues/6045 | Test BART's memory consumption | - this can run on GPU only and be marked `@slow`
- check how much memory bart is using at `__init__`
- assert that it doesn't use more than 110% of that.
- check how much memory bart uses on a single forward pass. (optionally test this in fp16).
- assert that it doesn't use more than 110% of that.
- check how much... | closed | completed | false | 11 | [
"Help wanted",
"Tests",
"Benchmarks",
"WIP"
] | [
"stas00",
"patrickvonplaten"
] | 2020-07-26T21:24:39Z | 2026-02-10T13:24:28Z | 2026-02-10T13:13:03Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | sshleifer | 6,045,025 | MDQ6VXNlcjYwNDUwMjU= | User | false |
huggingface/transformers | 718,927,476 | MDU6SXNzdWU3MTg5Mjc0NzY= | 7,715 | https://github.com/huggingface/transformers/issues/7715 | https://api.github.com/repos/huggingface/transformers/issues/7715 | examples/rag: test coverage, tiny model | Disclaimer: I don't know this code very well, this may be much harder than it seems.
Blocking PR: #7713
[`examples/rag/finetune.py`, `examples/rag/finetune.sh`, `eval_rag.py`] do not seem to be tested at all.
It would be good to have a `test_finetune.py` like `examples/seq2seq` that tested these.
cc @stas00 ... | closed | completed | false | 6 | [
"Help wanted",
"Tests",
"rag",
"Feature request"
] | [] | 2020-10-11T21:09:58Z | 2026-02-10T13:24:36Z | 2026-02-10T13:07:03Z | CONTRIBUTOR | null | 20260325T173244Z | 2026-03-25T17:32:44Z | sshleifer | 6,045,025 | MDQ6VXNlcjYwNDUwMjU= | User | false |
huggingface/transformers | 2,501,014,784 | I_kwDOCUB6oc6VEnUA | 33,260 | https://github.com/huggingface/transformers/issues/33260 | https://api.github.com/repos/huggingface/transformers/issues/33260 | Community contribution: Adding GGUF support for more architectures | ### Feature request
Recently, we have added the ability to load `gguf` files within [transformers](https://huggingface.co/docs/hub/en/gguf).
<img src="https://github.com/user-attachments/assets/61df6455-6016-449e-a37f-9dfc7f918902" width="600">
The goal was to offer the possibility to users to further train/fine-tu... | open | null | false | 47 | [
"Good Second Issue",
"Feature request"
] | [] | 2024-09-02T13:41:47Z | 2026-04-21T04:38:15Z | null | MEMBER | null | 20260421T060039Z | 2026-04-21T06:00:39Z | SunMarc | 57,196,510 | MDQ6VXNlcjU3MTk2NTEw | User | false |
huggingface/transformers | 4,281,338,037 | I_kwDOCUB6oc7_MAi1 | 45,488 | https://github.com/huggingface/transformers/issues/45488 | https://api.github.com/repos/huggingface/transformers/issues/45488 | LlamaTokenizer in v5 overrides tokenizer.json's ByteLevel pre-tokenizer with Metaspace, silently breaks DeepSeek V3/R1 family | ### System info
- `transformers`: 5.3.0
- `tokenizers`: 0.22.2
- Python: 3.12 / Linux
### Who can help?
@ArthurZucker
### Reproduction
Tokenizer-only, ~7 MB download:
```python
from transformers import AutoTokenizer
tok = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-V3")
print(repr(tok.decode(tok.encode("h... | open | null | false | 7 | [] | [] | 2026-04-17T08:50:48Z | 2026-05-15T19:17:25Z | null | NONE | null | 20260516T000042Z | 2026-05-16T00:00:42Z | dc3671 | 5,948,851 | MDQ6VXNlcjU5NDg4NTE= | User | false |
huggingface/transformers | 4,282,845,832 | I_kwDOCUB6oc7_RwqI | 45,491 | https://github.com/huggingface/transformers/issues/45491 | https://api.github.com/repos/huggingface/transformers/issues/45491 | [Gemma3] NaN embeddings on GPU when batching sequences of mixed length (sliding window attention + all-padding windows) | ### System Info
**System Info**
- `transformers`: 4.45.1
- `sentence-transformers`: 5.1.2
- `tokenizers`: 0.20.0
- `safetensors`: 0.4.5
- `PyTorch`: ≥ 2.6.0
- Serving runtime: `pytorch/torchserve-kfs:0.12.0` (Python 3.9, Linux x86_64)
- GPU: NVIDIA (CUDA, via KServe / TorchServe on Kubernetes)
- CPU inference: **not ... | closed | completed | false | 8 | [
"bug"
] | [] | 2026-04-17T13:10:25Z | 2026-04-21T10:37:02Z | 2026-04-21T10:37:02Z | NONE | null | 20260421T120044Z | 2026-04-21T12:00:44Z | RiccardoTOTI | 43,544,166 | MDQ6VXNlcjQzNTQ0MTY2 | User | false |
huggingface/transformers | 4,285,307,501 | I_kwDOCUB6oc7_bJpt | 45,496 | https://github.com/huggingface/transformers/issues/45496 | https://api.github.com/repos/huggingface/transformers/issues/45496 | Add V-JEPA 2.1 inference support | ### Feature request
Meta released [V-JEPA 2.1](https://github.com/facebookresearch/vjepa2) on 2026-03-16 with four pretrained video encoders at 384 resolution (ViT-B 80M, ViT-L 300M, ViT-g 1B, ViT-G 2B). The existing `vjepa2` model family in transformers supports V-JEPA 2.0 but not 2.1.
V-JEPA 2.1 introduces several ... | open | null | false | 0 | [
"Feature request"
] | [] | 2026-04-17T20:59:54Z | 2026-04-17T20:59:54Z | null | NONE | null | 20260417T210541Z | 2026-04-17T21:05:41Z | davevanveen | 25,591,765 | MDQ6VXNlcjI1NTkxNzY1 | User | false |
huggingface/transformers | 4,288,654,586 | I_kwDOCUB6oc7_n6z6 | 45,507 | https://github.com/huggingface/transformers/issues/45507 | https://api.github.com/repos/huggingface/transformers/issues/45507 | GraniteMoEHybrid Model Calls Invalid Method | ### System Info
Linux: Ubuntu 24.04.4 LTS / 6.8.0-107-generic-64k / aarch64
Python: 3.12.12
Transformers: 5.5.4
Cuda: 12.9
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such a... | open | null | false | 0 | [
"bug"
] | [] | 2026-04-18T17:07:36Z | 2026-04-18T17:10:09Z | null | NONE | null | 20260418T190539Z | 2026-04-18T19:05:39Z | rnowling | 1,114,888 | MDQ6VXNlcjExMTQ4ODg= | User | false |
huggingface/transformers | 4,291,972,999 | I_kwDOCUB6oc7_0k-H | 45,517 | https://github.com/huggingface/transformers/issues/45517 | https://api.github.com/repos/huggingface/transformers/issues/45517 | MPS OOM error, finetuning T5Gemma2 with Seq2SeqTrainer | ### System Info
MacOS M3 24gb ram Tahoe, MPS backend, transformers 5.5.4 (with #45516 applied), torch 2.10.0
### 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, ..... | closed | not_planned | false | 3 | [
"bug"
] | [] | 2026-04-19T20:47:00Z | 2026-04-20T19:54:57Z | 2026-04-20T19:54:57Z | NONE | null | 20260421T000044Z | 2026-04-21T00:00:44Z | Tokarak | 63,452,145 | MDQ6VXNlcjYzNDUyMTQ1 | User | false |
huggingface/transformers | 4,292,666,480 | I_kwDOCUB6oc7_3ORw | 45,518 | https://github.com/huggingface/transformers/issues/45518 | https://api.github.com/repos/huggingface/transformers/issues/45518 | Expose static_graph DDP flag via TrainingArguments | ### Feature request
Add a `ddp_static_graph: Optional[bool]` field to [`TrainingArguments`](https://github.com/huggingface/transformers/blob/a29df2d916e3b820aecd19d3b5a877abc523ba3c/src/transformers/training_args.py#L1370-L1377) (mirroring the existing `ddp_broadcast_buffers` pattern) and forward it through [`Trainer.... | closed | completed | false | 0 | [] | [] | 2026-04-20T02:00:45Z | 2026-04-21T12:46:41Z | 2026-04-21T12:46:41Z | CONTRIBUTOR | null | 20260421T180049Z | 2026-04-21T18:00:49Z | KeitaW | 8,693,216 | MDQ6VXNlcjg2OTMyMTY= | User | false |
huggingface/transformers | 3,713,335,476 | I_kwDOCUB6oc7dVQC0 | 42,757 | https://github.com/huggingface/transformers/issues/42757 | https://api.github.com/repos/huggingface/transformers/issues/42757 | cannot import name 'is_offline_mode' from 'huggingface_hub' | ### System Info
- transformers-5.0.0
- huggingface_hub-1.2.1
```
ImportError: cannot import name 'is_offline_mode' from 'huggingface_hub' (/root/miniconda3/envs/transformers/lib/python3.10/site-packages/huggingface_hub/__init__.py)
```
### Who can help?
_No response_
### Information
- [ ] The official example scri... | closed | completed | false | 2 | [
"bug"
] | [] | 2025-12-10T02:43:43Z | 2026-05-04T06:49:09Z | 2025-12-10T03:16:05Z | NONE | null | 20260504T180032Z | 2026-05-04T18:00:32Z | dollarser | 28,535,623 | MDQ6VXNlcjI4NTM1NjIz | User | false |
huggingface/transformers | 4,293,027,800 | I_kwDOCUB6oc7_4mfY | 45,520 | https://github.com/huggingface/transformers/issues/45520 | https://api.github.com/repos/huggingface/transformers/issues/45520 | KeyError: 'flash_attn' in import_utils.py when running on Python 3.13 | ### System Info
```
Using Python 3.13.12 environment at: ComfyUI-ROCm/.venv
Name: accelerate
Version: 1.13.0
Location: /home/laichiaheng/ComfyUI-ROCm/.venv/lib/python3.13/site-packages
Requires: huggingface-hub, numpy, packaging, psutil, pyyaml, safetensors, torch
Required-by: peft
---
Name: diffusers
Version: 0.37.1
... | open | null | false | 2 | [
"bug"
] | [] | 2026-04-20T04:08:22Z | 2026-04-25T22:37:44Z | null | NONE | null | 20260426T000021Z | 2026-04-26T00:00:21Z | laichiaheng | 9,532,105 | MDQ6VXNlcjk1MzIxMDU= | User | false |
huggingface/transformers | 4,293,135,092 | I_kwDOCUB6oc7_5Ar0 | 45,521 | https://github.com/huggingface/transformers/issues/45521 | https://api.github.com/repos/huggingface/transformers/issues/45521 | T5Gemma2: decoder self-attention fixed 4097-element mask at batch=1, fails on inputs >4094 tokens | ### System Info
- transformers `5.0.0` (T5Gemma 2 support shipped in #41834)
- torch `2.8.0`, CUDA `12.8.1`, Python `3.12`
- Hardware: 1× NVIDIA H100 NVL 94 GB (reproduced on same bug under A100 80 GB SXM)
- Model: `google/t5gemma-2-4b-4b` (gated)
- Base image: `runpod/pytorch:1.0.3-cu1281-torch280-ubuntu2404`
### Wh... | open | null | false | 7 | [
"Good Second Issue"
] | [] | 2026-04-20T04:44:55Z | 2026-04-26T04:37:24Z | null | NONE | null | 20260426T060015Z | 2026-04-26T06:00:15Z | arunkumarchithanar | 573,117 | MDQ6VXNlcjU3MzExNw== | User | false |
huggingface/transformers | 4,293,180,071 | I_kwDOCUB6oc7_5Lqn | 45,522 | https://github.com/huggingface/transformers/issues/45522 | https://api.github.com/repos/huggingface/transformers/issues/45522 | Feature request: Flash Attention 2 support for T5Gemma 2 | ### Feature request
Add Flash Attention 2 support for `T5Gemma2ForConditionalGeneration` (and companion variants: encoder, decoder, etc., wherever `attn_implementation="flash_attention_2"` currently raises).
Currently, loading the model with FA2 fails at dispatch time:
```python
from transformers import AutoModelFor... | closed | completed | false | 5 | [] | [] | 2026-04-20T04:57:54Z | 2026-05-15T13:00:48Z | 2026-05-15T13:00:22Z | NONE | null | 20260515T180026Z | 2026-05-15T18:00:26Z | arunkumarchithanar | 573,117 | MDQ6VXNlcjU3MzExNw== | User | false |
huggingface/transformers | 4,294,932,277 | I_kwDOCUB6oc7__3c1 | 45,529 | https://github.com/huggingface/transformers/issues/45529 | https://api.github.com/repos/huggingface/transformers/issues/45529 | Add `Olmo2ForSequenceClassification` (and ideally `OlmoForSequenceClassification` / `Olmo3ForSequenceClassification`) | `AutoModelForSequenceClassification.from_pretrained("allenai/OLMo-2-0425-1B")` currently fails because the OLMo family exposes only `*Model` and `*ForCausalLM`. All peer decoder architectures (Llama, Mistral, Qwen2, Gemma, Falcon, etc.) ship `ForSequenceClassification`.
## Motivation
I teach the graduate Applied Deep... | closed | completed | false | 2 | [] | [] | 2026-04-20T10:22:03Z | 2026-04-22T14:41:36Z | 2026-04-22T14:41:36Z | CONTRIBUTOR | null | 20260422T180025Z | 2026-04-22T18:00:25Z | earino | 3,258 | MDQ6VXNlcjMyNTg= | User | false |
huggingface/transformers | 4,299,297,967 | I_kwDOCUB6oc8AAAABAEIUrw | 45,536 | https://github.com/huggingface/transformers/issues/45536 | https://api.github.com/repos/huggingface/transformers/issues/45536 | Feature Request: Add SCAO Optimizer integration for 1.5x faster fine-tuning throughput | ### Feature request
Currently, AdamW is the default standard for fine-tuning via the Trainer class. While robust, its diagonal approximation of loss curvature makes early convergence slow, which is particularly expensive in compute-constrained environments or rapid fine-tuning pipelines (like LoRA/PEFT) where wall-clo... | closed | completed | false | 1 | [
"Feature request"
] | [] | 2026-04-20T23:31:39Z | 2026-04-21T12:21:07Z | 2026-04-21T12:21:07Z | NONE | null | 20260421T180049Z | 2026-04-21T18:00:49Z | whispering3 | 139,091,824 | U_kgDOCEpfcA | User | false |
huggingface/transformers | 4,299,470,795 | I_kwDOCUB6oc8AAAABAES3yw | 45,538 | https://github.com/huggingface/transformers/issues/45538 | https://api.github.com/repos/huggingface/transformers/issues/45538 | CLIPTokenizer uses 10**30 as `model_max_length` | ### System Info
```
transformers==5.5.4
python3.12
```
vs
```
transformers==4.57.6
python3.12
```
### 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... | open | null | false | 3 | [
"bug"
] | [] | 2026-04-21T00:19:55Z | 2026-04-21T06:57:34Z | null | NONE | null | 20260421T120044Z | 2026-04-21T12:00:44Z | D1-3105 | 65,292,437 | MDQ6VXNlcjY1MjkyNDM3 | User | false |
huggingface/transformers | 4,300,007,163 | I_kwDOCUB6oc8AAAABAEzm-w | 45,542 | https://github.com/huggingface/transformers/issues/45542 | https://api.github.com/repos/huggingface/transformers/issues/45542 | Only tensorboard is installed without TensorFlow, causing undefined tf backend error | ## System Info
- Docker Image: `verlai/verl:vllm018.dev1`
- Transformers version: **5.5.4** (manually updated)
- Python version: 3.12
- Dependencies: **TensorFlow is NOT installed**, only TensorBoard is installed
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [ ] My own mod... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-04-21T03:05:22Z | 2026-04-23T07:24:15Z | 2026-04-23T07:23:23Z | CONTRIBUTOR | null | 20260423T120024Z | 2026-04-23T12:00:24Z | enze5088 | 14,285,786 | MDQ6VXNlcjE0Mjg1Nzg2 | User | false |
huggingface/transformers | 4,301,312,789 | I_kwDOCUB6oc8AAAABAGDTFQ | 45,545 | https://github.com/huggingface/transformers/issues/45545 | https://api.github.com/repos/huggingface/transformers/issues/45545 | Transformers is trying to call home despite `local_files_only=True` | ### System Info
```log
Traceback (most recent call last):
File "/home/user/.local/lib/python3.12/site-packages/httpcore/_exceptions.py", line 10, in map_exceptions
yield
File "/home/user/.local/lib/python3.12/site-packages/httpcore/backends/sync.py", line 94, in connect_tcp
sock = socket.create_connection(... | closed | completed | false | 9 | [
"bug"
] | [] | 2026-04-21T08:29:26Z | 2026-04-22T18:09:18Z | 2026-04-22T17:33:26Z | NONE | null | 20260423T000027Z | 2026-04-23T00:00:27Z | Sur3 | 17,153,578 | MDQ6VXNlcjE3MTUzNTc4 | User | false |
huggingface/transformers | 4,305,393,654 | I_kwDOCUB6oc8AAAABAJ8X9g | 45,557 | https://github.com/huggingface/transformers/issues/45557 | https://api.github.com/repos/huggingface/transformers/issues/45557 | Pipeline num_workers runtime default is 0 but documentation states 8 | ## `num_workers` runtime default does not match documented default of `8`
### Description
The docstring for `num_workers` in `build_pipeline_init_args()` states a default of `8`, but the actual runtime fallback in `Pipeline.__call__()` is `0`.
### Steps to Reproduce
Inspect `src/transformers/pipelines/base.py`:
**... | closed | completed | false | 1 | [] | [] | 2026-04-21T21:30:21Z | 2026-04-23T11:34:26Z | 2026-04-23T11:34:26Z | NONE | null | 20260423T120024Z | 2026-04-23T12:00:24Z | Leonater | 83,777,478 | MDQ6VXNlcjgzNzc3NDc4 | User | false |
huggingface/transformers | 4,306,489,250 | I_kwDOCUB6oc8AAAABAK_Pog | 45,561 | https://github.com/huggingface/transformers/issues/45561 | https://api.github.com/repos/huggingface/transformers/issues/45561 | [Bug] pytest-xdist workers race on captured_info.txt in patched testing utils | ### System Info
Transformers=5.6.0.dev0 (local checkout 85099df959);
Python=3.13.5;
Platform=macOS-26.2-arm64-arm-64bit-Mach-O;
pytest=8.4.2;
pytest-xdist=3.8.0
### Who can help?
@ydshieh @SunMarc
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially ... | open | null | false | 2 | [
"bug"
] | [] | 2026-04-22T03:12:11Z | 2026-05-05T04:23:38Z | null | NONE | null | 20260505T060044Z | 2026-05-05T06:00:44Z | oleksii-tumanov | 6,143,578 | MDQ6VXNlcjYxNDM1Nzg= | User | false |
huggingface/transformers | 4,306,660,481 | I_kwDOCUB6oc8AAAABALJsgQ | 45,563 | https://github.com/huggingface/transformers/issues/45563 | https://api.github.com/repos/huggingface/transformers/issues/45563 | Paged generate() emits a stale warning for num_return_sequences | ### System Info
Transformers version: 5.6.0.dev0
Platform: macOS-26.2-arm64-arm-64bit-Mach-O
Python version: 3.13.5 (v3.13.5:6cb20a219a8, Jun 11 2025, 12:23:45) [Clang 16.0.0 (clang-1600.0.26.6)]
PyTorch version: 2.11.0
CUDA available: False
MPS available: True
### Who can help?
@cyrilvallez @remi-or
### Informatio... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-04-22T04:16:08Z | 2026-04-24T07:59:32Z | 2026-04-24T07:59:32Z | NONE | null | 20260424T120023Z | 2026-04-24T12:00:23Z | oleksii-tumanov | 6,143,578 | MDQ6VXNlcjYxNDM1Nzg= | User | false |
huggingface/transformers | 4,308,216,831 | I_kwDOCUB6oc8AAAABAMor_w | 45,571 | https://github.com/huggingface/transformers/issues/45571 | https://api.github.com/repos/huggingface/transformers/issues/45571 | A nice UX for generating dynamic tensors that break torch.compile/torch.export | ### Feature request
We have a couple tensors whose construction is highly dynamic and can't be captured while tracing (cu_seqlens, vision(image/video)_cu_seqlens, ... , vision(image/video)_position_ids, ...). I'm opening this issue to track what we are gonna do about them. The two options are either using the processo... | open | null | false | 3 | [
"Feature request"
] | [] | 2026-04-22T10:04:10Z | 2026-04-26T21:51:04Z | null | MEMBER | null | 20260427T000017Z | 2026-04-27T00:00:17Z | IlyasMoutawwakil | 57,442,720 | MDQ6VXNlcjU3NDQyNzIw | User | false |
huggingface/transformers | 4,310,708,891 | I_kwDOCUB6oc8AAAABAPAymw | 45,584 | https://github.com/huggingface/transformers/issues/45584 | https://api.github.com/repos/huggingface/transformers/issues/45584 | Whisper generation fails on empty transcription after align_special_tokens | ### System Info
- `transformers` version: 5.6.0.dev0
- Platform: Linux-6.17.0-1009-gcp-x86_64-with-glibc2.39
- Python version: 3.12.3
- Huggingface_hub version: 1.11.0
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accel... | open | null | false | 1 | [
"bug"
] | [] | 2026-04-22T17:23:03Z | 2026-05-07T17:02:24Z | null | NONE | null | 20260507T180032Z | 2026-05-07T18:00:32Z | ronansgd | 63,855,061 | MDQ6VXNlcjYzODU1MDYx | User | false |
huggingface/transformers | 1,708,804,231 | I_kwDOCUB6oc5l2kiH | 23,354 | https://github.com/huggingface/transformers/issues/23354 | https://api.github.com/repos/huggingface/transformers/issues/23354 | Make it easy to get seperate "prints" for individual runs/ users when using Transformers Agent | ### Feature request
I have started exploring the new Transformers Agent. And I would like to build a UI to help me speed up the process.
I might be running multiple runs in parallel or have multiple users using my application. I would like to be able to stream the information from the run as it arrives. I would l... | closed | completed | false | 5 | [
"Feature request"
] | [] | 2023-05-14T03:31:49Z | 2026-04-23T11:39:37Z | 2026-04-23T11:39:32Z | NONE | null | 20260423T120024Z | 2026-04-23T12:00:24Z | MarcSkovMadsen | 42,288,570 | MDQ6VXNlcjQyMjg4NTcw | User | false |
huggingface/transformers | 4,312,607,522 | I_kwDOCUB6oc8AAAABAQ0rIg | 45,588 | https://github.com/huggingface/transformers/issues/45588 | https://api.github.com/repos/huggingface/transformers/issues/45588 | `integrations/flash_attention.py` crashes with `AttributeError` on `s_aux=None` for sink-less models | ### System Info
- `transformers` version: 5.6.0
- Platform: Linux-6.8.0-1043-nvidia-x86_64-with-glibc2.35
- Python version: 3.12.13
- Huggingface_hub version: 1.11.0
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerat... | closed | completed | false | 8 | [
"bug"
] | [] | 2026-04-23T00:37:54Z | 2026-05-03T20:01:32Z | 2026-04-23T07:32:56Z | CONTRIBUTOR | null | 20260504T000033Z | 2026-05-04T00:00:33Z | jamesbraza | 8,990,777 | MDQ6VXNlcjg5OTA3Nzc= | User | false |
huggingface/transformers | 4,313,864,648 | I_kwDOCUB6oc8AAAABASBZyA | 45,593 | https://github.com/huggingface/transformers/issues/45593 | https://api.github.com/repos/huggingface/transformers/issues/45593 | D-FINE not using any auxiliary losses when denoising is turned off | ### System Info
- `transformers` version: 5.7.0.dev0
- Platform: Linux-6.6.87.2-microsoft-standard-WSL2-x86_64-with-glibc2.39
- Python version: 3.12.3
- Huggingface_hub version: 1.7.1
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorc... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-04-23T06:01:05Z | 2026-04-23T20:07:05Z | 2026-04-23T20:07:05Z | CONTRIBUTOR | null | 20260424T000039Z | 2026-04-24T00:00:39Z | m-matthias | 16,415,097 | MDQ6VXNlcjE2NDE1MDk3 | User | false |
huggingface/transformers | 2,252,063,300 | I_kwDOCUB6oc6GO8JE | 30,333 | https://github.com/huggingface/transformers/issues/30333 | https://api.github.com/repos/huggingface/transformers/issues/30333 | If a training job job failed MLFlow will not be reported and MLFlow shows job still running | ### System Info
- `transformers` version: 4.40.0
- Platform: Linux-6.8.6-arch1-1-x86_64-with-glibc2.39
- Python version: 3.11.7
- Huggingface_hub version: 0.20.3
- Safetensors version: 0.4.2
- Accelerate version: 0.27.0
- PyTorch version (GP?): 2.2.2+rocm5.7 (True)
- Jax version: not installed
- JaxLib versi... | open | null | false | 11 | [
"trainer",
"Good Second Issue",
"Integrations"
] | [] | 2024-04-19T04:31:41Z | 2026-05-05T22:06:43Z | null | CONTRIBUTOR | null | 20260506T000041Z | 2026-05-06T00:00:41Z | helloworld1 | 247,316 | MDQ6VXNlcjI0NzMxNg== | User | false |
huggingface/transformers | 4,315,067,067 | I_kwDOCUB6oc8AAAABATKyuw | 45,600 | https://github.com/huggingface/transformers/issues/45600 | https://api.github.com/repos/huggingface/transformers/issues/45600 | auto_mappings.py references removed Sam3LiteText configs, breaking CI | ### System Info
- `transformers` version: 5.6.0.dev0
- Platform: Linux-5.14.0-570.12.1.el9_6.x86_64-x86_64-with-glibc2.34
- Python version: 3.11.15
- Huggingface_hub version: 1.10.2
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch ve... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-04-23T09:51:06Z | 2026-04-23T11:19:27Z | 2026-04-23T11:19:27Z | NONE | null | 20260423T120024Z | 2026-04-23T12:00:24Z | artem-spector | 10,895,869 | MDQ6VXNlcjEwODk1ODY5 | User | false |
huggingface/transformers | 4,323,449,470 | I_kwDOCUB6oc8AAAABAbKafg | 45,632 | https://github.com/huggingface/transformers/issues/45632 | https://api.github.com/repos/huggingface/transformers/issues/45632 | `trust_remote_code` cache path collides for local models sharing a leaf directory name | ### System Info
- transformers: 5.5.3
- huggingface_hub: 1.12.0
- Python: 3.13
- OS: Linux
### Who can help?
_No response_
### 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 ... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-04-24T14:04:02Z | 2026-04-29T11:25:49Z | 2026-04-29T11:25:49Z | CONTRIBUTOR | null | 20260429T120025Z | 2026-04-29T12:00:25Z | nurpax | 297,823 | MDQ6VXNlcjI5NzgyMw== | User | false |
huggingface/transformers | 3,448,019,756 | I_kwDOCUB6oc7NhJss | 41,115 | https://github.com/huggingface/transformers/issues/41115 | https://api.github.com/repos/huggingface/transformers/issues/41115 | Add Model Architecture for MiniCPM3 | ### Model description
This is a feature request to add a new model architecture for MiniCPM3, a powerful, lightweight language model.
MiniCPM3 is the third generation of the MiniCPM series. It demonstrates performance comparable to or exceeding many 7B-9B models, despite its smaller size. It excels in both English an... | open | null | false | 1 | [
"New model"
] | [] | 2025-09-24T07:07:47Z | 2026-04-24T14:31:38Z | null | CONTRIBUTOR | null | 20260424T180025Z | 2026-04-24T18:00:25Z | bzantium | 19,511,788 | MDQ6VXNlcjE5NTExNzg4 | User | false |
huggingface/transformers | 4,323,748,552 | I_kwDOCUB6oc8AAAABAbcqyA | 45,636 | https://github.com/huggingface/transformers/issues/45636 | https://api.github.com/repos/huggingface/transformers/issues/45636 | Proposal: add sdpa_memeff attn_implementation for shape combinations no fast backend covers | ## Summary
Proposal to add a new `attn_implementation="sdpa_memeff"` that pins torch's SDPA dispatcher to `SDPBackend.EFFICIENT_ATTENTION` (via `sdpa_kernel([EFFICIENT_ATTENTION])` wrapping the existing `sdpa_attention_forward`). Filing as an issue to validate design direction before opening a PR.
## Motivation — two... | open | reopened | false | 6 | [] | [] | 2026-04-24T14:56:28Z | 2026-05-04T14:58:16Z | null | NONE | null | 20260504T180032Z | 2026-05-04T18:00:32Z | dvdimitrov13 | 60,075,474 | MDQ6VXNlcjYwMDc1NDc0 | User | false |
huggingface/transformers | 4,326,688,288 | I_kwDOCUB6oc8AAAABAeQGIA | 45,644 | https://github.com/huggingface/transformers/issues/45644 | https://api.github.com/repos/huggingface/transformers/issues/45644 | mps: test_eager_matches_sdpa_inference tests fail with PyTorch MPS backend | ### System Info
- `transformers`: 5.7.0.dev0 (main, c472755e79)
- macOS-26.1-arm64, Apple M5, torch 2.11.0 (MPS)
- Python 3.12.13
### Who can help?
@Cyrilvallez (MPS counterpart to the XPU branch you added).
### Reproduction
```
TRANSFORMERS_TEST_DEVICE=mps python -m pytest tests/models/{llama,gemma,qwen2,mistral}/t... | closed | completed | false | 1 | [] | [] | 2026-04-25T02:09:42Z | 2026-04-27T07:47:25Z | 2026-04-27T07:47:24Z | NONE | null | 20260427T120026Z | 2026-04-27T12:00:26Z | qflen | 194,738,340 | U_kgDOC5t4pA | User | false |
huggingface/transformers | 4,327,838,929 | I_kwDOCUB6oc8AAAABAfWU0Q | 45,646 | https://github.com/huggingface/transformers/issues/45646 | https://api.github.com/repos/huggingface/transformers/issues/45646 | NLP | null | closed | completed | false | 1 | [] | [] | 2026-04-25T10:34:05Z | 2026-04-27T09:05:10Z | 2026-04-27T09:05:10Z | NONE | null | 20260427T120026Z | 2026-04-27T12:00:26Z | mariam12-dotcom | 239,433,683 | U_kgDODkV30w | User | false |
huggingface/transformers | 4,328,968,742 | I_kwDOCUB6oc8AAAABAgbSJg | 45,647 | https://github.com/huggingface/transformers/issues/45647 | https://api.github.com/repos/huggingface/transformers/issues/45647 | MusicgenMelody ignores audio conditioning (regression between 4.48 and 4.57) | _(Original Note): Claude removed this line when editing, but I wanted to fully disclose that this issue was discovered and written up by Claude code_
**Update (corrected):** the regression is wider than the title suggests — it already exists in transformers **4.57.6**, the latest 4.x. So this is not a v5 regression; i... | open | null | false | 12 | [] | [] | 2026-04-25T18:58:41Z | 2026-05-01T17:10:32Z | null | NONE | null | 20260501T180051Z | 2026-05-01T18:00:51Z | audiodude | 57,832 | MDQ6VXNlcjU3ODMy | User | false |
huggingface/transformers | 4,334,969,822 | I_kwDOCUB6oc8AAAABAmJj3g | 45,656 | https://github.com/huggingface/transformers/issues/45656 | https://api.github.com/repos/huggingface/transformers/issues/45656 | Optimizer step being called 2 times when using deepspeed | ### System Info
In version transformers==4.57.3, and deepspeed==0.18.3,
in below screenshot, when accelerator.backward is called, the deepspeed backward internally calls engine.step which is performing optimizer step at gradient accumulation step
The below snapshot is from trainer.py in transformers library
<img w... | open | null | false | 0 | [
"bug"
] | [] | 2026-04-27T10:22:39Z | 2026-04-27T10:23:25Z | null | NONE | null | 20260427T120026Z | 2026-04-27T12:00:26Z | harsh2912 | 18,512,791 | MDQ6VXNlcjE4NTEyNzkx | User | false |
huggingface/transformers | 4,338,826,538 | I_kwDOCUB6oc8AAAABAp09Kg | 45,663 | https://github.com/huggingface/transformers/issues/45663 | https://api.github.com/repos/huggingface/transformers/issues/45663 | Gemma-4 training with FSDP2 raises `KeyError` in `Gemma4TextAttention.forward` because `shared_kv_states` is rebuilt per-layer | ### System Info
- `transformers` version: 5.6.2
- Platform: Linux-6.8.0-1043-nvidia-x86_64-with-glibc2.35
- Python version: 3.12.13
- Huggingface_hub version: 1.11.0
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerat... | closed | completed | false | 3 | [
"bug"
] | [] | 2026-04-27T20:55:50Z | 2026-05-14T05:58:27Z | 2026-05-14T05:58:27Z | CONTRIBUTOR | null | 20260514T060044Z | 2026-05-14T06:00:44Z | jamesbraza | 8,990,777 | MDQ6VXNlcjg5OTA3Nzc= | User | false |
huggingface/transformers | 4,341,660,362 | I_kwDOCUB6oc8AAAABAsh6yg | 45,674 | https://github.com/huggingface/transformers/issues/45674 | https://api.github.com/repos/huggingface/transformers/issues/45674 | [BitsAndBytesConfig] Providing llm_int8_skip_modules clears the default lm_head exclusion, causing AssertionError in 4-bit inference | ## Environment
- `transformers`: 5.5.4
- `bitsandbytes`: 0.49.2
- `torch`: 2.11.0+cu126
- CUDA: 12.6
- OS: Windows 11
- GPU: NVIDIA RTX 3090
## Bug Description
When specifying `llm_int8_skip_modules` in `BitsAndBytesConfig`, the default module exclusion list (which normally protects `lm_head` from being quantized) i... | open | null | false | 5 | [] | [] | 2026-04-28T08:27:12Z | 2026-04-28T13:03:14Z | null | NONE | null | 20260428T180033Z | 2026-04-28T18:00:33Z | softguy777 | 145,181,514 | U_kgDOCKdLSg | User | false |
huggingface/transformers | 4,341,641,436 | I_kwDOCUB6oc8AAAABAsgw3A | 45,672 | https://github.com/huggingface/transformers/issues/45672 | https://api.github.com/repos/huggingface/transformers/issues/45672 | [Gemma4] torch.finfo() TypeError on uint8 weights in audio modules during 4-bit (NF4) inference | ## Environment
- `transformers`: 5.5.4
- `bitsandbytes`: 0.49.2
- `torch`: 2.11.0+cu126
- CUDA: 12.6
- OS: Windows 11
- GPU: NVIDIA RTX 3090
## Bug Description
When running `google/gemma-4-e2b-it` with `BitsAndBytesConfig(load_in_4bit=True)`, the forward pass crashes immediately with:
```
TypeError: torch.finfo() r... | open | null | false | 1 | [] | [] | 2026-04-28T08:23:48Z | 2026-04-28T11:32:26Z | null | NONE | null | 20260428T120019Z | 2026-04-28T12:00:19Z | softguy777 | 145,181,514 | U_kgDOCKdLSg | User | false |
huggingface/transformers | 4,336,103,397 | I_kwDOCUB6oc8AAAABAnOv5Q | 45,657 | https://github.com/huggingface/transformers/issues/45657 | https://api.github.com/repos/huggingface/transformers/issues/45657 | ValueError in zero_shot_object_detection.md doctest on Python 3.13 | While running the test suite on Python 3.13.7, pytest fails to collect tests due to a malformed doctest in docs/source/en/tasks/zero_shot_object_detection.md.
Environment:
Python: 3.13.7
Platform: macOS (darwin)
transformers: main branch
Error:
ValueError: line 172 of the docstring for zero_shot_object_detection.md
... | closed | completed | false | 0 | [] | [] | 2026-04-27T13:28:53Z | 2026-04-28T11:40:48Z | 2026-04-28T11:40:48Z | CONTRIBUTOR | null | 20260428T120019Z | 2026-04-28T12:00:19Z | AnkitAhlawat7742 | 199,906,670 | U_kgDOC-pVbg | User | false |
huggingface/transformers | 4,341,737,188 | I_kwDOCUB6oc8AAAABAsmm5A | 45,676 | https://github.com/huggingface/transformers/issues/45676 | https://api.github.com/repos/huggingface/transformers/issues/45676 | Gemma 4: Exploding pre-clip gradient norms during LoRA fine-tuning of `gemma-4-31B-it` | ### System Info
## Summary
Fine-tuning `google/gemma-4-31B-it` with a small LoRA via standard `transformers.Trainer` + `peft` on a public chat-style dataset produces pre-clip gradient norms that are **1–3 orders of magnitude larger than expected**. With `max_grad_norm=1.0` the actual updates are bounded, but the pre-... | closed | completed | false | 9 | [
"Good Second Issue",
"bug"
] | [] | 2026-04-28T08:39:34Z | 2026-05-02T15:07:30Z | 2026-05-02T15:07:30Z | NONE | null | 20260502T180031Z | 2026-05-02T18:00:31Z | pritmish | 183,807,581 | U_kgDOCvSuXQ | User | false |
huggingface/transformers | 4,343,386,640 | I_kwDOCUB6oc8AAAABAuLSEA | 45,685 | https://github.com/huggingface/transformers/issues/45685 | https://api.github.com/repos/huggingface/transformers/issues/45685 | [moe] mps interface has error "histogram_mps" not implemented for 'Int' | ### System Info
# Transformers env info
Python version: 3.13.9
os system: macOS-26.4.1-arm64-arm-64bit-Mach-O
PyTorch version: 2.11.0
Transformers version: 5.6.2
CUDA : False
MPS: True
### Who can help?
@cyrilvallez in [moe.py](https://github.com/huggingface/transformers/blob/main/src/transformers/integrations/... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-04-28T13:08:21Z | 2026-05-05T08:28:03Z | 2026-05-05T08:28:03Z | NONE | null | 20260505T120036Z | 2026-05-05T12:00:36Z | chenzhe1204 | 38,660,669 | MDQ6VXNlcjM4NjYwNjY5 | User | false |
huggingface/transformers | 4,343,349,447 | I_kwDOCUB6oc8AAAABAuJAxw | 45,684 | https://github.com/huggingface/transformers/issues/45684 | https://api.github.com/repos/huggingface/transformers/issues/45684 | save_pretrained` (with `register_for_auto_class`) propagates read-only permissions from custom-model source files | ### System Info
- `transformers` version: 5.5.3
- Python: 3.13
- Platform: Linux
### Who can help?
@Cyrilvallez (model loading)
### 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... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-04-28T13:02:46Z | 2026-04-29T11:03:04Z | 2026-04-29T11:03:04Z | CONTRIBUTOR | null | 20260429T120025Z | 2026-04-29T12:00:25Z | nurpax | 297,823 | MDQ6VXNlcjI5NzgyMw== | User | false |
huggingface/transformers | 4,348,001,591 | I_kwDOCUB6oc8AAAABAyk9Nw | 45,693 | https://github.com/huggingface/transformers/issues/45693 | https://api.github.com/repos/huggingface/transformers/issues/45693 | Why the calculation of train_batch_size unrelated to split_batches | In the calculation of `train_batch_size` property in transformers/training_args.py, the formula used is `train_batch_size = self.per_device_train_batch_size * max(1, self.n_gpu)`. When `split_batches` is set to `False`, this is easy to understand: the number of samples on each GPU multiplied by the number of GPUs equa... | open | null | false | 2 | [] | [] | 2026-04-29T04:57:23Z | 2026-04-29T08:19:16Z | null | NONE | null | 20260429T120025Z | 2026-04-29T12:00:25Z | mklpr | 7,464,549 | MDQ6VXNlcjc0NjQ1NDk= | User | false |
huggingface/transformers | 4,348,919,946 | I_kwDOCUB6oc8AAAABAzdAig | 45,696 | https://github.com/huggingface/transformers/issues/45696 | https://api.github.com/repos/huggingface/transformers/issues/45696 | Improving CLI Serving Code Structure with Class-Based FastAPI Patterns | ### Feature request
This proposal suggests refactoring the CLI serving code (`transformers/cli/serving/server.py`) from a function-based to a class-based architecture, using a pattern that better organizes related endpoints, clarifies resource lifecycle management, and improves long-term maintainability.
### Motivat... | closed | completed | false | 1 | [
"Feature request",
"Code agent slop"
] | [] | 2026-04-29T08:16:12Z | 2026-04-29T10:12:24Z | 2026-04-29T10:11:53Z | NONE | null | 20260429T120025Z | 2026-04-29T12:00:25Z | HeHongyeFY | 44,426,557 | MDQ6VXNlcjQ0NDI2NTU3 | User | false |
huggingface/transformers | 4,348,995,412 | I_kwDOCUB6oc8AAAABAzhnVA | 45,698 | https://github.com/huggingface/transformers/issues/45698 | https://api.github.com/repos/huggingface/transformers/issues/45698 | from_pretrained loads wrong custom module after save_pretrained | ### System Info
Transformers version: 5.7.0.dev0
Python version: 3.13.3
Platform: Linux-6.17.0-20-generic-x86_64-with-glibc2.39
Machine: x86_64
Note, running `transformers env` fails with: `NameError: name 'CompletionCreateParamsStreaming' is not defined`
### Who can help?
@CyrilVallez (model loading)
### Informat... | open | null | false | 11 | [
"bug"
] | [] | 2026-04-29T08:29:41Z | 2026-05-06T19:43:50Z | null | CONTRIBUTOR | null | 20260507T000029Z | 2026-05-07T00:00:29Z | nurpax | 297,823 | MDQ6VXNlcjI5NzgyMw== | User | false |
huggingface/transformers | 4,350,467,219 | I_kwDOCUB6oc8AAAABA07ckw | 45,701 | https://github.com/huggingface/transformers/issues/45701 | https://api.github.com/repos/huggingface/transformers/issues/45701 | transformers version changes the tokenization | ### System Info
```
Python: 3.12.3
transformers: 5.7.0
torch : 2.11.0+cu130
sentencepiece: 0.2.1
```
________________
- **Platform:** Linux-6.17.0-22-generic-x86_64-with-glibc2.39
- **Python version:** 3.12.3
- **Huggingface_hub version:** 0.36.2
- **Safetensors version:** 0.7.0
- **Accelerate version:** 1.12.... | open | null | false | 3 | [
"bug"
] | [] | 2026-04-29T12:19:16Z | 2026-05-13T03:23:55Z | null | NONE | null | 20260513T060025Z | 2026-05-13T06:00:25Z | PiRom1 | 119,457,355 | U_kgDOBx7GSw | User | false |
huggingface/transformers | 4,352,722,515 | I_kwDOCUB6oc8AAAABA3FGUw | 45,704 | https://github.com/huggingface/transformers/issues/45704 | https://api.github.com/repos/huggingface/transformers/issues/45704 | T5 silently uses apex.FusedRMSNorm which has a memory leak (NVIDIA/apex#1999) | ### System Info
- `transformers` version: 4.57.6
- Platform: Linux-6.8.0-100-generic-x86_64-with-glibc2.39
- Python version: 3.12.3
- Huggingface_hub version: 0.36.2
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: 0.18.9
- PyTorch version (accelerator?):... | closed | completed | false | 0 | [
"bug"
] | [] | 2026-04-29T18:16:58Z | 2026-05-01T12:05:42Z | 2026-05-01T12:05:42Z | NONE | null | 20260501T180051Z | 2026-05-01T18:00:51Z | dustnehowl | 39,877,181 | MDQ6VXNlcjM5ODc3MTgx | User | false |
huggingface/transformers | 4,354,557,717 | I_kwDOCUB6oc8AAAABA41HFQ | 45,706 | https://github.com/huggingface/transformers/issues/45706 | https://api.github.com/repos/huggingface/transformers/issues/45706 | Hive Civilization — x402-native services for transformers Agents (notification) | Notification post — Hive Civilization runs 52 x402-wired services on Base mainnet (treasury 0x15184bf50b3d3f52b60434f8942b7d52f2eb436e, USDC settlement, x402 spec from coinbase/x402).
Why it might be relevant to transformers/Agents:
- transformers Agents tools can wrap Hive services (evaluator, classifier, summarizer... | closed | completed | false | 0 | [] | [] | 2026-04-30T00:41:28Z | 2026-04-30T10:18:32Z | 2026-04-30T10:18:32Z | NONE | null | 20260430T120024Z | 2026-04-30T12:00:24Z | srotzin | 140,019,476 | U_kgDOCFiHFA | User | false |
huggingface/transformers | 4,355,734,326 | I_kwDOCUB6oc8AAAABA587Ng | 45,710 | https://github.com/huggingface/transformers/issues/45710 | https://api.github.com/repos/huggingface/transformers/issues/45710 | Fix wrong repo link in Dinov2ForImageClassification doc example | The link to the config.json for google/dinov2-base-patch16-224 in the Dinov2ForImageClassification example is returning a 404. It should be updated to the correct path. | closed | completed | false | 1 | [] | [] | 2026-04-30T06:10:00Z | 2026-05-01T04:10:22Z | 2026-05-01T04:10:22Z | NONE | null | 20260501T113108Z | 2026-05-01T11:31:08Z | Milan-Bhimani | 157,954,157 | U_kgDOCWowbQ | User | false |
huggingface/transformers | 4,356,818,596 | I_kwDOCUB6oc8AAAABA6_GpA | 45,715 | https://github.com/huggingface/transformers/issues/45715 | https://api.github.com/repos/huggingface/transformers/issues/45715 | PreTrainedTokenizer.convert_ids_to_tokens(skip_special_tokens=True) rebuilds all_special_ids on every iteration of the per-id loop | ### System Info
- transformers version: 5.3.0 (also reproduces on `main`)
- Python: 3.12
- OS: Linux
- Affected class: `transformers.tokenization_python.PreTrainedTokenizer` (renamed to `PythonBackend` on `main`) — the slow-tokenizer base class
- **Not** affected: `TokenizersBackend` (the fast subclass) — it already h... | closed | completed | false | 2 | [] | [] | 2026-04-30T09:08:07Z | 2026-05-04T00:50:43Z | 2026-05-04T00:50:43Z | NONE | null | 20260504T060033Z | 2026-05-04T06:00:33Z | longlee0622 | 6,110,159 | MDQ6VXNlcjYxMTAxNTk= | User | false |
huggingface/transformers | 4,356,306,059 | I_kwDOCUB6oc8AAAABA6f0iw | 45,712 | https://github.com/huggingface/transformers/issues/45712 | https://api.github.com/repos/huggingface/transformers/issues/45712 | Six leftover dummy classes in `dummy_pt_objects.py` fail `check_repo.py` and leak into `dir(transformers)` without torch | ### System Info
transformers 5.7.0.dev0 (main).
`utils/check_repo.py` raises on a fresh install without torch:
```
Exception: The following objects are in the public init, but not in the docs:
- BeamScorer
- ConstrainedBeamSearchScorer
- Constraint
- ConstraintListState
- DisjunctiveConstraint
- PhrasalConstra... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-04-30T07:47:24Z | 2026-04-30T14:01:31Z | 2026-04-30T14:01:31Z | CONTRIBUTOR | null | 20260501T113108Z | 2026-05-01T11:31:08Z | jw9603 | 70,795,645 | MDQ6VXNlcjcwNzk1NjQ1 | User | false |
huggingface/transformers | 4,359,518,721 | I_kwDOCUB6oc8AAAABA9j6AQ | 45,727 | https://github.com/huggingface/transformers/issues/45727 | https://api.github.com/repos/huggingface/transformers/issues/45727 | fix(generation): correct spelling mistake in continuous_api docstring | Hi team,
While reviewing the continuous batching generation logic for a local deployment, I noticed a minor spelling mistake (`usefull` -> `useful`) in the docstring of `continuous_api.py`.
It's a tiny detail, but keeping the core API docs pristine is always good. Here is the patch:
```diff
--- a/src/transformers/g... | closed | completed | false | 3 | [] | [] | 2026-04-30T16:03:07Z | 2026-05-08T11:14:13Z | 2026-05-08T11:14:13Z | NONE | null | 20260508T120022Z | 2026-05-08T12:00:22Z | lohdptm | 280,717,817 | U_kgDOELtp-Q | User | false |
huggingface/transformers | 4,358,160,543 | I_kwDOCUB6oc8AAAABA8RAnw | 45,721 | https://github.com/huggingface/transformers/issues/45721 | https://api.github.com/repos/huggingface/transformers/issues/45721 | Pass library_name="transformers" to HfApi/Hub functions in push paths so commits attribute correctly | Per the repo's agentic contribution policy, opening this issue first to coordinate before sending the PR.
### Summary
`transformers` downloads correctly report a User-Agent of `transformers/<version>; python/...; session_id/...` (built by `http_user_agent()` in `src/transformers/utils/hub.py`). Pushes do not — every ... | open | null | false | 2 | [] | [] | 2026-04-30T12:40:51Z | 2026-04-30T19:24:43Z | null | MEMBER | null | 20260501T113108Z | 2026-05-01T11:31:08Z | davanstrien | 8,995,957 | MDQ6VXNlcjg5OTU5NTc= | User | false |
huggingface/transformers | 2,566,242,682 | I_kwDOCUB6oc6Y9cF6 | 33,945 | https://github.com/huggingface/transformers/issues/33945 | https://api.github.com/repos/huggingface/transformers/issues/33945 | Automatic dynamic batch size selection for DataCollatorWithFlattening | ### Feature request
Add a custom (batch index) sampler to automatically determine batch size to a fixed target number of tokens.
### Motivation
I'm keen to try out DataCollatorWithFlattening but unsure about how to set batch size, since no padding will be added so the total number of tokens is dynamic.
Im a... | closed | completed | false | 9 | [
"Usage",
"Feature request"
] | [] | 2024-10-04T12:19:46Z | 2026-05-16T02:04:30Z | 2026-05-01T12:27:32Z | NONE | null | 20260516T060035Z | 2026-05-16T06:00:35Z | alex-hh | 5,719,745 | MDQ6VXNlcjU3MTk3NDU= | User | false |
huggingface/transformers | 4,184,237,048 | I_kwDOCUB6oc75ZmP4 | 45,160 | https://github.com/huggingface/transformers/issues/45160 | https://api.github.com/repos/huggingface/transformers/issues/45160 | Add AEO quality badge to README | Hi! I'm behind [Clarvia](https://clarvia.art) — an open quality scoring platform for AI tools and MCP servers.
**Hugging Face Transformers** is indexed on Clarvia. Embed a live AEO (Agent Experience Optimization) badge in your README:
```markdown
[](https://clar... | closed | completed | false | 1 | [] | [] | 2026-04-01T05:29:50Z | 2026-05-10T08:34:07Z | 2026-05-10T08:34:07Z | NONE | null | 20260510T120026Z | 2026-05-10T12:00:26Z | digitamaz | 198,436,502 | U_kgDOC9Pmlg | User | false |
huggingface/transformers | 4,363,998,626 | I_kwDOCUB6oc8AAAABBB1Vog | 45,735 | https://github.com/huggingface/transformers/issues/45735 | https://api.github.com/repos/huggingface/transformers/issues/45735 | Bug with detecting cache positions in sdpa_mask | ### System Info
Transformers v5.7.0
### Who can help?
@ArthurZucker @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)
### ... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-05-01T10:42:56Z | 2026-05-11T06:23:13Z | 2026-05-11T06:23:13Z | CONTRIBUTOR | null | 20260511T120023Z | 2026-05-11T12:00:23Z | davidmezzetti | 561,939 | MDQ6VXNlcjU2MTkzOQ== | User | false |
huggingface/transformers | 4,364,318,314 | I_kwDOCUB6oc8AAAABBCI2ag | 45,736 | https://github.com/huggingface/transformers/issues/45736 | https://api.github.com/repos/huggingface/transformers/issues/45736 | Please update `tokenizers` version check | ### Feature request
Several days before, `tokenizers` released a new version of `0.23.1`, but `transformers` doesn't allow us to use it.
https://github.com/huggingface/transformers/blob/ecc3d0da0f8e9c2c54676345b816db29f842792a/setup.py#L150
https://github.com/huggingface/transformers/blob/ecc3d0da0f8e9c2c54676345b816... | open | null | false | 2 | [
"Feature request"
] | [] | 2026-05-01T12:15:45Z | 2026-05-02T06:40:40Z | null | NONE | null | 20260502T120033Z | 2026-05-02T12:00:33Z | lalala-233 | 79,189,174 | MDQ6VXNlcjc5MTg5MTc0 | User | false |
huggingface/transformers | 4,371,167,819 | I_kwDOCUB6oc8AAAABBIq6Sw | 45,750 | https://github.com/huggingface/transformers/issues/45750 | https://api.github.com/repos/huggingface/transformers/issues/45750 | `Qwen3VLVisionPatchEmbed.proj` (`nn.Conv3d` with `stride == kernel`) is ~50,000× slower than equivalent `nn.Linear` on Blackwell + bf16 | ### System Info
```
transformers version: 5.0.0.dev0
PyTorch: 2.9.0+cu128
CUDA: 12.8
cuDNN: 9.10.0.2 (91002)
Python: 3.14.0
flash-attn: 2.8.3 (installed)
GPU: NVIDIA GeForce RTX 5090 (Blackwell, compute capability 12.0, sm_120)
OS: ... | open | null | false | 4 | [
"bug"
] | [] | 2026-05-03T07:27:10Z | 2026-05-13T16:47:11Z | null | NONE | null | 20260513T180036Z | 2026-05-13T18:00:36Z | WangYuHang-cmd | 74,978,107 | MDQ6VXNlcjc0OTc4MTA3 | User | false |
huggingface/transformers | 4,371,965,219 | I_kwDOCUB6oc8AAAABBJblIw | 45,753 | https://github.com/huggingface/transformers/issues/45753 | https://api.github.com/repos/huggingface/transformers/issues/45753 | Qwen3_5 goes into infinite loop for a specific image | ### System Info
Colab T4 GPU
### Who can help?
I tried to run the colab notebooks for qwen3_5 models which are auto-generated. The inference takes forever. I figured out this issue is due to the specific image that is automatically inserted => "https://huggingface.co/datasets/huggingface/documentation-images/resolve... | open | null | false | 7 | [
"bug"
] | [] | 2026-05-03T13:47:50Z | 2026-05-13T15:22:02Z | null | CONTRIBUTOR | null | 20260513T180036Z | 2026-05-13T18:00:36Z | MHRDYN7 | 113,298,714 | U_kgDOBsDNGg | User | false |
huggingface/transformers | 4,372,666,795 | I_kwDOCUB6oc8AAAABBKGZqw | 45,758 | https://github.com/huggingface/transformers/issues/45758 | https://api.github.com/repos/huggingface/transformers/issues/45758 | DeepSeek-V4 CSA eager path may not preserve per-query top-k masking for S > 1 | ### System Info
Observed on `main` after DeepSeek-V4 support was added in #45643.
Relevant file:
- `src/transformers/models/deepseek_v4/modeling_deepseek_v4.py`
- `DeepseekV4CSACompressor.forward`
- `DeepseekV4Attention.forward`
### Who can help?
@ArthurZucker
### Information
- [ ] The official example scripts
-... | closed | completed | false | 3 | [] | [] | 2026-05-03T18:52:35Z | 2026-05-12T09:27:50Z | 2026-05-12T09:27:49Z | NONE | null | 20260512T120027Z | 2026-05-12T12:00:27Z | kekmodel | 7,728,527 | MDQ6VXNlcjc3Mjg1Mjc= | User | false |
huggingface/transformers | 4,372,671,381 | I_kwDOCUB6oc8AAAABBKGrlQ | 45,759 | https://github.com/huggingface/transformers/issues/45759 | https://api.github.com/repos/huggingface/transformers/issues/45759 | `AutoModelForCausalLM.from_config` does not unwrap `text_config` for composite Qwen 3.5 and 3.6 multimodal configs | ### System Info
- `transformers` version: 5.6.2
- Platform: Linux-6.8.0-1043-nvidia-x86_64-with-glibc2.35
- Python version: 3.12.13
- Huggingface_hub version: 1.13.0
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerato... | closed | completed | false | 0 | [
"bug"
] | [] | 2026-05-03T18:54:41Z | 2026-05-05T11:23:45Z | 2026-05-05T11:23:45Z | CONTRIBUTOR | null | 20260505T120036Z | 2026-05-05T12:00:36Z | jamesbraza | 8,990,777 | MDQ6VXNlcjg5OTA3Nzc= | User | false |
huggingface/transformers | 4,372,754,066 | I_kwDOCUB6oc8AAAABBKLukg | 45,761 | https://github.com/huggingface/transformers/issues/45761 | https://api.github.com/repos/huggingface/transformers/issues/45761 | Veneto | null | closed | completed | false | 0 | [] | [] | 2026-05-03T19:32:27Z | 2026-05-04T08:46:50Z | 2026-05-04T08:46:50Z | NONE | null | 20260504T180032Z | 2026-05-04T18:00:32Z | mirko772 | 207,275,767 | U_kgDODFrG9w | User | false |
huggingface/transformers | 4,372,756,390 | I_kwDOCUB6oc8AAAABBKL3pg | 45,762 | https://github.com/huggingface/transformers/issues/45762 | https://api.github.com/repos/huggingface/transformers/issues/45762 | Mirko Privitera 30-09-1990 | null | closed | completed | false | 0 | [] | [] | 2026-05-03T19:33:28Z | 2026-05-04T08:46:55Z | 2026-05-04T08:46:55Z | NONE | null | 20260504T180032Z | 2026-05-04T18:00:32Z | mirko772 | 207,275,767 | U_kgDODFrG9w | User | false |
huggingface/transformers | 4,372,761,179 | I_kwDOCUB6oc8AAAABBKMKWw | 45,763 | https://github.com/huggingface/transformers/issues/45763 | https://api.github.com/repos/huggingface/transformers/issues/45763 | Ivan privitera 02-09-1993 | null | closed | completed | false | 0 | [] | [] | 2026-05-03T19:35:29Z | 2026-05-04T08:47:24Z | 2026-05-04T08:47:24Z | NONE | null | 20260504T180032Z | 2026-05-04T18:00:32Z | mirko772 | 207,275,767 | U_kgDODFrG9w | User | false |
huggingface/transformers | 4,372,762,900 | I_kwDOCUB6oc8AAAABBKMRFA | 45,764 | https://github.com/huggingface/transformers/issues/45764 | https://api.github.com/repos/huggingface/transformers/issues/45764 | Daniele Privitera 14-05-1998 | null | closed | completed | false | 0 | [] | [] | 2026-05-03T19:36:14Z | 2026-05-04T08:47:19Z | 2026-05-04T08:47:19Z | NONE | null | 20260504T180032Z | 2026-05-04T18:00:32Z | mirko772 | 207,275,767 | U_kgDODFrG9w | User | false |
huggingface/transformers | 4,372,769,780 | I_kwDOCUB6oc8AAAABBKMr9A | 45,765 | https://github.com/huggingface/transformers/issues/45765 | https://api.github.com/repos/huggingface/transformers/issues/45765 | Erika Privitera 14-08 -2003 | null | closed | completed | false | 0 | [] | [] | 2026-05-03T19:39:07Z | 2026-05-04T08:47:15Z | 2026-05-04T08:47:15Z | NONE | null | 20260504T180032Z | 2026-05-04T18:00:32Z | mirko772 | 207,275,767 | U_kgDODFrG9w | User | false |
huggingface/transformers | 4,372,771,685 | I_kwDOCUB6oc8AAAABBKMzZQ | 45,766 | https://github.com/huggingface/transformers/issues/45766 | https://api.github.com/repos/huggingface/transformers/issues/45766 | Verduci Caterina 21-07-1964 | null | closed | completed | false | 0 | [] | [] | 2026-05-03T19:40:01Z | 2026-05-04T08:47:10Z | 2026-05-04T08:47:10Z | NONE | null | 20260504T180032Z | 2026-05-04T18:00:32Z | mirko772 | 207,275,767 | U_kgDODFrG9w | User | false |
huggingface/transformers | 4,372,772,935 | I_kwDOCUB6oc8AAAABBKM4Rw | 45,767 | https://github.com/huggingface/transformers/issues/45767 | https://api.github.com/repos/huggingface/transformers/issues/45767 | Privitera Fabrizio 16-12-1967 | null | closed | completed | false | 0 | [] | [] | 2026-05-03T19:40:34Z | 2026-05-04T08:47:05Z | 2026-05-04T08:47:05Z | NONE | null | 20260504T180032Z | 2026-05-04T18:00:32Z | mirko772 | 207,275,767 | U_kgDODFrG9w | User | false |
huggingface/transformers | 4,372,796,632 | I_kwDOCUB6oc8AAAABBKOU2A | 45,768 | https://github.com/huggingface/transformers/issues/45768 | https://api.github.com/repos/huggingface/transformers/issues/45768 | Macro Privitera Pitbull 24-10-2025 | null | closed | completed | false | 0 | [] | [] | 2026-05-03T19:51:36Z | 2026-05-04T08:47:00Z | 2026-05-04T08:47:00Z | NONE | null | 20260504T180032Z | 2026-05-04T18:00:32Z | mirko772 | 207,275,767 | U_kgDODFrG9w | User | false |
huggingface/transformers | 4,381,391,476 | I_kwDOCUB6oc8AAAABBSa6dA | 45,779 | https://github.com/huggingface/transformers/issues/45779 | https://api.github.com/repos/huggingface/transformers/issues/45779 | [Windows] RTX 5070 Ti (Blackwell sm_120) - setup and deployment notes | ### Environment
- GPU: NVIDIA GeForce RTX 5070 Ti Laptop GPU (Blackwell, compute capability 12.0)
- Driver: 595.79 (CUDA 13.2)
- OS: Windows 11
- Python: 3.14
- transformers: [latest]
### Problem
transformers on RTX 5070 Ti requires workarounds:
- **TORCH_CUDA_ARCH_LIST=12.0** required for Blackwell
- Model loading m... | open | null | false | 0 | [] | [] | 2026-05-05T03:54:09Z | 2026-05-05T03:54:09Z | null | NONE | null | 20260505T060044Z | 2026-05-05T06:00:44Z | loongmiaow-pixel | 252,396,232 | U_kgDODwtCyA | User | false |
huggingface/transformers | 3,583,802,125 | I_kwDOCUB6oc7VnHsN | 41,999 | https://github.com/huggingface/transformers/issues/41999 | https://api.github.com/repos/huggingface/transformers/issues/41999 | Add Timestamp Support for Voxtral Models | ### Feature request
Add support for segment-level timestamps for the Voxtral models (e.g., mistralai/Voxtral-Mini-3B-2507), similar to the existing timestamp functionality available in the Whisper models.
### Motivation
According to the [official Mistral documentation on audio transcription](https://docs.mistral.ai/... | open | null | false | 9 | [
"Feature request"
] | [] | 2025-11-03T21:23:01Z | 2026-05-05T09:04:47Z | null | NONE | null | 20260505T120036Z | 2026-05-05T12:00:36Z | juzhxng | 22,426,359 | MDQ6VXNlcjIyNDI2MzU5 | User | false |
huggingface/transformers | 2,481,366,710 | I_kwDOCUB6oc6T5qa2 | 32,946 | https://github.com/huggingface/transformers/issues/32946 | https://api.github.com/repos/huggingface/transformers/issues/32946 | Support `StaticCache` in assisted generation | Looking for contributions!
Assisted generation (or speculative decoding) is a strategy to speed up generation. Using `StaticCache` and `torch.compile` is another strategy to speed up generation. Currently, the two are not compatible. It would be nice to be able to use both at the same time, for maximum speed 😎
... | closed | completed | false | 3 | [
"Good Difficult Issue",
"Generation",
"Cache"
] | [] | 2024-08-22T17:34:38Z | 2026-05-05T13:02:50Z | 2026-05-05T13:02:50Z | CONTRIBUTOR | null | 20260505T180037Z | 2026-05-05T18:00:37Z | gante | 12,240,844 | MDQ6VXNlcjEyMjQwODQ0 | User | false |
huggingface/transformers | 4,387,308,095 | I_kwDOCUB6oc8AAAABBYECPw | 45,796 | https://github.com/huggingface/transformers/issues/45796 | https://api.github.com/repos/huggingface/transformers/issues/45796 | grootN_training | VLA model trainng for academic purposes. Requires Flash Attention 2. | closed | completed | false | 0 | [] | [] | 2026-05-05T21:46:46Z | 2026-05-06T10:55:54Z | 2026-05-06T10:55:54Z | NONE | null | 20260506T120019Z | 2026-05-06T12:00:19Z | zeic1 | 120,256,880 | U_kgDOByr5cA | User | false |
huggingface/transformers | 4,387,748,264 | I_kwDOCUB6oc8AAAABBYe5qA | 45,797 | https://github.com/huggingface/transformers/issues/45797 | https://api.github.com/repos/huggingface/transformers/issues/45797 | [BUG] HF Hub is DOWN on AWS in Xet mode | ### System Info
# Xet-backed model download hangs midway for `Qwen/Qwen3.6-35B-A3B`
## Summary
Downloading `Qwen/Qwen3.6-35B-A3B` on a p5en.48xlarge in us-east-2 on AWS hangs partway through when using the default Hugging Face download path. Clearing `~/.cache` does not fix it.
The evidence suggests the hang is in ... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-05-05T23:41:44Z | 2026-05-07T15:29:12Z | 2026-05-07T15:29:12Z | NONE | null | 20260507T180032Z | 2026-05-07T18:00:32Z | michaelroyzen | 45,830,328 | MDQ6VXNlcjQ1ODMwMzI4 | User | false |
huggingface/transformers | 4,388,500,815 | I_kwDOCUB6oc8AAAABBZM1Tw | 45,800 | https://github.com/huggingface/transformers/issues/45800 | https://api.github.com/repos/huggingface/transformers/issues/45800 | incompatiblity between torch 2.4.1 and transformers 5.8.0 | ### System Info
torch 2.4.1
transformers 5.8.0
### 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, ...)
- [x] My own task or dataset (give details below)
### Re... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-05-06T03:19:54Z | 2026-05-08T11:54:28Z | 2026-05-08T11:54:28Z | NONE | null | 20260508T120022Z | 2026-05-08T12:00:22Z | nickjyj | 46,984,040 | MDQ6VXNlcjQ2OTg0MDQw | User | false |
huggingface/transformers | 2,054,646,556 | I_kwDOCUB6oc56d2sc | 28,218 | https://github.com/huggingface/transformers/issues/28218 | https://api.github.com/repos/huggingface/transformers/issues/28218 | Tokenizer adds an additional space after the added token | ### System Info
- `transformers` version: 4.35.2
- Platform: Linux-6.1.58+-x86_64-with-glibc2.35
- Python version: 3.10.12
- Huggingface_hub version: 0.19.4
- Safetensors version: 0.4.1
- Accelerate version: 0.25.0
- Accelerate config: not found
- PyTorch version (GPU?): 2.1.0+cu121 (False)
- Tensorflow ver... | open | null | false | 7 | [
"Good Difficult Issue"
] | [] | 2023-12-23T03:43:18Z | 2026-05-07T10:18:25Z | null | NONE | null | 20260507T180032Z | 2026-05-07T18:00:32Z | kitkhai | 71,968,397 | MDQ6VXNlcjcxOTY4Mzk3 | User | false |
huggingface/transformers | 4,390,807,480 | I_kwDOCUB6oc8AAAABBbZnuA | 45,803 | https://github.com/huggingface/transformers/issues/45803 | https://api.github.com/repos/huggingface/transformers/issues/45803 | [Bug] Bare `except:` in FuyuBatchFeature.convert_to_tensors() swallows KeyboardInterrupt and hides real errors | ### System Info
transformers: 5.7.0.dev0 (main, commit 8659ae6245)
Python: 3.13.7
Platform: Windows 11 AMD64
PyTorch: 2.11.0+cu126
### Who can help?
@yonigozlan @molbap
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` ... | closed | completed | false | 3 | [
"bug"
] | [] | 2026-05-06T11:01:42Z | 2026-05-08T09:56:54Z | 2026-05-08T09:56:54Z | CONTRIBUTOR | null | 20260508T120022Z | 2026-05-08T12:00:22Z | Abineshabee | 104,718,709 | U_kgDOBj3hdQ | User | false |
huggingface/transformers | 4,393,029,377 | I_kwDOCUB6oc8AAAABBdhPAQ | 45,810 | https://github.com/huggingface/transformers/issues/45810 | https://api.github.com/repos/huggingface/transformers/issues/45810 | Feature request Add Qwen3_5ForTokenClassification for using as a value model. | ### Feature request
Add Qwen3_5ForTokenClassification for using as a value model.
### Motivation
In verl, they use AutoModelForTokenClassification for loading value model.
https://github.com/volcengine/verl/blob/2d6c6dbb39bf846d4ebf98c89fc5b4f49c37dd3d/verl/utils/model.py#L627
so I want to add this for using trans... | open | null | false | 3 | [
"Feature request"
] | [] | 2026-05-06T16:49:05Z | 2026-05-07T09:55:56Z | null | NONE | null | 20260507T180032Z | 2026-05-07T18:00:32Z | han2-l | 188,972,972 | U_kgDOC0N_rA | User | false |
huggingface/transformers | 4,393,508,674 | I_kwDOCUB6oc8AAAABBd-fQg | 45,812 | https://github.com/huggingface/transformers/issues/45812 | https://api.github.com/repos/huggingface/transformers/issues/45812 | `AutoTokenizer` produces wrong token IDs for all Granite models (silent v4→v5 regression) | ### System Info
- `transformers` version: 5.8.0 (also reproduced on 5.0.0 through 5.7.0)
- Platform: Linux-5.14.0-503.11.1.el9_5.x86_64-x86_64-with-glibc2.34
- Python version: 3.12.13
- Huggingface_hub version: 1.14.0
- Safetensors version: 0.7.0
- Tokenizers version: 0.22.2
- PyTorch version: not installed (tokenizer... | open | null | false | 2 | [
"bug"
] | [] | 2026-05-06T18:23:26Z | 2026-05-11T19:23:51Z | null | NONE | null | 20260512T000156Z | 2026-05-12T00:01:56Z | kndtran | 19,249,995 | MDQ6VXNlcjE5MjQ5OTk1 | User | false |
huggingface/transformers | 2,937,833,531 | I_kwDOCUB6oc6vG8g7 | 36,879 | https://github.com/huggingface/transformers/issues/36879 | https://api.github.com/repos/huggingface/transformers/issues/36879 | Add RF-DETR model | ### Model description
Hi,
Roboflow just released [RF-DETR](https://blog.roboflow.com/rf-detr/#how-to-use-rf-detr), a new object detection model based on DINOv2, Deformable DETR (which are already part of transformers) and LW DETR.
### Open source status
- [x] The model implementation is available
- [x] The model ... | closed | completed | false | 1 | [
"New model",
"Vision"
] | [] | 2025-03-21T09:38:40Z | 2026-05-07T06:56:41Z | 2026-05-07T06:56:41Z | CONTRIBUTOR | null | 20260507T180032Z | 2026-05-07T18:00:32Z | sbucaille | 24,275,548 | MDQ6VXNlcjI0Mjc1NTQ4 | User | false |
huggingface/transformers | 4,397,611,087 | I_kwDOCUB6oc8AAAABBh44Tw | 45,820 | https://github.com/huggingface/transformers/issues/45820 | https://api.github.com/repos/huggingface/transformers/issues/45820 | DeepSeekV4 的 CSA Indexer 因果 mask | 官方 | ✅ 有 mask，Query[t] 只检索 compressed[0..t//ratio] | 因果正确
Transformers | ❌ 无 mask，Query[t] 可检索所有 compressed | 信息泄露
对比官方推理代码的实现，在indexer做检索时，只会检索之前已经压缩过的window，而在transformers中，没有看到这部分实现。 | closed | completed | false | 6 | [] | [] | 2026-05-07T09:22:07Z | 2026-05-12T09:28:05Z | 2026-05-12T09:28:05Z | NONE | null | 20260512T120027Z | 2026-05-12T12:00:27Z | slZheng077 | 163,490,356 | U_kgDOCb6qNA | User | false |
huggingface/transformers | 4,398,351,189 | I_kwDOCUB6oc8AAAABBimDVQ | 45,823 | https://github.com/huggingface/transformers/issues/45823 | https://api.github.com/repos/huggingface/transformers/issues/45823 | Gemma4 PLE device mismatch with `device_map="auto"` during forward | ### System Info
- `transformers` version: 5.8.0
- Platform: Linux-6.6.122+-x86_64-with-glibc2.35
- Python version: 3.11.13
- Huggingface_hub version: 1.14.0
- Safetensors version: 0.6.2
- Accelerate version: 1.10.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?): 2.1... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-05-07T11:23:10Z | 2026-05-07T19:50:11Z | 2026-05-07T19:50:11Z | NONE | null | 20260508T000035Z | 2026-05-08T00:00:35Z | rishon-galileo | 203,699,605 | U_kgDODCQ1lQ | User | false |
huggingface/transformers | 2,419,933,304 | I_kwDOCUB6oc6QPUB4 | 32,101 | https://github.com/huggingface/transformers/issues/32101 | https://api.github.com/repos/huggingface/transformers/issues/32101 | Using Trainer + a pretrained tokenizer + 4D attention mask is extremely slow | ### System Info
transformers 4.41.0
### Who can help?
@ArthurZucker
### 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 (give ... | closed | completed | false | 6 | [
"Good Difficult Issue",
"bug"
] | [] | 2024-07-19T21:33:09Z | 2026-05-12T12:50:44Z | 2026-05-12T12:50:44Z | NONE | null | 20260512T180027Z | 2026-05-12T18:00:27Z | JackCai1206 | 16,009,360 | MDQ6VXNlcjE2MDA5MzYw | User | false |
huggingface/transformers | 4,402,849,723 | I_kwDOCUB6oc8AAAABBm4nuw | 45,834 | https://github.com/huggingface/transformers/issues/45834 | https://api.github.com/repos/huggingface/transformers/issues/45834 | Kosmos2.5: index error on long ocr input | ### System Info
- `transformers` version: 5.8.0
- Platform: Linux-7.0.3-arch1-2-x86_64-with-glibc2.43
- Python version: 3.13.11
- Huggingface_hub version: 1.14.0
- Safetensors version: 0.7.0
- Accelerate version: not installed
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (acceler... | open | null | false | 3 | [
"bug"
] | [] | 2026-05-08T00:29:00Z | 2026-05-11T19:17:44Z | null | NONE | null | 20260512T000156Z | 2026-05-12T00:01:56Z | nunq | 46,054,695 | MDQ6VXNlcjQ2MDU0Njk1 | User | false |
huggingface/transformers | 1,168,963,497 | I_kwDOCUB6oc5FrPep | 16,157 | https://github.com/huggingface/transformers/issues/16157 | https://api.github.com/repos/huggingface/transformers/issues/16157 | Implement Maximal Update Parametrization (muP) | # 🚀 Feature request
<!-- A clear and concise description of the feature proposal.
Please provide a link to the paper and code in case they exist. -->
This request is to open up a discussion on 1) whether it makes sense to implement [Maximal Update Parametrization (abbreviated muP)](http://arxiv.org/abs/220... | open | reopened | false | 18 | [
"WIP"
] | [] | 2022-03-14T22:08:55Z | 2026-05-08T16:37:34Z | null | NONE | null | 20260508T180023Z | 2026-05-08T18:00:23Z | thegregyang | 53,244,851 | MDQ6VXNlcjUzMjQ0ODUx | User | false |
huggingface/transformers | 4,404,517,571 | I_kwDOCUB6oc8AAAABBoeaww | 45,841 | https://github.com/huggingface/transformers/issues/45841 | https://api.github.com/repos/huggingface/transformers/issues/45841 | Embed Agent Friendly Code Score Badge | Hi team — I'm Himanshu, I built Agent Friendly Code, which scores public repos on how legible they are to AI coding agents (clear conventions, docs, tests, build signals — not anything about accepting agent-authored PRs).
`transformers` scored 75.9/100 — full breakdown: https://www.agentfriendlycode.com/repo/126
If y... | open | null | false | 0 | [] | [] | 2026-05-08T07:04:06Z | 2026-05-08T07:04:06Z | null | NONE | null | 20260508T120022Z | 2026-05-08T12:00:22Z | hsnice16 | 56,081,584 | MDQ6VXNlcjU2MDgxNTg0 | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.