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 ... | open | null | false | 7 | [
"bug"
] | [] | 2026-04-13T15:46:43Z | 2026-04-16T11:56:08Z | null | NONE | null | 20260416T222012Z | 2026-04-16T22:20:12Z | 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 | 1 | [] | [] | 2026-04-13T23:27:06Z | 2026-04-15T15:37:11Z | null | MEMBER | null | 20260415T224019Z | 2026-04-15T22:40:19Z | 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... | open | null | false | 0 | [
"bug"
] | [] | 2026-04-14T17:50:49Z | 2026-04-14T18:14:43Z | null | CONTRIBUTOR | null | 20260414T200457Z | 2026-04-14T20:04:57Z | 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 | 3 | [] | [] | 2026-04-15T06:13:12Z | 2026-04-15T12:17:47Z | null | NONE | null | 20260415T224019Z | 2026-04-15T22:40:19Z | 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... | open | null | false | 0 | [] | [
"tarekziade"
] | 2026-04-15T12:36:24Z | 2026-04-15T12:42:42Z | null | MEMBER | null | 20260415T224019Z | 2026-04-15T22:40:19Z | 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 ... | open | null | false | 4 | [
"bug"
] | [] | 2026-04-15T12:48:42Z | 2026-04-16T14:32:09Z | null | CONTRIBUTOR | null | 20260416T222012Z | 2026-04-16T22:20:12Z | 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 | 2 | [
"bug"
] | [] | 2026-04-15T16:36:31Z | 2026-04-16T10:49:09Z | null | NONE | null | 20260416T222012Z | 2026-04-16T22:20:12Z | 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 | 1 | [
"bug"
] | [] | 2026-04-16T06:52:43Z | 2026-04-16T09:30:36Z | null | CONTRIBUTOR | null | 20260416T222012Z | 2026-04-16T22:20:12Z | 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... | open | null | false | 3 | [
"Should Fix",
"bug"
] | [] | 2026-04-16T14:48:15Z | 2026-04-18T02:59:15Z | null | CONTRIBUTOR | null | 20260418T030535Z | 2026-04-18T03:05:35Z | 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 ... | open | null | false | 3 | [
"bug"
] | [] | 2026-04-16T14:58:54Z | 2026-04-17T11:48:17Z | null | NONE | null | 20260417T180542Z | 2026-04-17T18:05:42Z | 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 | 2 | [] | [] | 2026-04-16T15:57:28Z | 2026-04-17T17:12:08Z | null | NONE | null | 20260417T180542Z | 2026-04-17T18:05:42Z | 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 | 46 | [
"Good Second Issue",
"Feature request"
] | [] | 2024-09-02T13:41:47Z | 2026-04-17T00:09:50Z | null | MEMBER | null | 20260417T180542Z | 2026-04-17T18:05:42Z | 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 | 1 | [] | [] | 2026-04-17T08:50:48Z | 2026-04-17T11:43:10Z | null | NONE | null | 20260417T180542Z | 2026-04-17T18:05:42Z | 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 ... | open | null | false | 0 | [
"bug"
] | [] | 2026-04-17T13:10:25Z | 2026-04-17T13:10:25Z | null | NONE | null | 20260417T180542Z | 2026-04-17T18:05:42Z | 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 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.