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 | 4,887,996,003 | I_kwDOCUB6oc8AAAABI1jmYw | 47,328 | https://github.com/huggingface/transformers/issues/47328 | https://api.github.com/repos/huggingface/transformers/issues/47328 | Qwen2.5-Omni Token2Wav DiT uses mismatched RoPE layout (half-split cos/sin with interleaved rotate) — regression from #39847 | ### System Info
- `transformers`: main (reproduced at `ab1771c`); bug present since the v5 RoPE refactor (#39847, merged 2025-10-17)
- Code-level bug, weight-independent — reproducible on CPU, no GPU or pretrained weights needed
### Who can help?
@eustlb @ebezzam
### Description
`Qwen2_5OmniDiTRotaryEmbedding.forward... | open | null | false | 2 | [] | [] | 2026-07-14T23:55:23Z | 2026-07-18T21:25:53Z | null | NONE | null | 20260719T000028Z | 2026-07-19T00:00:28Z | HenryVarro666 | 48,014,495 | MDQ6VXNlcjQ4MDE0NDk1 | User | false |
huggingface/transformers | 4,889,536,762 | I_kwDOCUB6oc8AAAABI3Bo-g | 47,331 | https://github.com/huggingface/transformers/issues/47331 | https://api.github.com/repos/huggingface/transformers/issues/47331 | MetalConfig quantization: batched generate silently corrupts all rows after the first — affine_qmm_t computes only batch element 0 of 3D inputs (verified, one-line fix) | ### System Info
```
- `transformers` version: 5.13.1
- Platform: macOS-26.5.2-arm64-arm-64bit
- Python version: 3.11.15
- Huggingface_hub version: 1.23.0
- Safetensors version: 0.8.0
- Accelerate version: 1.14.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?): 2.13.0 ... | open | null | false | 4 | [] | [] | 2026-07-15T05:19:15Z | 2026-07-21T17:38:44Z | null | NONE | null | 20260721T180022Z | 2026-07-21T18:00:22Z | fenilsonani | 107,786,545 | U_kgDOBmyxMQ | User | false |
huggingface/transformers | 4,889,942,210 | I_kwDOCUB6oc8AAAABI3aYwg | 47,332 | https://github.com/huggingface/transformers/issues/47332 | https://api.github.com/repos/huggingface/transformers/issues/47332 | [Offloading] Cannot save disk-offloaded `Qwen3-VL-32B-Instruct` | ### System Info
- `transformers` version: 5.13.0.dev0
- Platform: Linux-5.14.0-611.41.1.el9_7.x86_64-x86_64-with-glibc2.34
- Python version: 3.12.12
- Huggingface_hub version: 1.18.0
- Safetensors version: 0.8.0
- Accelerate version: 1.15.0.dev0
- Accelerate config: not found
- DeepSpeed version: not installed
- Py... | closed | duplicate | false | 0 | [
"bug"
] | [] | 2026-07-15T06:29:35Z | 2026-07-15T07:06:10Z | 2026-07-15T07:06:10Z | CONTRIBUTOR | null | 20260715T120033Z | 2026-07-15T12:00:33Z | kylesayrs | 17,103,692 | MDQ6VXNlcjE3MTAzNjky | User | false |
huggingface/transformers | 4,890,015,845 | I_kwDOCUB6oc8AAAABI3e4ZQ | 47,333 | https://github.com/huggingface/transformers/issues/47333 | https://api.github.com/repos/huggingface/transformers/issues/47333 | [Offloading] Cannot save disk-offloaded `Qwen/Qwen3-30B-A3B` | ### System Info
- `transformers` version: 5.13.0.dev0
- Platform: Linux-5.14.0-611.41.1.el9_7.x86_64-x86_64-with-glibc2.34
- Python version: 3.12.12
- Huggingface_hub version: 1.18.0
- Safetensors version: 0.8.0
- Accelerate version: 1.15.0.dev0
- Accelerate config: not found
- DeepSpeed version: not installed
- Py... | open | null | false | 5 | [
"bug"
] | [] | 2026-07-15T06:42:42Z | 2026-07-16T15:09:48Z | null | CONTRIBUTOR | null | 20260717T000052Z | 2026-07-17T00:00:52Z | kylesayrs | 17,103,692 | MDQ6VXNlcjE3MTAzNjky | User | false |
huggingface/transformers | 4,890,401,779 | I_kwDOCUB6oc8AAAABI32b8w | 47,337 | https://github.com/huggingface/transformers/issues/47337 | https://api.github.com/repos/huggingface/transformers/issues/47337 | ## Flaky test: `IndexError` — bbox coordinate values out of 0-1000 range | ### System Info
### Affected tests
Multiple model tests that use random bounding box generation fail with:
```
IndexError: The `bbox` coordinate values should be within 0-1000 range.
```
This affects `layoutlm`, `layoutlmv2`, `layoutlmv3`, `lilt`, `pp_doclayout_v2` — any model whose `modeling_*.py` validates bbox c... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-07-15T07:42:19Z | 2026-07-15T11:45:28Z | 2026-07-15T11:45:09Z | NONE | null | 20260715T120033Z | 2026-07-15T12:00:33Z | gygdh-001 | 43,057,576 | MDQ6VXNlcjQzMDU3NTc2 | User | false |
huggingface/transformers | 2,744,622,076 | I_kwDOCUB6oc6jl5v8 | 35,308 | https://github.com/huggingface/transformers/issues/35308 | https://api.github.com/repos/huggingface/transformers/issues/35308 | Multi-GPU training crashes with IterableDataset and different length input (e.g. Next token prediction) | ### System Info
- `transformers` version: 4.46.3
- Platform: Linux-6.6.20-aufs-1-x86_64-with-glibc2.36
- Python version: 3.11.2
- Huggingface_hub version: 0.26.1
- Safetensors version: 0.4.5
- Accelerate version: 1.2.1
- Accelerate config: not found
- PyTorch version (GPU?): 2.5.1+cu124 (True)
- Tensorflow ... | open | null | false | 21 | [
"Good Second Issue",
"bug"
] | [] | 2024-12-17T11:03:19Z | 2026-07-16T15:26:32Z | null | CONTRIBUTOR | null | 20260717T000052Z | 2026-07-17T00:00:52Z | avishaiElmakies | 36,810,152 | MDQ6VXNlcjM2ODEwMTUy | User | false |
huggingface/transformers | 4,898,204,693 | I_kwDOCUB6oc8AAAABI_SsFQ | 47,355 | https://github.com/huggingface/transformers/issues/47355 | https://api.github.com/repos/huggingface/transformers/issues/47355 | GlmMoeDsaConfig: legacy `num_experts` unconditionally overrides explicit `n_routed_experts` (sibling of #46739) | ### System info
- `transformers` `main` (verified against `src/transformers/models/glm_moe_dsa/configuration_glm_moe_dsa.py` today)
- Python 3.12, torch 2.11, vLLM 0.25.1
- Affected checkpoint: [`0xSero/GLM-5.2-504B-Nvidia`](https://huggingface.co/0xSero/GLM-5.2-504B-Nvidia) —
`GlmMoeDsaForCausalLM`, **no `auto_map`... | open | null | false | 5 | [] | [] | 2026-07-16T02:50:44Z | 2026-07-19T03:43:26Z | null | NONE | null | 20260719T060027Z | 2026-07-19T06:00:27Z | hermia-ai | 227,936,971 | U_kgDODZYKyw | User | false |
huggingface/transformers | 1,171,036,545 | I_kwDOCUB6oc5FzJmB | 16,195 | https://github.com/huggingface/transformers/issues/16195 | https://api.github.com/repos/huggingface/transformers/issues/16195 | Gpt2 large for onnx exportation and int8 quantization | Hi, for model big as 7GB, does transformers support export to onnx?? Any tutorial about big model? | closed | completed | false | 5 | [
"New model"
] | [
"lewtun"
] | 2022-03-16T13:36:40Z | 2026-07-16T03:20:02Z | 2026-07-16T03:20:02Z | NONE | null | 20260716T060036Z | 2026-07-16T06:00:36Z | lucasjinreal | 21,303,438 | MDQ6VXNlcjIxMzAzNDM4 | User | false |
huggingface/transformers | 3,505,413,795 | I_kwDOCUB6oc7Q8F6j | 41,519 | https://github.com/huggingface/transformers/issues/41519 | https://api.github.com/repos/huggingface/transformers/issues/41519 | Add Shared Vision–Language Negative Sampling Utility for BLIP / LLaVA / IDEFICS | ### Feature request
Add a unified vision-language negative sampling utility for contrastive learning. Support hard negative mining and in-batch negative sampling. Compatible with BLIP, LLaVA, and IDEFICS training pipelines. Modular design for easy integration into existing VLM frameworks
### Motivation
VLM training ... | closed | not_planned | false | 2 | [
"Feature request"
] | [] | 2025-10-11T08:13:34Z | 2026-07-16T03:23:57Z | 2026-07-16T03:23:57Z | CONTRIBUTOR | null | 20260716T060036Z | 2026-07-16T06:00:36Z | Aki-07 | 95,642,646 | U_kgDOBbNkFg | User | false |
huggingface/transformers | 2,975,754,834 | I_kwDOCUB6oc6xXmpS | 37,334 | https://github.com/huggingface/transformers/issues/37334 | https://api.github.com/repos/huggingface/transformers/issues/37334 | Support LLMDet in transformers | ### Model description
Could you please kindly consider add [LLMDet](https://arxiv.org/abs/2501.18954) (CVPR2025 Highlight) to transformers, which is a next-generation open-vocabulary object detector. The architecture of LLMDet is similar to GroundingDino and most of the code can be reused. We have provide the code and... | closed | completed | false | 7 | [
"New model"
] | [] | 2025-04-07T06:23:44Z | 2026-07-16T03:36:51Z | 2026-07-16T03:32:44Z | NONE | null | 20260716T060036Z | 2026-07-16T06:00:36Z | fushh | 71,886,569 | MDQ6VXNlcjcxODg2NTY5 | User | false |
huggingface/transformers | 1,058,212,993 | I_kwDOCUB6oc4_EwyB | 14,456 | https://github.com/huggingface/transformers/issues/14456 | https://api.github.com/repos/huggingface/transformers/issues/14456 | [New Model] DocFormer: End-to-End Transformer for Document Understanding | # 🌟 New model addition
## Model description
See _"DocFormer: End-to-End Transformer for Document Understanding", Appalaraju et al (ICCV 2021)_ on [CVF](https://openaccess.thecvf.com/content/ICCV2021/papers/Appalaraju_DocFormer_End-to-End_Transformer_for_Document_Understanding_ICCV_2021_paper.pdf) and [arXiv](htt... | closed | not_planned | false | 19 | [
"New model"
] | [] | 2021-11-19T07:08:44Z | 2026-07-16T13:31:22Z | 2026-07-16T13:31:22Z | NONE | null | 20260717T000052Z | 2026-07-17T00:00:52Z | athewsey | 19,252,478 | MDQ6VXNlcjE5MjUyNDc4 | User | false |
huggingface/transformers | 1,902,199,616 | I_kwDOCUB6oc5xYUNA | 26,242 | https://github.com/huggingface/transformers/issues/26242 | https://api.github.com/repos/huggingface/transformers/issues/26242 | WhisperForCTC | ### Feature request
Request to add WhisperForCTC model.
### Motivation
it would be cool if we had custom WhisperForCTC with Whisper encoder and ctc head, just like Wav2vec2, but since whisper is based on mel spectograms, I think it may bring better results.
### Your contribution
Here is my implementatio... | closed | not_planned | false | 14 | [
"Feature request"
] | [] | 2023-09-19T04:50:05Z | 2026-07-16T13:34:49Z | 2026-07-16T13:34:49Z | NONE | null | 20260717T000052Z | 2026-07-17T00:00:52Z | DavraYoung | 33,338,429 | MDQ6VXNlcjMzMzM4NDI5 | User | false |
huggingface/transformers | 3,375,632,692 | I_kwDOCUB6oc7JNBE0 | 40,621 | https://github.com/huggingface/transformers/issues/40621 | https://api.github.com/repos/huggingface/transformers/issues/40621 | [Feature Proposal] Add GGUF support for Korean-specific LLM architectures | ### Feature request
This is a feature proposal to extend GGUF support in `transformers` for several Korean-specific LLM architectures that have recently been released on the Hugging Face Hub.
Reference issue: [Community contribution: Adding GGUF support for more architectures](https://github.com/huggingface/transform... | open | null | false | 3 | [
"Feature request"
] | [] | 2025-09-02T11:33:31Z | 2026-07-18T14:56:45Z | null | CONTRIBUTOR | null | 20260718T180026Z | 2026-07-18T18:00:26Z | Kim-Ju-won | 81,630,351 | MDQ6VXNlcjgxNjMwMzUx | User | false |
huggingface/transformers | 941,301,047 | MDU6SXNzdWU5NDEzMDEwNDc= | 12,627 | https://github.com/huggingface/transformers/issues/12627 | https://api.github.com/repos/huggingface/transformers/issues/12627 | Add Flax Models to Pipelines | # 🚀 Feature request
Hi y'all, I am trying a GPTNeo Flax model and am wanting to use it in the text generation pipeline. However, it is currently not supported. From looking at the current implementation of flax models and the text generation pipeline, it should be a relatively easy (famous last words) addition.
... | closed | not_planned | false | 1 | [
"Feature request"
] | [] | 2021-07-10T18:26:54Z | 2026-07-16T15:29:19Z | 2026-07-16T15:29:18Z | CONTRIBUTOR | null | 20260717T000052Z | 2026-07-17T00:00:52Z | ncoop57 | 7,613,470 | MDQ6VXNlcjc2MTM0NzA= | User | false |
huggingface/transformers | 1,639,989,623 | I_kwDOCUB6oc5hwEF3 | 22,372 | https://github.com/huggingface/transformers/issues/22372 | https://api.github.com/repos/huggingface/transformers/issues/22372 | Add Restormer | ### Model description
**Restormer: Efficient Transformer for High-Resolution Image Restoration** was published in CVPR 2022, which introduced a new Vision Transformer based architecture for Image Restoration tasks like Deraining, Motion Deblurring, Defocus Deblurring and Denoising. It reduced the time complexity of Se... | closed | not_planned | false | 2 | [
"New model"
] | [] | 2023-03-24T20:25:27Z | 2026-07-16T17:50:05Z | 2026-07-16T17:50:05Z | NONE | null | 20260717T000052Z | 2026-07-17T00:00:52Z | tushdon2 | 76,245,823 | MDQ6VXNlcjc2MjQ1ODIz | User | false |
huggingface/transformers | 4,905,685,497 | I_kwDOCUB6oc8AAAABJGbR-Q | 47,373 | https://github.com/huggingface/transformers/issues/47373 | https://api.github.com/repos/huggingface/transformers/issues/47373 | examples/training/distributed_training.py crashes on import: LOCAL_RANK/WORLD_SIZE/RANK are overwritten by OMPI_* vars | ### System Info
- `transformers`: main (as of 2026-07-16)
- File: [`examples/training/distributed_training.py`](https://github.com/huggingface/transformers/blob/main/examples/training/distributed_training.py)
### Who can help?
@ArthurZucker @muellerzr — this looks like leftover WIP from the distributed training exam... | closed | completed | false | 1 | [] | [] | 2026-07-16T20:19:50Z | 2026-07-16T20:23:04Z | 2026-07-16T20:23:04Z | NONE | null | 20260717T000052Z | 2026-07-17T00:00:52Z | berkgungor | 48,314,701 | MDQ6VXNlcjQ4MzE0NzAx | User | false |
huggingface/transformers | 1,756,577,492 | I_kwDOCUB6oc5osz7U | 24,272 | https://github.com/huggingface/transformers/issues/24272 | https://api.github.com/repos/huggingface/transformers/issues/24272 | Finetuning Whisper with prompts | ### Feature request
Training code implementation for finetuning Whisper using prompts.
Hi All,
I’m trying to finetune Whisper by resuming its pre-training task and adding initial prompts as part of the model’s forward pass. I saw this [amazing tutorial](https://huggingface.co/blog/fine-tune-whisper), however, i... | closed | not_planned | false | 41 | [
"Feature request"
] | [] | 2023-06-14T10:36:04Z | 2026-07-16T20:29:12Z | 2026-07-16T20:29:12Z | CONTRIBUTOR | null | 20260717T000052Z | 2026-07-17T00:00:52Z | AvivSham | 43,371,254 | MDQ6VXNlcjQzMzcxMjU0 | User | false |
huggingface/transformers | 4,905,724,008 | I_kwDOCUB6oc8AAAABJGdoaA | 47,375 | https://github.com/huggingface/transformers/issues/47375 | https://api.github.com/repos/huggingface/transformers/issues/47375 | Two systematic bugs in examples/pytorch/*_no_trainer.py: broken --resume_from_checkpoint auto-discover branch, and wrong eval loss/perplexity on partial batches | ### System Info
- `transformers`: `main` (as of 2026-07-16)
- Affected: **every `*_no_trainer.py` example** under `examples/pytorch/`
- No dependency-specific reproducer needed — these are pure Python control‑flow / accounting bugs.
I hit two independent, systematic bugs in the Accelerate‑based `no_trainer` examples.... | open | null | false | 1 | [] | [] | 2026-07-16T20:25:08Z | 2026-07-18T04:56:16Z | null | NONE | null | 20260718T060036Z | 2026-07-18T06:00:36Z | berkgungor | 48,314,701 | MDQ6VXNlcjQ4MzE0NzAx | User | false |
huggingface/transformers | 4,906,985,357 | I_kwDOCUB6oc8AAAABJHqnjQ | 47,378 | https://github.com/huggingface/transformers/issues/47378 | https://api.github.com/repos/huggingface/transformers/issues/47378 | Question about overriding bnb_4bit_compute_dtype for pre-quantized checkpoints | ### System information
- `transformers`: 5.13.1
- `bitsandbytes`: 0.49.2
- `accelerate`: 1.14.0
- `huggingface_hub`: 1.23.0
- `safetensors`: 0.8.0
- PyTorch: 2.11.0+cu128
- CUDA: 12.8
- Python: 3.12.13
- Platform: Google Colab / Linux
- GPU: NVIDIA Tesla T4
- CUDA capability: 7.5
- Native BF16 support: false
- Distrib... | open | null | false | 1 | [] | [] | 2026-07-17T00:27:34Z | 2026-07-18T15:50:34Z | null | NONE | null | 20260718T180026Z | 2026-07-18T18:00:26Z | John6666cat | 186,692,226 | U_kgDOCyCygg | User | false |
huggingface/transformers | 2,437,894,963 | I_kwDOCUB6oc6RT1Mz | 32,326 | https://github.com/huggingface/transformers/issues/32326 | https://api.github.com/repos/huggingface/transformers/issues/32326 | Length grouping for iterable datasets | ### Feature request
Similar to non-iterable datasets I would like functionality to group batches by similar length inputs.
Example of how this could be implemented is using a buffer to preload batches and sort in IterableDatasetShard in `accelerate` using the defined column for length grouping.
```
class Iter... | closed | not_planned | false | 3 | [
"Feature request"
] | [] | 2024-07-30T13:50:09Z | 2026-07-17T00:35:13Z | 2026-07-17T00:35:13Z | CONTRIBUTOR | null | 20260717T060031Z | 2026-07-17T06:00:31Z | ssharpe42 | 8,136,905 | MDQ6VXNlcjgxMzY5MDU= | User | false |
huggingface/transformers | 4,906,983,463 | I_kwDOCUB6oc8AAAABJHqgJw | 47,377 | https://github.com/huggingface/transformers/issues/47377 | https://api.github.com/repos/huggingface/transformers/issues/47377 | [serge] integration failure triage - 2026-07-16 | <!-- serge-triage-run:integration-failure-triage:2026-07-16 -->
Automated **integration-failure triage** for the daily CI window `2026-07-08 → 2026-07-16`.
This issue was generated by AI-assisted automation. The grouping, summaries, and recommended follow-up can be incomplete or misleading; verify the failures before... | closed | not_planned | false | 1 | [] | [
"tarekziade"
] | 2026-07-17T00:27:11Z | 2026-07-18T00:25:19Z | 2026-07-18T00:25:19Z | CONTRIBUTOR | null | 20260718T060036Z | 2026-07-18T06:00:36Z | github-actions[bot] | 41,898,282 | MDM6Qm90NDE4OTgyODI= | Bot | false |
huggingface/transformers | 2,024,486,636 | I_kwDOCUB6oc54qzbs | 27,837 | https://github.com/huggingface/transformers/issues/27837 | https://api.github.com/repos/huggingface/transformers/issues/27837 | torch CUDA graphs with HF generate | ### Feature request
In my experiments, I cannot get torch CUDA graphs to work with HF generate. CUDA graphs work fine when calling the forward pass of a model, but either due to static input/output sizes or something else, stream capture fails when calling .generate(). Can support for torch CUDA graphs be added?
### ... | closed | completed | false | 12 | [
"WIP"
] | [] | 2023-12-04T18:39:35Z | 2026-07-17T02:55:17Z | 2026-07-17T02:55:17Z | NONE | null | 20260717T060031Z | 2026-07-17T06:00:31Z | tsengalb99 | 33,385,672 | MDQ6VXNlcjMzMzg1Njcy | User | false |
huggingface/transformers | 2,062,055,346 | I_kwDOCUB6oc566Hey | 28,306 | https://github.com/huggingface/transformers/issues/28306 | https://api.github.com/repos/huggingface/transformers/issues/28306 | LLaMA-MoE | ### Model description
LLaMA-MoE is a series of token-choice based Mixture-of-Experts models on LLaMA2. It first partition LLaMA2's FFNs into multiple experts, then apply continual pre-training to recover its language abilities.
We believe LLaMA-MoE is a good start for MoE research under limited computing resources.... | closed | not_planned | false | 1 | [
"New model"
] | [] | 2024-01-02T07:49:56Z | 2026-07-17T13:11:32Z | 2026-07-17T13:11:32Z | NONE | null | 20260717T180032Z | 2026-07-17T18:00:32Z | Spico197 | 22,840,952 | MDQ6VXNlcjIyODQwOTUy | User | false |
huggingface/transformers | 2,562,317,569 | I_kwDOCUB6oc6Yud0B | 33,898 | https://github.com/huggingface/transformers/issues/33898 | https://api.github.com/repos/huggingface/transformers/issues/33898 | Flex attention support with arbitrary 4d mask for LlamaModel | ### Feature request
It would be nice to combine the benefits of flex attention and 4d masking.
Perhaps the llama model could be a first case, allowing arbitrary 4d masks to be handled via an efficient flex attention path.
### Motivation
Custom attention masking/biasing patterns lead to considerable improvements i... | closed | completed | false | 9 | [
"Feature request",
"Flash Attention"
] | [] | 2024-10-02T18:01:39Z | 2026-07-17T13:12:48Z | 2026-07-17T13:12:48Z | NONE | null | 20260717T180032Z | 2026-07-17T18:00:32Z | alex-hh | 5,719,745 | MDQ6VXNlcjU3MTk3NDU= | User | false |
huggingface/transformers | 1,148,962,454 | I_kwDOCUB6oc5Ee8aW | 15,813 | https://github.com/huggingface/transformers/issues/15813 | https://api.github.com/repos/huggingface/transformers/issues/15813 | Add OFA to transformers | # 🌟 New model addition
We recently proposed OFA, a unified model for multimodal pretraining, which achieves multiple SoTAs on downstream tasks, including image captioning, text-to-image generation, referring expression comprehension, etc. We would like to implement OFA on transformers if it is possible.
## Model d... | closed | not_planned | false | 5 | [
"New model"
] | [
"gchhablani"
] | 2022-02-24T07:42:42Z | 2026-07-17T13:14:27Z | 2026-07-17T13:14:27Z | CONTRIBUTOR | null | 20260717T180032Z | 2026-07-17T18:00:32Z | JustinLin610 | 27,664,428 | MDQ6VXNlcjI3NjY0NDI4 | User | false |
huggingface/transformers | 1,458,578,372 | I_kwDOCUB6oc5W8CPE | 20,362 | https://github.com/huggingface/transformers/issues/20362 | https://api.github.com/repos/huggingface/transformers/issues/20362 | Add FlexiBERT | ### Model description
FlexiBERT is a suite of *flexible* and *heterogeneous* models. The design space was proposed in this [paper](https://arxiv.org/abs/2205.11656) accepted for publication in the Journal of Artificial Intelligence Research.
### Open source status
- [X] The model implementation is available
- [X] Th... | closed | not_planned | false | 2 | [
"New model"
] | [] | 2022-11-21T21:08:10Z | 2026-07-17T13:16:42Z | 2026-07-17T13:16:42Z | CONTRIBUTOR | null | 20260717T180032Z | 2026-07-17T18:00:32Z | shikhartuli | 40,000,988 | MDQ6VXNlcjQwMDAwOTg4 | User | false |
huggingface/transformers | 2,242,114,332 | I_kwDOCUB6oc6Fo_Mc | 30,240 | https://github.com/huggingface/transformers/issues/30240 | https://api.github.com/repos/huggingface/transformers/issues/30240 | [i18n-PL] Translating docs to Polish | <!--
Note: Please search to see if an issue already exists for the language you are trying to translate.
-->
Hi!
Let's bring the documentation to all the Polish-speaking community 🌐 (currently 0 out of 267 complete)
Who would want to translate? Please follow the 🤗 [TRANSLATING guide](https://github.com/hug... | closed | not_planned | false | 4 | [
"WIP"
] | [] | 2024-04-14T11:45:12Z | 2026-07-17T15:32:35Z | 2026-07-17T15:32:35Z | NONE | null | 20260717T180032Z | 2026-07-17T18:00:32Z | RooTender | 24,989,290 | MDQ6VXNlcjI0OTg5Mjkw | User | false |
huggingface/transformers | 2,055,478,265 | I_kwDOCUB6oc56hBv5 | 28,238 | https://github.com/huggingface/transformers/issues/28238 | https://api.github.com/repos/huggingface/transformers/issues/28238 | Finetuning Whisper with adapter with MAML | ### Feature request
MAML is a widely used meta-learning method for reinitializing model parameters, which can effectively cope with low-resource situations.
As Whisper is a pre-trained model, its parameters cannot be reinitialized, but the bottleneck structure adapter can be applied to the encoder and decoder layers ... | closed | not_planned | false | 3 | [
"Feature request",
"Audio"
] | [] | 2023-12-25T07:43:42Z | 2026-07-17T17:46:47Z | 2026-07-17T17:46:47Z | NONE | null | 20260717T180032Z | 2026-07-17T18:00:32Z | LYPinASR | 112,866,899 | U_kgDOBro2Uw | User | false |
huggingface/transformers | 4,913,597,821 | I_kwDOCUB6oc8AAAABJN-NfQ | 47,388 | https://github.com/huggingface/transformers/issues/47388 | https://api.github.com/repos/huggingface/transformers/issues/47388 | [New model] PersonaPlex (full-duplex speech model, based on Moshi) | ### Model description
# [Moshi] Modernize for PersonaPlex: modular + processor + WeightConverter + #36928 fix
# What does this PR do?
**Our research team:** @b-re-w, @un1876, @jookite, @capyjin, @haesol-shin
We are using [NVIDIA PersonaPlex](https://huggingface.co/nvidia/personaplex-7b-v1) in an ongoing research pr... | open | null | false | 0 | [
"New model"
] | [] | 2026-07-17T17:48:39Z | 2026-07-17T17:48:53Z | null | NONE | null | 20260717T180032Z | 2026-07-17T18:00:32Z | un1876 | 105,284,212 | U_kgDOBkaCdA | User | false |
huggingface/transformers | 2,238,058,637 | I_kwDOCUB6oc6FZhCN | 30,191 | https://github.com/huggingface/transformers/issues/30191 | https://api.github.com/repos/huggingface/transformers/issues/30191 | Tranformers documentation translation to Persian | Hi!
Let's bring the documentation to all the Persian speaking community 🌐
Who would want to translate? Please follow the 🤗 [TRANSLATING guide](https://github.com/huggingface/transformers/blob/main/docs/TRANSLATING.md). Here is a list of the files ready for translation. Let us know in this issue if you'd like t... | closed | not_planned | false | 2 | [
"WIP"
] | [] | 2024-04-11T15:51:09Z | 2026-07-18T02:13:36Z | 2026-07-18T02:13:36Z | NONE | null | 20260718T060036Z | 2026-07-18T06:00:36Z | AmirHosseinMaleki | 40,566,488 | MDQ6VXNlcjQwNTY2NDg4 | User | false |
huggingface/transformers | 2,177,160,783 | I_kwDOCUB6oc6BxNZP | 29,553 | https://github.com/huggingface/transformers/issues/29553 | https://api.github.com/repos/huggingface/transformers/issues/29553 | Can we change mistral attention bias=False to bias=config.attention_bias ？ | just like llama attention 
| closed | not_planned | false | 6 | [
"Feature request"
] | [] | 2024-03-09T09:53:46Z | 2026-07-18T02:15:33Z | 2026-07-18T02:15:33Z | NONE | null | 20260718T060036Z | 2026-07-18T06:00:36Z | Minami-su | 76,865,636 | MDQ6VXNlcjc2ODY1NjM2 | User | false |
huggingface/transformers | 1,638,488,620 | I_kwDOCUB6oc5hqVos | 22,352 | https://github.com/huggingface/transformers/issues/22352 | https://api.github.com/repos/huggingface/transformers/issues/22352 | XVector Finetuning process - Whisper XVector | ### Model description
The idea is to apply XVector to Whisper and, In the process, generate documentation to Finetune or Adapt XVector (Maybe something similar to SetFit for Audio) @vaibva
### Open source status
- [ ] The model implementation is available
- [ ] The model weights are available
### Provide useful li... | closed | not_planned | false | 1 | [
"New model"
] | [] | 2023-03-23T23:34:38Z | 2026-07-18T02:16:42Z | 2026-07-18T02:16:42Z | NONE | null | 20260718T060036Z | 2026-07-18T06:00:36Z | rafael-ariascalles | 45,745,870 | MDQ6VXNlcjQ1NzQ1ODcw | User | false |
huggingface/transformers | 4,915,789,113 | I_kwDOCUB6oc8AAAABJQD9OQ | 47,390 | https://github.com/huggingface/transformers/issues/47390 | https://api.github.com/repos/huggingface/transformers/issues/47390 | [serge] integration failure triage - 2026-07-17 | <!-- serge-triage-run:integration-failure-triage:2026-07-17 -->
Automated **integration-failure triage** for the daily CI window `2026-07-09 → 2026-07-17`.
This issue was generated by AI-assisted automation. The grouping, summaries, and recommended follow-up can be incomplete or misleading; verify the failures before... | closed | not_planned | false | 1 | [] | [
"tarekziade"
] | 2026-07-18T00:25:02Z | 2026-07-19T00:26:53Z | 2026-07-19T00:26:53Z | CONTRIBUTOR | null | 20260719T060027Z | 2026-07-19T06:00:27Z | github-actions[bot] | 41,898,282 | MDM6Qm90NDE4OTgyODI= | Bot | false |
huggingface/transformers | 2,634,580,690 | I_kwDOCUB6oc6dCILS | 34,610 | https://github.com/huggingface/transformers/issues/34610 | https://api.github.com/repos/huggingface/transformers/issues/34610 | Saving logs to file | ### Feature request
Add a `FileHandler` to root logger.
### Motivation
Transformers is only using StreamHandler for logging, which makes debug analysis hard across different experiments. Many other frameworks support both logging to terminal and to file.
For example: fairseq: https://github.com/facebookresearch/fai... | closed | not_planned | false | 2 | [
"Feature request"
] | [] | 2024-11-05T06:50:03Z | 2026-07-18T02:40:59Z | 2026-07-18T02:40:59Z | CONTRIBUTOR | null | 20260718T060036Z | 2026-07-18T06:00:36Z | techkang | 22,956,582 | MDQ6VXNlcjIyOTU2NTgy | User | false |
huggingface/transformers | 4,917,408,637 | I_kwDOCUB6oc8AAAABJRmzfQ | 47,394 | https://github.com/huggingface/transformers/issues/47394 | https://api.github.com/repos/huggingface/transformers/issues/47394 | GitForCausalLM double-shifts the training loss and leaks targets across batch rows (regression from #35875) | ### System Info
- `transformers`: main (`7ea2320c76`), also present in every release since v4.49
- Code-level bug, weight-independent — reproducible on CPU, no GPU or pretrained weights needed
### Who can help?
@zucchini-nlp
### Description
`GitForCausalLM.forward` shifts the labels manually and then hands the res... | closed | completed | false | 0 | [] | [] | 2026-07-18T08:03:46Z | 2026-07-21T15:39:03Z | 2026-07-21T15:39:03Z | CONTRIBUTOR | null | 20260721T180022Z | 2026-07-21T18:00:22Z | Yigtwxx | 156,921,875 | U_kgDOCVpwEw | User | false |
huggingface/transformers | 4,917,570,385 | I_kwDOCUB6oc8AAAABJRwrUQ | 47,396 | https://github.com/huggingface/transformers/issues/47396 | https://api.github.com/repos/huggingface/transformers/issues/47396 | MPS: DynamicCache produces NaN logits mid-decode in generate() (transformers >=5.2, regression from 5.0) | ## Summary
On Apple Silicon (MPS), transformers **>= 5.2.0** produces NaN logits during autoregressive `generate()` with the default cache (DynamicCache), surfacing as:
```
RuntimeError: probability tensor contains either `inf`, `nan` or element < 0
File ".../transformers/generation/utils.py", line 2808, in _sample... | closed | completed | false | 0 | [] | [] | 2026-07-18T08:52:09Z | 2026-07-18T08:55:47Z | 2026-07-18T08:55:47Z | NONE | null | 20260718T120032Z | 2026-07-18T12:00:32Z | scp7 | 17,938,009 | MDQ6VXNlcjE3OTM4MDA5 | User | false |
huggingface/transformers | 4,920,046,350 | I_kwDOCUB6oc8AAAABJUHzDg | 47,402 | https://github.com/huggingface/transformers/issues/47402 | https://api.github.com/repos/huggingface/transformers/issues/47402 | [i18n-<languageCode>] Translating docs to <languageName> | Duplicate of #33260 <!--
Note: Please search to see if an issue already exists for the language you are trying to translate.
-->
Hi!
Let's bring the documentation to all the <languageName>-speaking community 🌐 (currently 0 out of 267 complete)
Who would want to translate? Please follow the 🤗 [TRANSLATING guide](ht... | closed | completed | false | 0 | [
"WIP"
] | [] | 2026-07-18T21:04:06Z | 2026-07-20T15:23:23Z | 2026-07-20T15:23:23Z | NONE | null | 20260721T000056Z | 2026-07-21T00:00:56Z | diplrcdi-cmd | 306,608,811 | U_kgDOEkZ6qw | User | false |
huggingface/transformers | 4,920,152,712 | I_kwDOCUB6oc8AAAABJUOSiA | 47,404 | https://github.com/huggingface/transformers/issues/47404 | https://api.github.com/repos/huggingface/transformers/issues/47404 | device_map loading OOMs on conversion-op transients: WeightConverter contiguous() runs on destination GPU, unbudgeted by max_memory packing | ### System Info
- `transformers` version: 5.14.1
- Platform: Linux-6.17.0-1007-aws-x86_64-with-glibc2.39
- Python version: 3.12.3
- Huggingface_hub version: 1.24.0
- Safetensors version: 0.8.0
- Accelerate version: 1.14.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?... | open | null | false | 0 | [
"bug"
] | [] | 2026-07-18T21:38:55Z | 2026-07-18T21:38:55Z | null | NONE | null | 20260719T000028Z | 2026-07-19T00:00:28Z | SelfSL | 90,419,525 | MDQ6VXNlcjkwNDE5NTI1 | User | false |
huggingface/transformers | 4,920,157,819 | I_kwDOCUB6oc8AAAABJUOmew | 47,405 | https://github.com/huggingface/transformers/issues/47405 | https://api.github.com/repos/huggingface/transformers/issues/47405 | Silent conversion skip on model_type mismatch: loading a converted checkpoint via the wrong Auto class yields randomly-initialized weights that generate fluently | ### System Info
- `transformers` version: 5.14.1
- Platform: Linux-6.17.0-1007-aws-x86_64-with-glibc2.39
- Python version: 3.12.3
- Huggingface_hub version: 1.24.0
- Safetensors version: 0.8.0
- Accelerate version: 1.14.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?... | open | null | false | 1 | [
"bug"
] | [] | 2026-07-18T21:40:19Z | 2026-07-20T10:12:49Z | null | NONE | null | 20260720T120037Z | 2026-07-20T12:00:37Z | SelfSL | 90,419,525 | MDQ6VXNlcjkwNDE5NTI1 | User | false |
huggingface/transformers | 4,920,808,008 | I_kwDOCUB6oc8AAAABJU2SSA | 47,407 | https://github.com/huggingface/transformers/issues/47407 | https://api.github.com/repos/huggingface/transformers/issues/47407 | CompressedTensorsConfig(run_compressed=False) silently random-initializes unmapped fused-MoE expert weights on 5.10.x (works on 5.13.1) — request hard error for unmapped packed tensors | **Summary**: Loading a compressed-tensors W4A16 checkpoint of `Qwen/Qwen3.6-35B-A3B` (fused-expert MoE) with `from_pretrained(ckpt, quantization_config=CompressedTensorsConfig(run_compressed=False))` on **transformers 5.10.1** prints:
```
model.language_model.layers.{0...39}.mlp.experts.gate_up_proj_packed | UNEXPECTE... | open | null | false | 2 | [] | [] | 2026-07-19T01:30:29Z | 2026-07-20T09:39:06Z | null | NONE | null | 20260720T120037Z | 2026-07-20T12:00:37Z | narfian | 42,402,856 | MDQ6VXNlcjQyNDAyODU2 | User | false |
huggingface/transformers | 4,920,636,609 | I_kwDOCUB6oc8AAAABJUr0wQ | 47,406 | https://github.com/huggingface/transformers/issues/47406 | https://api.github.com/repos/huggingface/transformers/issues/47406 | [serge] integration failure triage - 2026-07-18 | <!-- serge-triage-run:integration-failure-triage:2026-07-18 -->
Automated **integration-failure triage** for the daily CI window `2026-07-10 → 2026-07-18`.
This issue was generated by AI-assisted automation. The grouping, summaries, and recommended follow-up can be incomplete or misleading; verify the failures before... | closed | not_planned | false | 1 | [] | [
"tarekziade"
] | 2026-07-19T00:26:37Z | 2026-07-20T02:30:54Z | 2026-07-20T02:30:54Z | CONTRIBUTOR | null | 20260720T060031Z | 2026-07-20T06:00:31Z | github-actions[bot] | 41,898,282 | MDM6Qm90NDE4OTgyODI= | Bot | false |
huggingface/transformers | 2,068,828,967 | I_kwDOCUB6oc57T9Mn | 28,370 | https://github.com/huggingface/transformers/issues/28370 | https://api.github.com/repos/huggingface/transformers/issues/28370 | Missing `vocab_file` Attribute When Using Custom SentencePiece Models | ### System Info
- `transformers` version: 4.36.2
- Platform: Linux-6.1.69-1-lts-x86_64-with-glibc2.38
- Python version: 3.11.6
- Huggingface_hub version: 0.19.4
- Safetensors version: 0.4.0
- Accelerate version: not installed
- Accelerate config: not found
- PyTorch version (GPU?): 2.1.2+cpu (False)
- Tensor... | closed | completed | false | 7 | [
"Feature request",
"Good Difficult Issue"
] | [] | 2024-01-06T20:43:07Z | 2026-07-19T17:56:36Z | 2026-07-19T17:56:35Z | NONE | null | 20260719T180031Z | 2026-07-19T18:00:31Z | teleprint-me | 77,757,836 | MDQ6VXNlcjc3NzU3ODM2 | User | false |
huggingface/transformers | 2,212,350,240 | I_kwDOCUB6oc6D3ckg | 29,925 | https://github.com/huggingface/transformers/issues/29925 | https://api.github.com/repos/huggingface/transformers/issues/29925 | SigLIP tokenizer not enforcing use_fast=True | ### System Info
- `transformers` version: 4.38.2
- Platform: Linux-4.18.0-477.27.1.el8_8.x86_64-x86_64-with-glibc2.28
- Python version: 3.10.13
- Huggingface_hub version: 0.21.4
- Safetensors version: 0.4.2
- Accelerate version: not installed
- Accelerate config: not found
- PyTorch version (GPU?): 2.2.1+cu121 ... | open | null | false | 7 | [
"Feature request",
"Good Difficult Issue"
] | [] | 2024-03-28T04:56:26Z | 2026-07-21T12:12:28Z | null | NONE | null | 20260721T180022Z | 2026-07-21T18:00:22Z | yxchng | 10,518,587 | MDQ6VXNlcjEwNTE4NTg3 | User | false |
huggingface/transformers | 1,856,207,291 | I_kwDOCUB6oc5uo3m7 | 25,586 | https://github.com/huggingface/transformers/issues/25586 | https://api.github.com/repos/huggingface/transformers/issues/25586 | SwahBERT | ### Model description
<html>
<body>
<!--StartFragment--><h1 tabindex="-1" dir="auto" style="box-sizing: border-box; font-size: 2em; margin-top: 0px !important; margin-right: 0px; margin-bottom: 16px; margin-left: 0px; font-weight: var(--base-text-weight-semibold, 600); line-height: 1.25; padding-bottom: 0.3em; borde... | closed | completed | false | 2 | [
"New model"
] | [] | 2023-08-18T07:25:56Z | 2026-07-19T18:00:46Z | 2026-07-19T18:00:46Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | Adilatec | 130,484,182 | U_kgDOB8cH1g | User | false |
huggingface/transformers | 1,422,083,233 | I_kwDOCUB6oc5Uw0Sh | 19,865 | https://github.com/huggingface/transformers/issues/19865 | https://api.github.com/repos/huggingface/transformers/issues/19865 | Add VATT model | ### Model description
Hey,
as discussed with @NielsRogge a few weeks back, I'd like to work on adding the "VATT: Transformers for Multimodal
Self-Supervised Learning from Raw Video, Audio and Text" model from Google.
It is basically three transformers(Video/Audio/Text) that are trained jointly in an unsupervised ... | closed | not_planned | false | 9 | [
"New model"
] | [] | 2022-10-25T08:36:37Z | 2026-07-19T18:02:23Z | 2026-07-19T18:02:22Z | CONTRIBUTOR | null | 20260720T060031Z | 2026-07-20T06:00:31Z | johko | 2,843,485 | MDQ6VXNlcjI4NDM0ODU= | User | false |
huggingface/transformers | 1,949,588,386 | I_kwDOCUB6oc50NFui | 26,901 | https://github.com/huggingface/transformers/issues/26901 | https://api.github.com/repos/huggingface/transformers/issues/26901 | Add quantization_config in AutoModelForCausalLM.from_config() | ### Feature request
Add quantization_config feature to AutoModelForCausalLM from config .
I am trying to pretrain a model from scratch and use bits and bytes so that It can be trained on less computation expensive machines.
Below is my quantization config :
```
bnb_config = BitsAndBytesConfig(
load_in_4b... | closed | not_planned | false | 16 | [
"Feature request",
"Quantization"
] | [] | 2023-10-18T12:07:54Z | 2026-07-19T18:04:22Z | 2026-07-19T18:04:22Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | ishaansharma | 8,963,395 | MDQ6VXNlcjg5NjMzOTU= | User | false |
huggingface/transformers | 2,249,869,273 | I_kwDOCUB6oc6GGkfZ | 30,308 | https://github.com/huggingface/transformers/issues/30308 | https://api.github.com/repos/huggingface/transformers/issues/30308 | Badam support | ### Feature request
https://arxiv.org/pdf/2404.02827.pdf
Memory efficient optimizer like galore but with less hyperparameters
### Motivation
https://github.com/hiyouga/LLaMA-Factory/pull/3287
LlamaFactory supports it, but having it in transformers is much more convenient
### Your contribution
- | closed | not_planned | false | 3 | [
"Feature request",
"optimization"
] | [] | 2024-04-18T06:36:22Z | 2026-07-19T18:05:36Z | 2026-07-19T18:05:35Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | Theodotus1243 | 32,220,358 | MDQ6VXNlcjMyMjIwMzU4 | User | false |
huggingface/transformers | 3,340,056,210 | I_kwDOCUB6oc7HFTaS | 40,330 | https://github.com/huggingface/transformers/issues/40330 | https://api.github.com/repos/huggingface/transformers/issues/40330 | open-qwen2vl-base | ### Model description
is there any plan to add open-qwen2vl-base model?
### Open source status
- [x] The model implementation is available
- [x] The model weights are available
### Provide useful links for the implementation
_No response_ | closed | not_planned | false | 6 | [
"New model"
] | [] | 2025-08-21T02:24:01Z | 2026-07-19T18:07:16Z | 2026-07-19T18:07:16Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | olccihyeon | 36,918,246 | MDQ6VXNlcjM2OTE4MjQ2 | User | false |
huggingface/transformers | 1,456,554,082 | I_kwDOCUB6oc5W0UBi | 20,334 | https://github.com/huggingface/transformers/issues/20334 | https://api.github.com/repos/huggingface/transformers/issues/20334 | longformer_content | ### Model description
This model will generate a content score for a summary given the context and the summary itself.
### Open source status
- [X] The model implementation is available
- [X] The model weights are available
### Provide useful links for the implementation
_No response_ | closed | completed | false | 2 | [
"New model"
] | [] | 2022-11-19T15:57:10Z | 2026-07-19T18:08:07Z | 2026-07-19T18:08:07Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | wesleygmorris | 82,518,780 | MDQ6VXNlcjgyNTE4Nzgw | User | false |
huggingface/transformers | 2,358,743,023 | I_kwDOCUB6oc6Ml4_v | 31,462 | https://github.com/huggingface/transformers/issues/31462 | https://api.github.com/repos/huggingface/transformers/issues/31462 | A parameter in TrainingArguments: sample_output=True | ### Feature request
It would be nice to have the model output some sample generations while it is training, so that I can see if my training is going well, and at what point I should stop it.
### Motivation
For example let's say I set my training for 5 epochs. I see the samples from the first epoch are bad, the sec... | closed | not_planned | false | 4 | [
"Feature request"
] | [] | 2024-06-18T02:45:41Z | 2026-07-19T18:09:34Z | 2026-07-19T18:09:33Z | CONTRIBUTOR | null | 20260720T060031Z | 2026-07-20T06:00:31Z | Tylersuard | 41,713,505 | MDQ6VXNlcjQxNzEzNTA1 | User | false |
huggingface/transformers | 2,309,436,533 | I_kwDOCUB6oc6JpzR1 | 30,947 | https://github.com/huggingface/transformers/issues/30947 | https://api.github.com/repos/huggingface/transformers/issues/30947 | [LLaMA3] 'add_bos_token=True, add_eos_token=True' seems not taking effect | ### System Info
Platform = Windows
PyTorch = 2.3.0
Transformers = 4.41.0
### 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, ...)
- [x] My own ... | closed | completed | false | 7 | [
"Core: Tokenization",
"Feature request"
] | [] | 2024-05-22T02:22:34Z | 2026-07-19T18:11:19Z | 2026-07-19T18:11:13Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | kiva12138 | 40,737,639 | MDQ6VXNlcjQwNzM3NjM5 | User | false |
huggingface/transformers | 2,550,722,098 | I_kwDOCUB6oc6YCO4y | 33,728 | https://github.com/huggingface/transformers/issues/33728 | https://api.github.com/repos/huggingface/transformers/issues/33728 | Will Trainer.predict() return data in the same order as the original dataset during multi-machine and multi-gpus inference? | ### Feature request
I want to use accelerate for multi-machine and multi-gpus inference. Since using trainer.predict does not return the original inference data, only the inference results, I am not sure if their order can still be maintained in the case of multi-machine and multi-gpus.
### Motivation
I want to use ... | closed | completed | false | 4 | [
"trainer",
"Feature request",
"Accelerate"
] | [] | 2024-09-26T14:12:32Z | 2026-07-19T18:15:27Z | 2026-07-19T18:12:49Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | deepwebney | 49,145,634 | MDQ6VXNlcjQ5MTQ1NjM0 | User | false |
huggingface/transformers | 925,627,493 | MDU6SXNzdWU5MjU2Mjc0OTM= | 12,273 | https://github.com/huggingface/transformers/issues/12273 | https://api.github.com/repos/huggingface/transformers/issues/12273 | [Deepspeed] [performance] inefficient load with `from_pretrained` w/ zero3 | # 🚀 Feature request
Currently under Deepspeed stage3 with `from_pretrained` we:
a. loop over each sub-module in zero.Init
1. init the sub-module
2. shard and scatter the shards
b. then to load pre-trained weights we loop over each sub-module:
1. gather the shards
2. `load_state_dict` for the one layer lay... | closed | not_planned | false | 2 | [
"DeepSpeed",
"WIP"
] | [
"stas00"
] | 2021-06-20T15:49:30Z | 2026-07-20T15:55:51Z | 2026-07-19T18:16:33Z | CONTRIBUTOR | null | 20260721T000056Z | 2026-07-21T00:00:56Z | stas00 | 10,676,103 | MDQ6VXNlcjEwNjc2MTAz | User | false |
huggingface/transformers | 1,719,258,680 | I_kwDOCUB6oc5mec44 | 23,647 | https://github.com/huggingface/transformers/issues/23647 | https://api.github.com/repos/huggingface/transformers/issues/23647 | Wandb sweeps integraition: custom objective | ### System Info
- `transformers` version: 4.29.2
- Platform: Linux-5.14.0-162.6.1.el9_1.0.1.x86_64-x86_64-with-glibc2.34
- Python version: 3.10.10
- Huggingface_hub version: 0.14.1
- Safetensors version: not installed
- PyTorch version (GPU?): 2.0.1+cu117 (True) -- using torch in my experiments
- Tensorflow vers... | closed | completed | false | 18 | [
"WIP",
"Integrations"
] | [] | 2023-05-22T09:11:55Z | 2026-07-19T18:18:46Z | 2026-07-19T18:18:29Z | COLLABORATOR | null | 20260720T060031Z | 2026-07-20T06:00:31Z | BramVanroy | 2,779,410 | MDQ6VXNlcjI3Nzk0MTA= | User | false |
huggingface/transformers | 1,246,769,470 | I_kwDOCUB6oc5KUDE- | 17,399 | https://github.com/huggingface/transformers/issues/17399 | https://api.github.com/repos/huggingface/transformers/issues/17399 | [RFC] Scan & Gradient checkpointing in Flax | ### Feature request
We should add scan and remat (gradient checkpointing) to the most important Flax/JAX models (BERT, GPT2, OPT, T5, BART, Wav2Vec2).
### Motivation
Scan allows for much faster compilation and memory savings and `remat` is the equivalent of `gradient_checkpointing` in PyTorch.
@sanchit-gandhi alr... | closed | not_planned | false | 6 | [
"WIP"
] | [] | 2022-05-24T15:53:36Z | 2026-07-19T18:21:16Z | 2026-07-19T18:21:16Z | CONTRIBUTOR | null | 20260720T060031Z | 2026-07-20T06:00:31Z | patrickvonplaten | 23,423,619 | MDQ6VXNlcjIzNDIzNjE5 | User | false |
huggingface/transformers | 1,766,630,109 | I_kwDOCUB6oc5pTKLd | 24,392 | https://github.com/huggingface/transformers/issues/24392 | https://api.github.com/repos/huggingface/transformers/issues/24392 | Allow `TextClassificationPipeline` to handle input longer than `model_max_length` tokens | ### Problem
Running a `TextClassificationPipeline` on a text with more tokens than its model's maximum position embeddings (e.g. 512 for BERT) like so:
```python
from transformers import pipeline
classifier = pipeline('sentiment-analysis')
classifier("Hello, world! " * 1000)
```
will lead to this error:
... | closed | not_planned | false | 6 | [
"Feature request"
] | [] | 2023-06-21T03:51:47Z | 2026-07-19T18:23:42Z | 2026-07-19T18:23:42Z | CONTRIBUTOR | null | 20260720T060031Z | 2026-07-20T06:00:31Z | boyleconnor | 6,520,892 | MDQ6VXNlcjY1MjA4OTI= | User | false |
huggingface/transformers | 2,207,875,036 | I_kwDOCUB6oc6DmX_c | 29,875 | https://github.com/huggingface/transformers/issues/29875 | https://api.github.com/repos/huggingface/transformers/issues/29875 | Added HelpingAI model type in it | ### Model description
Introduction
HelpingAI-3B is a state-of-the-art AI model designed to assist with day-to-day tasks. It's trained on a diverse range of datasets, making it versatile and adaptable to various applications.
Model Overview
HelpingAI-3B is the latest model in the HelpingAI series. It's built on ... | closed | not_planned | false | 2 | [
"New model"
] | [] | 2024-03-26T10:38:54Z | 2026-07-19T22:59:38Z | 2026-07-19T22:59:38Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | OEvortex | 158,988,478 | U_kgDOCXn4vg | User | false |
huggingface/transformers | 2,256,742,258 | I_kwDOCUB6oc6Ggydy | 30,396 | https://github.com/huggingface/transformers/issues/30396 | https://api.github.com/repos/huggingface/transformers/issues/30396 | Add HelpingAI-3B-v2.2: Emotionally Intelligent Conversational AI | ### Model description
## Introduction
HelpingAI-3B-v2.2 is a state-of-the-art large language model specializing in emotionally intelligent conversation. With advanced emotional understanding capabilities, it can engage in empathetic dialogue tailored to the user's emotional state and context.
## Emotional Inte... | closed | not_planned | false | 1 | [
"New model"
] | [] | 2024-04-22T14:49:29Z | 2026-07-19T23:00:05Z | 2026-07-19T23:00:05Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | OEvortex | 158,988,478 | U_kgDOCXn4vg | User | false |
huggingface/transformers | 2,322,561,054 | I_kwDOCUB6oc6Kb3ge | 31,103 | https://github.com/huggingface/transformers/issues/31103 | https://api.github.com/repos/huggingface/transformers/issues/31103 | Add HelpingAIForCausalLM to it | ### Model description
HelpingAI is a state-of-the-art large language model specializing in emotionally intelligent conversation. With advanced emotional understanding capabilities, it can engage in empathetic dialogue tailored to the user's emotional state and context.
## Emotional Intelligence Capabilities
He... | closed | duplicate | false | 8 | [
"New model"
] | [] | 2024-05-29T07:19:36Z | 2026-07-19T23:01:48Z | 2026-07-19T23:01:48Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | OEvortex | 158,988,478 | U_kgDOCXn4vg | User | false |
huggingface/transformers | 1,864,901,414 | I_kwDOCUB6oc5vKCMm | 25,725 | https://github.com/huggingface/transformers/issues/25725 | https://api.github.com/repos/huggingface/transformers/issues/25725 | Meta-Transformer | ### Model description
Introducing a groundbreaking research paper that explores the potential of unified multimodal learning, revolutionizing the way we process and integrate diverse data types such as text, images, audio, and more! 🌟
[meta-transformer](https://arxiv.org/pdf/2307.10802.pdf)
### Open source status
... | closed | not_planned | false | 9 | [
"New model"
] | [] | 2023-08-24T10:53:42Z | 2026-07-20T01:48:26Z | 2026-07-19T23:02:53Z | CONTRIBUTOR | null | 20260720T060031Z | 2026-07-20T06:00:31Z | rajveer43 | 64,583,161 | MDQ6VXNlcjY0NTgzMTYx | User | false |
huggingface/transformers | 4,925,290,639 | I_kwDOCUB6oc8AAAABJZH4jw | 47,419 | https://github.com/huggingface/transformers/issues/47419 | https://api.github.com/repos/huggingface/transformers/issues/47419 | [serge] integration failure triage - 2026-07-19 | <!-- serge-triage-run:integration-failure-triage:2026-07-19 -->
Automated **integration-failure triage** for the daily CI window `2026-07-11 → 2026-07-19`.
This issue was generated by AI-assisted automation. The grouping, summaries, and recommended follow-up can be incomplete or misleading; verify the failures before... | closed | not_planned | false | 1 | [] | [
"tarekziade"
] | 2026-07-20T02:30:37Z | 2026-07-21T00:27:36Z | 2026-07-21T00:27:36Z | CONTRIBUTOR | null | 20260721T060018Z | 2026-07-21T06:00:18Z | github-actions[bot] | 41,898,282 | MDM6Qm90NDE4OTgyODI= | Bot | false |
huggingface/transformers | 1,853,507,992 | I_kwDOCUB6oc5uekmY | 25,546 | https://github.com/huggingface/transformers/issues/25546 | https://api.github.com/repos/huggingface/transformers/issues/25546 | Andromeda | ### Model description
## **Andromeda Specs**: Unveiling Mastery
**Overview**
Elegantly marrying craftsmanship and technology, Andromeda is not just another step in AI evolution. It's a giant leap. Driven by precision, powered by innovation, and defined by excellence, Andromeda is the epitome of intelligence realiz... | closed | not_planned | false | 1 | [
"New model"
] | [] | 2023-08-16T15:50:06Z | 2026-07-20T04:34:01Z | 2026-07-20T04:34:01Z | NONE | null | 20260720T060031Z | 2026-07-20T06:00:31Z | kyegomez | 98,760,976 | U_kgDOBeL5EA | User | false |
huggingface/transformers | 4,926,575,614 | I_kwDOCUB6oc8AAAABJaWT_g | 47,424 | https://github.com/huggingface/transformers/issues/47424 | https://api.github.com/repos/huggingface/transformers/issues/47424 | Several slow tokenizers read all_special_tokens or all_special_ids once per token in decode loops | ### Who can help?
@itazap
### Summary
Same pattern as #46315 and #46396, third instance. `all_special_tokens` and `all_special_ids` are properties that rebuild their list on every access (`all_special_ids` additionally calls `convert_tokens_to_ids` on each special token). The base class knows this and hoists the loo... | closed | completed | false | 0 | [] | [] | 2026-07-20T07:02:09Z | 2026-07-20T15:56:54Z | 2026-07-20T15:56:54Z | CONTRIBUTOR | null | 20260721T000056Z | 2026-07-21T00:00:56Z | ishan-1010 | 98,383,932 | U_kgDOBd04PA | User | false |
huggingface/transformers | 4,927,152,262 | I_kwDOCUB6oc8AAAABJa5ghg | 47,427 | https://github.com/huggingface/transformers/issues/47427 | https://api.github.com/repos/huggingface/transformers/issues/47427 | `_is_hf_initialized` per-param check doesn't skip redundant init for built-in models on non-CUDA hardware | ### System Info
- `transformers` version: 4.57.6
- Platform: Ascend NPU 910B, CANN 8.5.0
- Python version: 3.11
- PyTorch version: 2.9.0, torch_npu 2.9.0
- Hardware: 16 × Ascend 910B, FSDP2
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
... | open | null | false | 0 | [
"bug"
] | [] | 2026-07-20T08:28:45Z | 2026-07-20T17:13:39Z | null | NONE | null | 20260721T000056Z | 2026-07-21T00:00:56Z | gygdh-001 | 43,057,576 | MDQ6VXNlcjQzMDU3NTc2 | User | false |
huggingface/transformers | 941,593,275 | MDU6SXNzdWU5NDE1OTMyNzU= | 12,635 | https://github.com/huggingface/transformers/issues/12635 | https://api.github.com/repos/huggingface/transformers/issues/12635 | Long-Short Transformer | # 🌟 New model addition
## Model description
https://arxiv.org/abs/2107.02192
In this paper, they propose Long-Short Transformer, an efficient self-attention mechanism for modeling long sequences with linear complexity for both language and vision tasks. It aggregates a novel long-range attention with dynamic ... | closed | not_planned | false | 6 | [
"New model"
] | [] | 2021-07-12T01:47:30Z | 2026-07-20T14:42:22Z | 2026-07-20T14:42:22Z | CONTRIBUTOR | null | 20260721T000056Z | 2026-07-21T00:00:56Z | schmidek | 442,328 | MDQ6VXNlcjQ0MjMyOA== | User | false |
huggingface/transformers | 1,872,110,603 | I_kwDOCUB6oc5vliQL | 25,839 | https://github.com/huggingface/transformers/issues/25839 | https://api.github.com/repos/huggingface/transformers/issues/25839 | Add D_Nikud model | ### Model description
Model description
The repository is dedicated to the implementation of our innovative D-Nikud model, which use the TavBERT architecture and Bi-LSTM to predict and apply diacritics (nikud) to Hebrew text. Diacritics play a crucial role in accurately conveying pronunciation and interpretation, m... | closed | not_planned | false | 11 | [
"New model"
] | [] | 2023-08-29T17:30:40Z | 2026-07-20T14:48:43Z | 2026-07-20T14:48:43Z | NONE | null | 20260721T000056Z | 2026-07-21T00:00:56Z | NadavShaked | 19,892,483 | MDQ6VXNlcjE5ODkyNDgz | User | false |
huggingface/transformers | 4,929,706,155 | I_kwDOCUB6oc8AAAABJdVYqw | 47,429 | https://github.com/huggingface/transformers/issues/47429 | https://api.github.com/repos/huggingface/transformers/issues/47429 | tokenizers<=0.23.0 upper bound blocks the only existing 0.23.x release | ### System Info
tokenizers==0.23.0 was never published — the tokenizers team skipped straight to 0.23.1. This means transformers' metadata constraint tokenizers>=0.22.0,<=0.23.0 effectively blocks 0.23.1, the only 0.23.x release that exists.
This is a security issue in practice: 0.23.1 contains the fix for an XSS vul... | open | null | false | 3 | [
"bug"
] | [] | 2026-07-20T14:03:10Z | 2026-07-21T12:39:15Z | null | NONE | null | 20260721T180022Z | 2026-07-21T18:00:22Z | EfiSiapiti | 82,284,420 | MDQ6VXNlcjgyMjg0NDIw | User | false |
huggingface/transformers | 937,864,452 | MDU6SXNzdWU5Mzc4NjQ0NTI= | 12,534 | https://github.com/huggingface/transformers/issues/12534 | https://api.github.com/repos/huggingface/transformers/issues/12534 | [Flax] from_pretrained does not consider the passed dtype | ## Environment info
When loading a flax model with `from_pretrained` the type argument is not used. The weights are initialized with the dtype of saved weights.
So if you do
```python
model = FlaxGPT2ForCausalLM.from_pretrained("gpt2", dtype=jnp.dtype("bfloat16"))
# check the dtype of one of the params
mo... | closed | not_planned | false | 8 | [
"WIP"
] | [
"patil-suraj"
] | 2021-07-06T12:51:26Z | 2026-07-20T15:31:12Z | 2026-07-20T15:31:12Z | CONTRIBUTOR | null | 20260721T000056Z | 2026-07-21T00:00:56Z | patil-suraj | 27,137,566 | MDQ6VXNlcjI3MTM3NTY2 | User | false |
huggingface/transformers | 4,931,044,012 | I_kwDOCUB6oc8AAAABJenCrA | 47,433 | https://github.com/huggingface/transformers/issues/47433 | https://api.github.com/repos/huggingface/transformers/issues/47433 | DynamicCache.crop(0) retains the full KV cache and oversized negative crops retain stale tokens | # `DynamicCache.crop(0)` retains the full KV cache and oversized negative crops retain stale tokens
## System Info
- Transformers source revision: `71554973d2c3123a23f2a1d6b7b63568b4e4df62`
- OS: Windows 10 (`10.0.26200`)
- Python: 3.11.9
- PyTorch: 2.13.0+cpu
## Who can help?
Generation/cache maintainer... | open | null | false | 4 | [] | [] | 2026-07-20T16:43:35Z | 2026-07-22T07:43:58Z | null | NONE | null | 20260722T120022Z | 2026-07-22T12:00:22Z | HoneyTyagii | 78,690,656 | MDQ6VXNlcjc4NjkwNjU2 | User | false |
huggingface/transformers | 2,906,128,386 | I_kwDOCUB6oc6tOAAC | 36,624 | https://github.com/huggingface/transformers/issues/36624 | https://api.github.com/repos/huggingface/transformers/issues/36624 | [deepspeed] any plans for deepspeed-domino? | ### Feature request
## deepspeed-domino
[sample code](https://github.com/deepspeedai/DeepSpeedExamples/blob/master/training/DeepSpeed-Domino/pretrain_llama.py)
### Motivation
deepspeed released a plog post about deepspeed-domino, which is designed to significantly reduce communication burdens btw nodes[(link)](https... | closed | not_planned | false | 2 | [
"Feature request"
] | [] | 2025-03-10T05:30:33Z | 2026-07-20T17:37:45Z | 2026-07-20T17:37:45Z | NONE | null | 20260721T000056Z | 2026-07-21T00:00:56Z | SoundProvider | 48,939,336 | MDQ6VXNlcjQ4OTM5MzM2 | User | false |
huggingface/transformers | 4,931,420,119 | I_kwDOCUB6oc8AAAABJe9_1w | 47,436 | https://github.com/huggingface/transformers/issues/47436 | https://api.github.com/repos/huggingface/transformers/issues/47436 | AttributeError: 'Qwen2_5OmniConfig' object has no attribute 'max_position_embeddings' | ### System Info
File "/venv/main/lib/python3.12/site-packages/transformers/modeling_utils.py", line 3443, in save_pretrained
model_to_save.config.save_pretrained(save_directory)
File "/venv/main/lib/python3.12/site-packages/transformers/configuration_utils.py", line 545, in save_pretrained
self.validate()
... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-07-20T17:31:25Z | 2026-07-20T17:39:47Z | 2026-07-20T17:39:14Z | NONE | null | 20260721T000056Z | 2026-07-21T00:00:56Z | VietDunghacker | 29,811,693 | MDQ6VXNlcjI5ODExNjkz | User | false |
huggingface/transformers | 4,932,810,785 | I_kwDOCUB6oc8AAAABJgS4IQ | 47,439 | https://github.com/huggingface/transformers/issues/47439 | https://api.github.com/repos/huggingface/transformers/issues/47439 | Added-token encoder cache goes stale when tokenizers mutate _added_tokens_decoder directly, giving ids that do not round-trip | ### Who can help?
@itazap
### Summary
Slow tokenizers keep added tokens in two structures that have to be kept in sync by hand: `_added_tokens_decoder` (the source of truth) and `_added_tokens_encoder` (a lookup cache, on the hot path since #46323). Five model tokenizers mutate `_added_tokens_decoder` directly after... | open | null | false | 0 | [] | [] | 2026-07-20T20:28:42Z | 2026-07-20T20:28:42Z | null | CONTRIBUTOR | null | 20260721T000056Z | 2026-07-21T00:00:56Z | ishan-1010 | 98,383,932 | U_kgDOBd04PA | User | false |
huggingface/transformers | 1,942,164,704 | I_kwDOCUB6oc5zwxTg | 26,794 | https://github.com/huggingface/transformers/issues/26794 | https://api.github.com/repos/huggingface/transformers/issues/26794 | [i18n-<kan>] Translating docs to <Kannada> | <!--
Note: Please search to see if an issue already exists for the language you are trying to translate.
-->
Hi!
Let's bring the documentation to all the <Kannada>-speaking community 🌐 (currently 0 out of 267 complete)
Who would want to translate? Please follow the 🤗 [TRANSLATING guide](https://github.com/... | closed | not_planned | false | 2 | [
"WIP"
] | [] | 2023-10-13T15:25:06Z | 2026-07-20T20:30:49Z | 2026-07-20T20:30:49Z | NONE | null | 20260721T000056Z | 2026-07-21T00:00:56Z | praj-pawar | 99,607,337 | U_kgDOBe_jKQ | User | false |
huggingface/transformers | 3,025,348,585 | I_kwDOCUB6oc60Uyfp | 37,835 | https://github.com/huggingface/transformers/issues/37835 | https://api.github.com/repos/huggingface/transformers/issues/37835 | Add HindiCausalLM: A specialized Hindi language model (~102M parameters) | ### Model description
HindiCausalLM model, a causal language model specifically designed for Hindi text generation. The model was developed by ConvaiInnovations and the original implementation is available on the [Hugging Face Hub](https://huggingface.co/convaiinnovations/hindi-causal-lm).
HindiCausalLM addresses the... | closed | not_planned | false | 2 | [
"New model"
] | [] | 2025-04-28T15:30:32Z | 2026-07-21T00:08:51Z | 2026-07-21T00:07:30Z | NONE | null | 20260721T060018Z | 2026-07-21T06:00:18Z | NandhaKishorM | 48,623,612 | MDQ6VXNlcjQ4NjIzNjEy | User | false |
huggingface/transformers | 1,331,914,135 | I_kwDOCUB6oc5PY2WX | 18,525 | https://github.com/huggingface/transformers/issues/18525 | https://api.github.com/repos/huggingface/transformers/issues/18525 | [Summary] Regarding memory issue in tests | ### Description
This is a short summary of the memory issue in our tests
### The following tests definitely have memory issues
- PyTorch (increase ~`15 MB` each call):
- test_torch_fx
- test_torch_fx_output_loss
- TensorFlow:
- test_xla_fit
- test_xla_generate_fast (increase ~`100 MB` each call)
... | closed | completed | false | 10 | [
"WIP",
"bug"
] | [
"ydshieh"
] | 2022-08-08T13:56:20Z | 2026-07-21T00:10:00Z | 2026-07-21T00:10:00Z | MEMBER | null | 20260721T060018Z | 2026-07-21T06:00:18Z | ydshieh | 2,521,628 | MDQ6VXNlcjI1MjE2Mjg= | User | false |
huggingface/transformers | 4,934,170,088 | I_kwDOCUB6oc8AAAABJhl16A | 47,442 | https://github.com/huggingface/transformers/issues/47442 | https://api.github.com/repos/huggingface/transformers/issues/47442 | [serge] integration failure triage - 2026-07-20 | <!-- serge-triage-run:integration-failure-triage:2026-07-20 -->
Automated **integration-failure triage** for the daily CI window `2026-07-12 → 2026-07-20`.
This issue was generated by AI-assisted automation. The grouping, summaries, and recommended follow-up can be incomplete or misleading; verify the failures before... | closed | not_planned | false | 1 | [] | [
"tarekziade"
] | 2026-07-21T00:27:19Z | 2026-07-22T00:25:37Z | 2026-07-22T00:25:37Z | CONTRIBUTOR | null | 20260722T060034Z | 2026-07-22T06:00:34Z | github-actions[bot] | 41,898,282 | MDM6Qm90NDE4OTgyODI= | Bot | false |
huggingface/transformers | 2,004,233,116 | I_kwDOCUB6oc53diuc | 27,631 | https://github.com/huggingface/transformers/issues/27631 | https://api.github.com/repos/huggingface/transformers/issues/27631 | Add the learning rate (in exponential representation, like "9.197948717948718e-08") to the model training output | ### Feature request
Good day. Can you add current learning rate to output of training process.
Now, if we train model with Trainer we get those output (without leraning rate):
<img width="302" alt="image" src="https://github.com/huggingface/transformers/assets/17560478/239c3ce1-c7dc-4d79-9664-52ac4e3ea605">
... | open | null | false | 13 | [
"Feature request"
] | [
"muellerzr"
] | 2023-11-21T12:58:15Z | 2026-07-21T02:43:35Z | null | NONE | null | 20260721T060018Z | 2026-07-21T06:00:18Z | artyomboyko | 17,560,478 | MDQ6VXNlcjE3NTYwNDc4 | User | false |
huggingface/transformers | 4,936,009,634 | I_kwDOCUB6oc8AAAABJjWHog | 47,448 | https://github.com/huggingface/transformers/issues/47448 | https://api.github.com/repos/huggingface/transformers/issues/47448 | gemma4 12B arch is not recognized! | ### System Info
### i had tried transformers==[5.11.0, 5.13.0, 5.14.1, 5.14.2, 5.15.0.dev] (`pip install git+transformer_url.git`)
- `transformers` version: 5.15.0.dev0
- Platform: Linux-6.12.90+-x86_64-with-glibc2.35
- Python version: 3.12.13
- Huggingface_hub version: 1.11.0
- Safetensors version: 0.8.0
- Accelerat... | open | null | false | 2 | [
"bug"
] | [] | 2026-07-21T06:21:09Z | 2026-07-21T09:54:56Z | null | NONE | null | 20260721T120029Z | 2026-07-21T12:00:29Z | HuangChiEn | 52,521,165 | MDQ6VXNlcjUyNTIxMTY1 | User | false |
huggingface/transformers | 4,938,874,710 | I_kwDOCUB6oc8AAAABJmE_Vg | 47,455 | https://github.com/huggingface/transformers/issues/47455 | https://api.github.com/repos/huggingface/transformers/issues/47455 | The kernel's version doesn't properly check. | ### System Info
If the kernels version is not 0.15.2 and we want to use MXFP4, a warning will print, which implies the kernels version must be greater than 0.12.0.
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially suppor... | open | null | false | 2 | [
"bug"
] | [] | 2026-07-21T12:33:46Z | 2026-07-21T19:10:52Z | null | NONE | null | 20260722T000038Z | 2026-07-22T00:00:38Z | Mohsen-Liaghat | 81,348,397 | MDQ6VXNlcjgxMzQ4Mzk3 | User | false |
huggingface/transformers | 3,293,398,290 | I_kwDOCUB6oc7ETUUS | 39,921 | https://github.com/huggingface/transformers/issues/39921 | https://api.github.com/repos/huggingface/transformers/issues/39921 | [Gemma3N] Not able to add new special tokens to model/tokenizer due to projection error | ### System Info
```
- transformers==4.54.1
- Platform: Linux-5.15.0-1084-aws-x86_64-with-glibc2.31
- Python version: 3.13
- TRL version: 0.19.1
- Huggingface_hub version: 0.33.4
- Safetensors version: 0.5.3
- Accelerate version: 1.9.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch versio... | open | null | false | 15 | [
"Usage",
"Good Second Issue",
"bug"
] | [] | 2025-08-05T14:43:37Z | 2026-07-21T15:33:42Z | null | NONE | null | 20260721T180022Z | 2026-07-21T18:00:22Z | debasisdwivedy | 20,119,757 | MDQ6VXNlcjIwMTE5NzU3 | User | false |
huggingface/transformers | 4,940,589,451 | I_kwDOCUB6oc8AAAABJntpiw | 47,458 | https://github.com/huggingface/transformers/issues/47458 | https://api.github.com/repos/huggingface/transformers/issues/47458 | [Bug] GPT2 c_proj weights incorrectly initialised | ### System Info
- `transformers` version: 5.14.1
- Platform: macOS-15.7.3-arm64-arm-64bit
- Python version: 3.12.12
- Huggingface_hub version: 1.24.0
- Safetensors version: 0.8.0
- Accelerate version: not installed
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?): 2.12... | open | null | false | 0 | [
"bug"
] | [] | 2026-07-21T15:41:25Z | 2026-07-21T15:41:25Z | null | NONE | null | 20260721T180022Z | 2026-07-21T18:00:22Z | DavidJohnQuinlan | 54,847,160 | MDQ6VXNlcjU0ODQ3MTYw | User | false |
huggingface/transformers | 2,496,721,109 | I_kwDOCUB6oc6U0PDV | 33,210 | https://github.com/huggingface/transformers/issues/33210 | https://api.github.com/repos/huggingface/transformers/issues/33210 | The model's address is https://huggingface.co/Xenova/nllb-200-distilled-600M/tree/main/onnx。I don't know how to load encode.onnx and decoder.onnx, and successfully translate a sentence into another language. Can you help me write an inference code to achieve the translation effect through the encoder and decoder? thank... | ### Feature request
hello，The model's address is [https://huggingface.co/Xenova/nllb-200-distilled-600M/tree/main/onnx。I](https://huggingface.co/Xenova/nllb-200-distilled-600M/tree/main/onnx%E3%80%82I) don't know how to load encode.onnx and decoder.onnx, and successfully translate a sentence into another language. Can... | closed | completed | false | 3 | [
"Feature request"
] | [] | 2024-08-30T09:33:01Z | 2026-07-21T19:20:51Z | 2026-07-21T19:20:51Z | NONE | null | 20260722T000038Z | 2026-07-22T00:00:38Z | pengpengtao | 63,086,628 | MDQ6VXNlcjYzMDg2NjI4 | User | false |
huggingface/transformers | 1,598,198,203 | I_kwDOCUB6oc5fQpG7 | 21,778 | https://github.com/huggingface/transformers/issues/21778 | https://api.github.com/repos/huggingface/transformers/issues/21778 | Add TensorFlow Wav2Vec2 for sequence classification | ### Feature request
Wav2Vec2 is one of the most popular speech recognition models, used over 2 million times monthly. In the PyTorch modelling code, we have Wav2Vec2 for speech recognition _and_ Wav2Vec2 for audio classification. However, in TensorFlow, we only have Wav2Vec2 for speech recognition. It would be great... | closed | completed | false | 10 | [
"TensorFlow",
"Good Second Issue",
"Feature request"
] | [] | 2023-02-24T08:53:01Z | 2026-07-21T19:21:49Z | 2026-07-21T19:21:49Z | CONTRIBUTOR | null | 20260722T000038Z | 2026-07-22T00:00:38Z | sanchit-gandhi | 93,869,735 | U_kgDOBZhWpw | User | false |
huggingface/transformers | 3,085,061,733 | I_kwDOCUB6oc634k5l | 38,312 | https://github.com/huggingface/transformers/issues/38312 | https://api.github.com/repos/huggingface/transformers/issues/38312 | LagKV for key-value compression | ### Feature request
Integrated the LagKV for KV compression.
https://github.com/AI-Lab-China-Merchants-Bank/LagKV
### Motivation
Right now, the Cache utils is expanding rapidly. But, it still lack an efficient token eviction algorithm except SinkCache and SlidingWindowCache, which show limited performance in inform... | closed | not_planned | false | 12 | [
"Feature request"
] | [] | 2025-05-23T03:54:42Z | 2026-07-21T20:12:33Z | 2026-07-21T20:12:32Z | NONE | null | 20260722T000038Z | 2026-07-22T00:00:38Z | JoelSeniorLiang | 5,005,500 | MDQ6VXNlcjUwMDU1MDA= | User | false |
huggingface/transformers | 1,688,575,885 | I_kwDOCUB6oc5kpZ-N | 23,050 | https://github.com/huggingface/transformers/issues/23050 | https://api.github.com/repos/huggingface/transformers/issues/23050 | [New model] 🐸TTS advanced Text-to-Speech | ### Model description
🐸TTS is a library for advanced Text-to-Speech generation. It's built on the latest research, was designed to achieve the best trade-off among ease-of-training, speed and quality. 🐸TTS comes with pretrained models, tools for measuring dataset quality and already used in 20+ languages for product... | closed | not_planned | false | 21 | [
"New model"
] | [] | 2023-04-28T13:37:38Z | 2026-07-21T20:14:41Z | 2026-07-21T20:14:41Z | NONE | null | 20260722T000038Z | 2026-07-22T00:00:38Z | jozefchutka | 750,041 | MDQ6VXNlcjc1MDA0MQ== | User | false |
huggingface/transformers | 856,166,780 | MDU6SXNzdWU4NTYxNjY3ODA= | 11,209 | https://github.com/huggingface/transformers/issues/11209 | https://api.github.com/repos/huggingface/transformers/issues/11209 | [RFC] introduce `config.trained_precision` | # 🚀 Feature request
As we are discovering that `bf16`-pretrained models don't do well on `fp16` "regime" (and surely vice-versa), and some models are pre-trained in `fp32` and surely won't do well on either `bf16` or `fp16`, and the problem is going to grow as more `bf16`-supporting hardware comes out, I propose we... | closed | not_planned | false | 16 | [
"WIP"
] | [
"stas00"
] | 2021-04-12T16:54:09Z | 2026-07-21T22:40:57Z | 2026-07-21T22:40:57Z | CONTRIBUTOR | null | 20260722T000038Z | 2026-07-22T00:00:38Z | stas00 | 10,676,103 | MDQ6VXNlcjEwNjc2MTAz | User | false |
huggingface/transformers | 2,689,299,821 | I_kwDOCUB6oc6gS3Vt | 34,905 | https://github.com/huggingface/transformers/issues/34905 | https://api.github.com/repos/huggingface/transformers/issues/34905 | WSD Scheduler to auto infer training steps | ### Feature request
WSD Scheduler should calculate stable steps in `trainer.py`. And if num_warmup_steps is provided in kwargs, schedule_func should respect the kwargs.
My guess is that the intention is it to decay till min and stay there till the end of training, but `min_lr_ratio` is set to the default of 0, wou... | closed | completed | false | 2 | [
"Feature request"
] | [] | 2024-11-25T06:01:17Z | 2026-07-21T22:43:31Z | 2026-07-21T22:43:31Z | NONE | null | 20260722T000038Z | 2026-07-22T00:00:38Z | wheynelau | 90,563,815 | MDQ6VXNlcjkwNTYzODE1 | User | false |
huggingface/transformers | 4,944,207,236 | I_kwDOCUB6oc8AAAABJrKdhA | 47,465 | https://github.com/huggingface/transformers/issues/47465 | https://api.github.com/repos/huggingface/transformers/issues/47465 | [serge] integration failure triage - 2026-07-21 | <!-- serge-triage-run:integration-failure-triage:2026-07-21 -->
Automated **integration-failure triage** for the daily CI window `2026-07-13 → 2026-07-21`.
This issue was generated by AI-assisted automation. The grouping, summaries, and recommended follow-up can be incomplete or misleading; verify the failures before... | open | null | false | 0 | [] | [
"tarekziade"
] | 2026-07-22T00:25:20Z | 2026-07-22T01:08:17Z | null | CONTRIBUTOR | null | 20260722T060034Z | 2026-07-22T06:00:34Z | github-actions[bot] | 41,898,282 | MDM6Qm90NDE4OTgyODI= | Bot | false |
huggingface/transformers | 4,944,530,545 | I_kwDOCUB6oc8AAAABJreMcQ | 47,467 | https://github.com/huggingface/transformers/issues/47467 | https://api.github.com/repos/huggingface/transformers/issues/47467 | Qwen3.5-MoE (Qwen3_5MoeExperts) data-dependent expert dispatch breaks torch.utils.checkpoint reproducibility — CheckpointError / silent NaN gradients | ### Summary
Training (LoRA fine-tuning) `Qwen3_5MoeForCausalLM` / `Qwen3_5MoeForConditionalGeneration` with `model.gradient_checkpointing_enable()` produces either a hard `torch.utils.checkpoint.CheckpointError` or silent NaN gradients during `backward()`, depending on run conditions.
Root cause: the MoE expert-dispa... | open | null | false | 0 | [] | [] | 2026-07-22T01:34:37Z | 2026-07-22T01:34:37Z | null | NONE | null | 20260722T060034Z | 2026-07-22T06:00:34Z | sternryan | 206,953,196 | U_kgDODFXa7A | User | false |
huggingface/transformers | 4,945,806,062 | I_kwDOCUB6oc8AAAABJssC7g | 47,470 | https://github.com/huggingface/transformers/issues/47470 | https://api.github.com/repos/huggingface/transformers/issues/47470 | ESMFold throws error when explicitly cast to fp16 | ### System Info
- `transformers` version: 5.14.1
- Platform: macOS-15.7.3-arm64-arm-64bit
- Python version: 3.12.13
- Huggingface_hub version: 1.24.0
- Safetensors version: 0.8.0
- Accelerate version: 1.14.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?): 2.13.0 (NA... | open | null | false | 0 | [
"bug"
] | [] | 2026-07-22T05:33:56Z | 2026-07-22T05:33:56Z | null | CONTRIBUTOR | null | 20260722T060034Z | 2026-07-22T06:00:34Z | crackalamoo | 37,599,247 | MDQ6VXNlcjM3NTk5MjQ3 | User | false |
huggingface/transformers | 4,946,578,803 | I_kwDOCUB6oc8AAAABJtbNcw | 47,473 | https://github.com/huggingface/transformers/issues/47473 | https://api.github.com/repos/huggingface/transformers/issues/47473 | SentencePieceBackend should decode byte-fallback tokens | ### System Info
- `transformers` version: 5.14.1
- Platform: Linux-6.18.33.2-microsoft-standard-WSL2-x86_64-with-glibc2.39
- Python version: 3.14.5
- Huggingface_hub version: 1.24.0
- Safetensors version: 0.8.0
- Accelerate version: not installed
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTo... | open | null | false | 0 | [
"bug"
] | [] | 2026-07-22T07:29:55Z | 2026-07-22T07:29:55Z | null | NONE | null | 20260722T120022Z | 2026-07-22T12:00:22Z | toslunar | 1,964,281 | MDQ6VXNlcjE5NjQyODE= | User | false |
huggingface/transformers | 4,948,214,733 | I_kwDOCUB6oc8AAAABJu_DzQ | 47,475 | https://github.com/huggingface/transformers/issues/47475 | https://api.github.com/repos/huggingface/transformers/issues/47475 | [Zamba2] Causality violation in torch_forward: inter-chunk state recurrence reduces over the wrong axis | ### System Info
- transformers version: 5.6.0.dev0
- Platform: Linux (x86_64)
- Python version: 3.10
- PyTorch version: 2.10.0+cu126
- Device used for this report: CPU
- mamba_ssm / causal_conv1d: not installed, so the pure-PyTorch path (torch_forward) is taken
### Who can help?
_No response_
### Information
- [ ]... | open | null | false | 0 | [
"bug"
] | [] | 2026-07-22T11:00:56Z | 2026-07-22T11:04:54Z | null | NONE | null | 20260722T120022Z | 2026-07-22T12:00:22Z | openaiteams | 125,527,015 | U_kgDOB3tj5w | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.