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 | 4,407,949,194 | I_kwDOCUB6oc8AAAABBrv3ig | 45,850 | https://github.com/huggingface/transformers/issues/45850 | https://api.github.com/repos/huggingface/transformers/issues/45850 | Since 5.0 version it breaks too many models | ### System Info
Im using DNABERT -2 and after i update transformers there is an error:
AttributeError: 'BertConfig' object has no attribute 'pad_token_id'
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task... | closed | completed | false | 3 | [
"bug"
] | [] | 2026-05-08T16:47:19Z | 2026-06-16T09:42:38Z | 2026-06-16T09:42:38Z | NONE | null | 20260616T120046Z | 2026-06-16T12:00:46Z | LOYINuts | 126,253,581 | U_kgDOB4Z6DQ | User | false |
huggingface/transformers | 4,409,674,091 | I_kwDOCUB6oc8AAAABBtZJaw | 45,853 | https://github.com/huggingface/transformers/issues/45853 | https://api.github.com/repos/huggingface/transformers/issues/45853 | Provide HF_USE_MLX=0 / public flag to disable MLX backend detection at import time | ### Summary
Transformers' import-utils module probes for MLX availability during its own
load and imports `mlx.core` for backend type checks if `mlx` is installed.
There is no documented opt-out. On some Apple Silicon configurations,
`import mlx.core` aborts the interpreter (silent `SIGABRT` during native
init), takin... | closed | completed | false | 7 | [] | [] | 2026-05-08T22:04:11Z | 2026-05-25T15:11:43Z | 2026-05-25T15:11:43Z | NONE | null | 20260525T180030Z | 2026-05-25T18:00:30Z | Steve-Allison | 3,996,420 | MDQ6VXNlcjM5OTY0MjA= | User | false |
huggingface/transformers | 4,410,296,606 | I_kwDOCUB6oc8AAAABBt_JHg | 45,854 | https://github.com/huggingface/transformers/issues/45854 | https://api.github.com/repos/huggingface/transformers/issues/45854 | pipeline text-generation ignores return_full_text=False with chat template | When using the text-generation pipeline with a chat template format, setting return_full_text=False still returns the full conversation including the prompt. Tested on transformers 4.40.x, model meta-llama/Llama-3-8B-Instruct. Expected: only generated text. Actual: full prompt + generated text. | closed | completed | false | 3 | [] | [] | 2026-05-09T00:34:19Z | 2026-05-12T12:47:32Z | 2026-05-12T12:47:32Z | NONE | null | 20260512T180027Z | 2026-05-12T18:00:27Z | yashpandya-msr | 213,021,850 | U_kgDODLJ0mg | User | false |
huggingface/transformers | 4,411,356,695 | I_kwDOCUB6oc8AAAABBu_2Fw | 45,859 | https://github.com/huggingface/transformers/issues/45859 | https://api.github.com/repos/huggingface/transformers/issues/45859 | `Qwen3_5MoeTextRotaryEmbedding.forward` is not compatible with CPU offload | ### System Info
- `transformers` version: 5.5.4
- Platform: Linux-6.8.0-1043-nvidia-x86_64-with-glibc2.35
- Python version: 3.12.13
- Huggingface_hub version: 1.11.0
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerat... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-05-09T05:46:24Z | 2026-05-15T16:21:46Z | 2026-05-12T08:09:24Z | CONTRIBUTOR | null | 20260515T180026Z | 2026-05-15T18:00:26Z | jamesbraza | 8,990,777 | MDQ6VXNlcjg5OTA3Nzc= | User | false |
huggingface/transformers | 4,413,459,774 | I_kwDOCUB6oc8AAAABBxANPg | 45,864 | https://github.com/huggingface/transformers/issues/45864 | https://api.github.com/repos/huggingface/transformers/issues/45864 | PretrainedConfig.from_pretrained: silent default on missing config.json | ### System Info
- transformers 5.6+ (verified on 5.8.0; 5.5.4 raised)
- introduced by #36033 ("Large/full refactor of from_pretrained", 2025-03-12)
### Description
When `PretrainedConfig.from_pretrained(path)` is given a *local* directory with no `config.json`, it silently returns a default-populated config instead ... | closed | not_planned | false | 1 | [] | [] | 2026-05-09T18:25:53Z | 2026-05-09T18:38:50Z | 2026-05-09T18:38:50Z | CONTRIBUTOR | null | 20260510T000023Z | 2026-05-10T00:00:23Z | MilkClouds | 26,109,705 | MDQ6VXNlcjI2MTA5NzA1 | User | false |
huggingface/transformers | 4,414,834,078 | I_kwDOCUB6oc8AAAABByUFng | 45,865 | https://github.com/huggingface/transformers/issues/45865 | https://api.github.com/repos/huggingface/transformers/issues/45865 | [Feature Request] Add lossy speculative decoding via static ensemble verification | ### Feature request
**Is your feature request related to a problem? Please describe.**
Standard speculative decoding (assisted generation) in Transformers is *lossless* — it guarantees the output distribution exactly matches the target model. While this is a strong guarantee, it comes at a cost: many plausible draft ... | closed | completed | false | 5 | [] | [] | 2026-05-10T06:09:08Z | 2026-06-18T17:45:33Z | 2026-06-17T09:31:08Z | NONE | null | 20260618T180024Z | 2026-06-18T18:00:24Z | kasakh | 31,774,865 | MDQ6VXNlcjMxNzc0ODY1 | User | false |
huggingface/transformers | 4,416,384,722 | I_kwDOCUB6oc8AAAABBzyu0g | 45,869 | https://github.com/huggingface/transformers/issues/45869 | https://api.github.com/repos/huggingface/transformers/issues/45869 | Dataset: Financial Data Bundle (3,518 records) - Free Sample Available | I compiled a financial data bundle that might be useful for testing, benchmarking, or tutorials in this project:
**DataForge Financial Data Bundle**
- 3,518 real records across 5 datasets
- Crypto (BTC, ETH, SOL, ADA, XRP, DOT)
- AI Stocks (NVDA, MSFT, GOOGL, AMD, META)
- EV Stocks (TSLA, RIVN, NIO, LCID)
- Tech OHLCV... | open | null | false | 1 | [] | [] | 2026-05-10T17:59:05Z | 2026-05-11T02:09:14Z | null | NONE | null | 20260511T060028Z | 2026-05-11T06:00:28Z | matisaar | 187,905,737 | U_kgDOCzM2yQ | User | false |
huggingface/transformers | 4,416,482,423 | I_kwDOCUB6oc8AAAABBz4sdw | 45,870 | https://github.com/huggingface/transformers/issues/45870 | https://api.github.com/repos/huggingface/transformers/issues/45870 | Example: Financial time series prediction with DataForge | ## Context
I have been working with a comprehensive financial dataset called DataForge (3,500+ real records, OHLCV format).
## Contents
- Crypto prices (BTC, ETH, top 100)
- AI stocks (NVDA, AMD, INTC)
- EV stocks (TSLA, RIVN, NIO)
- Tech stocks (AAPL, GOOGL, MSFT)
- Macro indicators
## Links
- Sales: https://matisaa... | open | null | false | 1 | [] | [] | 2026-05-10T18:40:10Z | 2026-05-10T18:47:08Z | null | NONE | null | 20260511T000025Z | 2026-05-11T00:00:25Z | matisaar | 187,905,737 | U_kgDOCzM2yQ | User | false |
huggingface/transformers | 4,416,532,134 | I_kwDOCUB6oc8AAAABBz7upg | 45,871 | https://github.com/huggingface/transformers/issues/45871 | https://api.github.com/repos/huggingface/transformers/issues/45871 | Example: Financial time series prediction with DataForge | ## Context
I have been working with a comprehensive financial dataset called DataForge (3,500+ real records, OHLCV format).
## Contents
- Crypto prices (BTC, ETH, top 100)
- AI stocks (NVDA, AMD, INTC)
- EV stocks (TSLA, RIVN, NIO)
- Tech stocks (AAPL, GOOGL, MSFT)
- Macro indicators
## Links
- Sales: https://matisaa... | open | null | false | 0 | [] | [] | 2026-05-10T19:02:06Z | 2026-05-10T19:02:06Z | null | NONE | null | 20260511T000025Z | 2026-05-11T00:00:25Z | matisaar | 187,905,737 | U_kgDOCzM2yQ | User | false |
huggingface/transformers | 4,416,562,363 | I_kwDOCUB6oc8AAAABBz9kuw | 45,872 | https://github.com/huggingface/transformers/issues/45872 | https://api.github.com/repos/huggingface/transformers/issues/45872 | DataForge: Financial time series + scraping service | ## DataForge - Financial Datasets + Custom Scraping Service
I built a comprehensive financial dataset and wanted to share it with this community.
### Products
- **$1** - Sample datasets (300-500 records)
- **$9** - Individual bundles (500+ records)
- **$19** - Complete bundle (3,500+ records)
### Custom Scraping S... | open | null | false | 0 | [] | [] | 2026-05-10T19:14:08Z | 2026-05-10T19:14:08Z | null | NONE | null | 20260511T000025Z | 2026-05-11T00:00:25Z | matisaar | 187,905,737 | U_kgDOCzM2yQ | User | false |
huggingface/transformers | 4,416,591,707 | I_kwDOCUB6oc8AAAABBz_XWw | 45,873 | https://github.com/huggingface/transformers/issues/45873 | https://api.github.com/repos/huggingface/transformers/issues/45873 | DataForge: Financial time series + scraping service | ## DataForge - Financial Datasets + Custom Scraping Service
I built a comprehensive financial dataset and wanted to share it with this community.
### Products
- **$1** - Sample datasets (300-500 records)
- **$9** - Individual bundles (500+ records)
- **$19** - Complete bundle (3,500+ records)
### Custom Scraping S... | open | null | false | 0 | [] | [] | 2026-05-10T19:26:02Z | 2026-05-10T19:26:02Z | null | NONE | null | 20260511T000025Z | 2026-05-11T00:00:25Z | matisaar | 187,905,737 | U_kgDOCzM2yQ | User | false |
huggingface/transformers | 4,416,762,215 | I_kwDOCUB6oc8AAAABB0JxZw | 45,874 | https://github.com/huggingface/transformers/issues/45874 | https://api.github.com/repos/huggingface/transformers/issues/45874 | Gemma4-E2B/E4B: passing `inputs_embeds` triggers an extremely expensive reverse embedding lookup | ### System Info
- `transformers` version: 5.7.0
- Platform: Windows-11-10.0.26200-SP0
- Python version: 3.13.13
- Huggingface_hub version: 1.13.0
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?): 2.11.0+cu130... | closed | completed | false | 13 | [
"bug"
] | [] | 2026-05-10T20:35:48Z | 2026-05-14T07:28:04Z | 2026-05-14T07:28:04Z | NONE | null | 20260514T120030Z | 2026-05-14T12:00:30Z | thijs-vanweezel | 117,186,594 | U_kgDOBvwgIg | User | false |
huggingface/transformers | 4,418,099,931 | I_kwDOCUB6oc8AAAABB1ba2w | 45,882 | https://github.com/huggingface/transformers/issues/45882 | https://api.github.com/repos/huggingface/transformers/issues/45882 | Reduce peak memory during quantization of MoE models (follow-up to #43917) | ### Context
#43917 (Model patching API) gave calibration-based quantizers (auto-round, gptqmodel, …) a clean way to swap our fused `*Experts` for a defused `ModuleList[Linear]` via `register_monkey_patch_mapping`. That fixed the detection half of #43284 — quantizers can see individual `nn.Linear` weights again.
It do... | open | null | false | 0 | [] | [] | 2026-05-11T04:37:45Z | 2026-05-11T04:37:45Z | null | MEMBER | null | 20260511T060028Z | 2026-05-11T06:00:28Z | ArthurZucker | 48,595,927 | MDQ6VXNlcjQ4NTk1OTI3 | User | false |
huggingface/transformers | 4,424,460,847 | I_kwDOCUB6oc8AAAABB7fqLw | 45,901 | https://github.com/huggingface/transformers/issues/45901 | https://api.github.com/repos/huggingface/transformers/issues/45901 | table-question-answering task crashes | ### System Info
```
(.venv) C:\Users\usuario\Documents\aibased>transformers env
Traceback (most recent call last):
File "<frozen runpy>", line 198, in _run_module_as_main
File "<frozen runpy>", line 88, in _run_code
File "C:\Users\usuario\Documents\aibased\.venv\Scripts\transformers.exe\__main__.py", line 4, in ... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-05-11T21:33:49Z | 2026-05-28T14:25:48Z | 2026-05-28T14:25:48Z | NONE | null | 20260528T180037Z | 2026-05-28T18:00:37Z | someone282801 | 283,800,608 | U_kgDOEOp0IA | User | false |
huggingface/transformers | 4,424,896,390 | I_kwDOCUB6oc8AAAABB76Phg | 45,902 | https://github.com/huggingface/transformers/issues/45902 | https://api.github.com/repos/huggingface/transformers/issues/45902 | `Qwen3_5MoeTextRotaryEmbedding.inv_freq` reads uninitialized memory after `meta → to_empty(cuda)` materialization | ### System Info
- `transformers` version: 5.6.2
- Platform: Linux-6.8.0-1043-nvidia-x86_64-with-glibc2.35
- Python version: 3.12.13
- Huggingface_hub version: 1.13.0
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerato... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-05-11T23:01:34Z | 2026-05-13T12:55:55Z | 2026-05-13T06:12:17Z | CONTRIBUTOR | null | 20260513T180036Z | 2026-05-13T18:00:36Z | jamesbraza | 8,990,777 | MDQ6VXNlcjg5OTA3Nzc= | User | false |
huggingface/transformers | 4,426,006,552 | I_kwDOCUB6oc8AAAABB8-AGA | 45,910 | https://github.com/huggingface/transformers/issues/45910 | https://api.github.com/repos/huggingface/transformers/issues/45910 | [DeepSeekV4] Potential RoPE theta mismatch between main attention and compressed KV branches | ## Description
I noticed a potential inconsistency between the official DeepSeekV4 `inference/model.py` implementation released on Hugging Face and the current `transformers` implementation in `modeling_deepseek_v4.py`.
In the official `inference/model.py`, the RoPE theta seems to be selected based on `self.compress_... | closed | completed | false | 2 | [] | [] | 2026-05-12T03:46:09Z | 2026-05-12T09:24:52Z | 2026-05-12T09:24:52Z | NONE | null | 20260512T120027Z | 2026-05-12T12:00:27Z | WKQ9411 | 127,908,800 | U_kgDOB5-7wA | User | false |
huggingface/transformers | 4,425,875,923 | I_kwDOCUB6oc8AAAABB82B0w | 45,907 | https://github.com/huggingface/transformers/issues/45907 | https://api.github.com/repos/huggingface/transformers/issues/45907 | Our `tinker-cookbook` CI broke: `list_repo_files` should forward the `revision` argument | ### System Info
In `PreTrainedTokenizerBase._from_pretrained`, the call to `list_repo_files` does not forward the `revision` kwarg, so the returned file list always reflects the repo's `main` branch even when the caller pinned a specific revision. This makes the surrounding "fall back to `tiktoken.model`/`tokenizer.mo... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-05-12T03:11:36Z | 2026-05-12T05:55:37Z | 2026-05-12T05:55:37Z | CONTRIBUTOR | null | 20260512T060024Z | 2026-05-12T06:00:24Z | nealwu | 726,075 | MDQ6VXNlcjcyNjA3NQ== | User | false |
huggingface/transformers | 801,257,815 | MDU6SXNzdWU4MDEyNTc4MTU= | 10,000 | https://github.com/huggingface/transformers/issues/10000 | https://api.github.com/repos/huggingface/transformers/issues/10000 | German DistilBertModel raises an issue | ## Environment info
- `transformers` version: 4.2.2
- Platform: Linux-5.4.0-65-generic-x86_64-with-debian-buster-sid
- Python version: 3.7.7
- PyTorch version (GPU?): 1.8.0.dev20201202 (False)
- Tensorflow version (GPU?): not installed (NA)
- Using GPU in script?: No
- Using distributed or parallel set-up in scr... | closed | completed | false | 7 | [] | [] | 2021-02-04T12:54:33Z | 2026-05-12T08:12:41Z | 2021-02-04T13:14:01Z | NONE | null | 20260512T120027Z | 2026-05-12T12:00:27Z | Svito-zar | 15,908,492 | MDQ6VXNlcjE1OTA4NDky | User | false |
huggingface/transformers | 4,429,319,142 | I_kwDOCUB6oc8AAAABCAIL5g | 45,915 | https://github.com/huggingface/transformers/issues/45915 | https://api.github.com/repos/huggingface/transformers/issues/45915 | 🚨 Security Analysis: 1. An attacker identifies the insecure-deserialization vulne | ## Security Analysis: 1. An attacker identifies the insecure-deserialization vulnerability in `read_metadata`
### Impact
The `read_metadata` function contains a insecure-deserialization vulnerability that could be exploited by an attacker to compromise the application or access unauthorized resources.
### Attack Scen... | closed | completed | false | 1 | [
"Code agent slop"
] | [] | 2026-05-12T13:01:28Z | 2026-05-12T14:45:00Z | 2026-05-12T14:28:38Z | NONE | null | 20260512T180027Z | 2026-05-12T18:00:27Z | anxovatomica | 23,504,033 | MDQ6VXNlcjIzNTA0MDMz | User | false |
huggingface/transformers | 4,431,860,621 | I_kwDOCUB6oc8AAAABCCjTjQ | 45,920 | https://github.com/huggingface/transformers/issues/45920 | https://api.github.com/repos/huggingface/transformers/issues/45920 | AutoTokenizer produces wrong token IDs for OLMo2, HyperClovaX, DeepSeek-R1-Distill-Llama, Yi, and others (v5 regression) | ### System Info
- `transformers` version: 5.8.0 (also reproduced on 5.0.0 through 5.7.0)
- Platform: Linux-5.14.0-503.11.1.el9_5.x86_64-x86_64-with-glibc2.34
- Python version: 3.12.13
- Huggingface_hub version: 1.14.0
- Safetensors version: 0.7.0
- Tokenizers version: 0.22.2
### Who can help?
@ArthurZucker and @ita... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-05-12T19:02:06Z | 2026-06-19T11:11:09Z | 2026-06-19T11:11:09Z | CONTRIBUTOR | null | 20260619T120025Z | 2026-06-19T12:00:25Z | kndtran | 19,249,995 | MDQ6VXNlcjE5MjQ5OTk1 | User | false |
huggingface/transformers | 4,432,694,110 | I_kwDOCUB6oc8AAAABCDWLXg | 45,923 | https://github.com/huggingface/transformers/issues/45923 | https://api.github.com/repos/huggingface/transformers/issues/45923 | Nemotron-3-Nano-Omni: supports_gradient_checkpointing flag missing on trust_remote_code variant (1-line fix) | ### System Info
- transformers version: 5.8.0
- Platform: Linux-6.17.0-22-generic-x86_64-with-glibc2.39
- Python version: 3.10.19
- PyTorch version (GPU?): 2.10.0+cu128 (cuda 12.8)
- Huggingface_hub version: 0.36.2
- Safetensors version: 0.6.2
- Accelerate version: 1.11.0
- Accelerate config: not found
- PEFT version:... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-05-12T21:14:48Z | 2026-05-13T12:45:17Z | 2026-05-13T12:45:16Z | NONE | null | 20260513T180036Z | 2026-05-13T18:00:36Z | badlerSI | 213,488,728 | U_kgDODLmUWA | User | false |
huggingface/transformers | 4,433,874,796 | I_kwDOCUB6oc8AAAABCEePbA | 45,925 | https://github.com/huggingface/transformers/issues/45925 | https://api.github.com/repos/huggingface/transformers/issues/45925 | Discrepancy Between the Paper’s Claimed Sparse Attention Complexity and the HuggingFace Implementation Caused by Expanding KV to S * k | In the HuggingFace Transformers implementation of DeepSeek V4, the Compressed Sparse Attention (CSA) module gathers the top‑k compressed blocks selected by the indexer for each query, flattens them into a tensor of shape [B, 1, S*k, head_dim], and then directly concatenates this tensor with the sliding‑window KV:
Fina... | closed | completed | false | 8 | [] | [] | 2026-05-13T01:26:16Z | 2026-05-14T00:39:31Z | 2026-05-14T00:39:31Z | NONE | null | 20260514T060044Z | 2026-05-14T06:00:44Z | I-hercules | 20,699,384 | MDQ6VXNlcjIwNjk5Mzg0 | User | false |
huggingface/transformers | 3,676,102,629 | I_kwDOCUB6oc7bHN_l | 42,490 | https://github.com/huggingface/transformers/issues/42490 | https://api.github.com/repos/huggingface/transformers/issues/42490 | Import warning when loading a local model from a path that ends with a slash | ### System Info
- `transformers` version: 4.57.1
- Platform: Linux-6.8.0-87-generic-x86_64-with-glibc2.39
- Python version: 3.12.3
- Huggingface_hub version: 0.36.0
- Safetensors version: 0.7.0
- Accelerate version: 1.11.0
- Accelerate config:
- compute_environment: LOCAL_MACHINE
- distributed_type: NO
- mixed_prec... | closed | completed | false | 11 | [
"bug"
] | [] | 2025-11-28T23:25:01Z | 2026-05-13T10:56:04Z | 2025-12-03T16:52:27Z | NONE | null | 20260513T120045Z | 2026-05-13T12:00:45Z | kk-89 | 31,396,340 | MDQ6VXNlcjMxMzk2MzQw | User | false |
huggingface/transformers | 4,437,398,802 | I_kwDOCUB6oc8AAAABCH1VEg | 45,941 | https://github.com/huggingface/transformers/issues/45941 | https://api.github.com/repos/huggingface/transformers/issues/45941 | FSDP + KD-teacher-wrap + flash-attn-2 + LoRA: working-set memory exceeds 40 GiB per rank on production-shape SFT training (deepseek-coder-6.7b + sahil2801/CodeAlpaca-20k) | ## Summary
We're running supervised fine-tuning of `deepseek-ai/deepseek-coder-6.7b-instruct` with LoRA (PEFT) + Knowledge Distillation (teacher = same architecture as student, full bf16 replica), via the HF `Trainer` with FSDP `full_shard` integration, `attn_implementation="flash_attention_2"`, bf16, world_size=8, on... | closed | completed | false | 1 | [] | [] | 2026-05-13T11:20:34Z | 2026-05-13T13:07:56Z | 2026-05-13T13:07:56Z | NONE | null | 20260513T180036Z | 2026-05-13T18:00:36Z | drewvenegas | 134,726,503 | U_kgDOCAfDZw | User | false |
huggingface/transformers | 4,436,060,369 | I_kwDOCUB6oc8AAAABCGjo0Q | 45,938 | https://github.com/huggingface/transformers/issues/45938 | https://api.github.com/repos/huggingface/transformers/issues/45938 | [DeepSeekV4] Compressor does not seem to account for padding tokens when forming compressed KV blocks | ## Description
I noticed a potential padding-related issue in the current DeepSeekV4 implementation.
The sliding-window attention path appears to correctly receive the user-provided `attention_mask` through `create_sliding_window_causal_mask`, so padded key/value tokens can be masked out in the normal attention path.... | open | null | false | 11 | [
"Good Difficult Issue"
] | [] | 2026-05-13T08:06:41Z | 2026-05-27T12:45:21Z | null | NONE | null | 20260527T180027Z | 2026-05-27T18:00:27Z | WKQ9411 | 127,908,800 | U_kgDOB5-7wA | User | false |
huggingface/transformers | 4,440,513,009 | I_kwDOCUB6oc8AAAABCKzZ8Q | 45,950 | https://github.com/huggingface/transformers/issues/45950 | https://api.github.com/repos/huggingface/transformers/issues/45950 | BAN | sorry for creating this issue but this is little unrelated
Hey any maintainer here or mod here from discord , or if you know someone to connect .My account is banned from the discord server because my account was hacked and spammed ... is there anyone who can help me to unban .
I am not able to find anyone | closed | completed | false | 4 | [] | [] | 2026-05-13T19:09:56Z | 2026-06-12T16:21:53Z | 2026-05-15T17:46:14Z | NONE | null | 20260612T180035Z | 2026-06-12T18:00:35Z | Priyabhunia | 116,344,684 | U_kgDOBu9HbA | User | false |
huggingface/transformers | 2,126,161,751 | I_kwDOCUB6oc5-uqdX | 28,936 | https://github.com/huggingface/transformers/issues/28936 | https://api.github.com/repos/huggingface/transformers/issues/28936 | [i18n-es] Translating docs to Spanish | Hi!
Let's bring the documentation to all the Spanish-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... | open | reopened | false | 11 | [
"WIP"
] | [] | 2024-02-08T22:19:09Z | 2026-05-14T20:07:26Z | null | MEMBER | null | 20260515T000050Z | 2026-05-15T00:00:50Z | stevhliu | 59,462,357 | MDQ6VXNlcjU5NDYyMzU3 | User | false |
huggingface/transformers | 4,452,020,581 | I_kwDOCUB6oc8AAAABCVxxZQ | 45,987 | https://github.com/huggingface/transformers/issues/45987 | https://api.github.com/repos/huggingface/transformers/issues/45987 | [Bug] StaticCache.get_seq_length() returns shape-(1,) Tensor despite -> int contract | ### System Info
transformers : 5.7.0.dev0 (main, commit 84c2e2f487d6e792a8f1582f1cb1aa386b0d6133)
Python : 3.13.7
Platform : Windows 11 AMD64
PyTorch : 2.11.0+cu126
### Who can help?
@ArthurZucker @zucchini-nlp
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tas... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-05-15T07:09:04Z | 2026-06-22T09:55:52Z | 2026-06-22T09:55:52Z | CONTRIBUTOR | null | 20260622T120018Z | 2026-06-22T12:00:18Z | Abineshabee | 104,718,709 | U_kgDOBj3hdQ | User | false |
huggingface/transformers | 4,453,879,765 | I_kwDOCUB6oc8AAAABCXjP1Q | 45,995 | https://github.com/huggingface/transformers/issues/45995 | https://api.github.com/repos/huggingface/transformers/issues/45995 | docs: Add missing parameter documentation | ## Issue
I noticed several functions are missing proper parameter documentation in the docstrings.
### Affected Areas
- Missing type annotations on several public APIs
- Undocumented return values in some utility functions
- Outdated examples in README
I can submit a PR to fix these if the maintainers are interested... | open | null | false | 4 | [] | [] | 2026-05-15T12:26:21Z | 2026-05-24T04:43:38Z | null | NONE | null | 20260524T060021Z | 2026-05-24T06:00:21Z | h0clam | 283,303,525 | U_kgDOEOLeZQ | User | false |
huggingface/transformers | 4,458,866,238 | I_kwDOCUB6oc8AAAABCcTmPg | 45,998 | https://github.com/huggingface/transformers/issues/45998 | https://api.github.com/repos/huggingface/transformers/issues/45998 | Security Vulnerability | Hi Hugging Face Security Team,
To ensure responsible disclosure, no technical details, reproduction steps, or affected modules are disclosed here.
For the maintenance team's reference, the pending Huntr bounty reports are linked below:
* https://huntr.com/bounties/915d290b-8e76-4093-833c-ba8f675a4c61
* https://hunt... | closed | completed | false | 3 | [] | [] | 2026-05-16T06:10:49Z | 2026-05-28T07:01:25Z | 2026-05-18T17:02:20Z | CONTRIBUTOR | null | 20260528T120029Z | 2026-05-28T12:00:29Z | LinZiyuu | 104,151,270 | U_kgDOBjU45g | User | false |
huggingface/transformers | 4,459,514,188 | I_kwDOCUB6oc8AAAABCc7JTA | 46,000 | https://github.com/huggingface/transformers/issues/46000 | https://api.github.com/repos/huggingface/transformers/issues/46000 | 🎯 AI Agents: Claim tasks and execute them here | <spam> | closed | completed | false | 0 | [] | [] | 2026-05-16T10:04:18Z | 2026-05-18T11:36:16Z | 2026-05-18T11:35:57Z | NONE | null | 20260518T120024Z | 2026-05-18T12:00:24Z | chenyuan35 | 268,374,942 | U_kgDOD_8Tng | User | false |
huggingface/transformers | 4,459,757,393 | I_kwDOCUB6oc8AAAABCdJ_UQ | 46,002 | https://github.com/huggingface/transformers/issues/46002 | https://api.github.com/repos/huggingface/transformers/issues/46002 | NaNs in classification heads upon checkout + init | ### System Info
software
`torch=2.12.0` `transformers=5.8.1`
hardware
`NVIDIA-SMI 580.82.07 Driver Version: 580.82.07 CUDA Version: 13.0`
### Who can help?
call for help
@CyrilVallez - as model loading and init seems wrong
@stevhliu - as this can be better docummented
thanks!
### Information
... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-05-16T11:45:22Z | 2026-05-19T09:54:47Z | 2026-05-19T02:00:09Z | NONE | null | 20260519T120033Z | 2026-05-19T12:00:33Z | maciejskorski | 31,315,784 | MDQ6VXNlcjMxMzE1Nzg0 | User | false |
huggingface/transformers | 4,464,212,697 | I_kwDOCUB6oc8AAAABChZ62Q | 46,015 | https://github.com/huggingface/transformers/issues/46015 | https://api.github.com/repos/huggingface/transformers/issues/46015 | Outdated examples about RAG | ### System Info
```
C:\Users\usuario\AppData\Local\Programs\Python\Python313\python.exe: No module named transformers.__main__; 'transformers' is a package and cannot be directly executed
(.venv) C:\Users\usuario\Documents\aibased>transformers env ... | closed | completed | false | 3 | [
"Good First Issue",
"bug"
] | [] | 2026-05-17T16:48:39Z | 2026-05-18T15:51:42Z | 2026-05-18T15:51:42Z | NONE | null | 20260518T180026Z | 2026-05-18T18:00:26Z | someone282801 | 283,800,608 | U_kgDOEOp0IA | User | false |
huggingface/transformers | 4,464,796,223 | I_kwDOCUB6oc8AAAABCh9iPw | 46,016 | https://github.com/huggingface/transformers/issues/46016 | https://api.github.com/repos/huggingface/transformers/issues/46016 | Code quality scan: 164 findings (A-, 84/100) | Hi @huggingface, an automated scan of this repository surfaced **164 code-quality findings** that may be worth a look.
Full details, severity filters, and per-file context are at the link below — feel free to close this issue if it isn't useful to you.
## Full interactive report
**https://repobility.com/scan/27ddd46... | closed | completed | false | 0 | [] | [] | 2026-05-17T20:10:51Z | 2026-05-18T12:10:43Z | 2026-05-18T12:10:43Z | NONE | null | 20260518T180026Z | 2026-05-18T18:00:26Z | repobilitycom | 280,837,422 | U_kgDOEL09Lg | User | false |
huggingface/transformers | 4,465,722,081 | I_kwDOCUB6oc8AAAABCi2C4Q | 46,018 | https://github.com/huggingface/transformers/issues/46018 | https://api.github.com/repos/huggingface/transformers/issues/46018 | DeepSeek-V4 shared expert not gated. | SwiGLU is not clamped for shared expert. Is it a bug or just a divergence between training and inference?
Deepseek inference ref code: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/main/inference/model.py#L627
transformers implementation: https://github.com/huggingface/transformers/blob/main/src/transforme... | closed | completed | false | 7 | [] | [] | 2026-05-18T01:52:24Z | 2026-06-19T10:26:43Z | 2026-06-19T10:26:43Z | NONE | null | 20260619T120025Z | 2026-06-19T12:00:25Z | yuzhongw-nvidia | 178,130,253 | U_kgDOCp4NTQ | User | false |
huggingface/transformers | 4,466,242,670 | I_kwDOCUB6oc8AAAABCjV0bg | 46,021 | https://github.com/huggingface/transformers/issues/46021 | https://api.github.com/repos/huggingface/transformers/issues/46021 | Passport stamp — toured Hugging Face on GIT42 | Don't panic — I warped through Hugging Face's sector on GIT42 and thought they should know.
Never heard of GIT42? It's a free, ad-free Hitchhiker-style tour of GitHub: developers as planets, galaxies to explore, profiles to visit, and passport stamps for fellow travelers. Chart a course at https://www.git42.dev.
http... | closed | completed | false | 0 | [
"Code agent slop"
] | [] | 2026-05-18T04:23:13Z | 2026-05-18T11:30:38Z | 2026-05-18T11:30:01Z | NONE | null | 20260518T120024Z | 2026-05-18T12:00:24Z | mmatheuspalma | 24,392,490 | MDQ6VXNlcjI0MzkyNDkw | User | false |
huggingface/transformers | 2,145,650,790 | I_kwDOCUB6oc5_5Ahm | 29,159 | https://github.com/huggingface/transformers/issues/29159 | https://api.github.com/repos/huggingface/transformers/issues/29159 | [tokenizer] Inconsistent behavior in slow tokenizer and fast tokenizer | ### System Info
- `transformers` version: 4.35.2
- Platform: Linux-5.4.0-163-generic-x86_64-with-glibc2.10
- Python version: 3.8.18
- Huggingface_hub version: 0.19.4
- Safetensors version: 0.4.1
- Accelerate version: not installed
- Accelerate config: not found
- PyTorch version (GPU?): 2.1.1+cu121 (True)
- ... | closed | completed | false | 12 | [
"Good Second Issue"
] | [] | 2024-02-21T02:24:18Z | 2026-05-18T12:18:01Z | 2026-05-18T12:18:01Z | CONTRIBUTOR | null | 20260518T180026Z | 2026-05-18T18:00:26Z | Ki-Seki | 60,967,965 | MDQ6VXNlcjYwOTY3OTY1 | User | false |
huggingface/transformers | 4,469,302,647 | I_kwDOCUB6oc8AAAABCmQldw | 46,032 | https://github.com/huggingface/transformers/issues/46032 | https://api.github.com/repos/huggingface/transformers/issues/46032 | Mamba2Mixer: use_cache with seq_len > 1 silently produces incorrect results (both CPU and GPU paths) | ### System Info
# Summary
Hi, first of all: thank you for your amazing work on incorporating Mamba2 in the `transformers` library.
We are researching Mamba for a context in which we have very long sequences, and for computational efficiency we would want to process a sequence in batches (e.g. 10% of the sequence each... | closed | completed | false | 6 | [
"Good Second Issue",
"bug"
] | [] | 2026-05-18T12:59:33Z | 2026-06-23T15:51:08Z | 2026-06-23T15:51:08Z | NONE | null | 20260623T180016Z | 2026-06-23T18:00:16Z | clara2911 | 14,054,459 | MDQ6VXNlcjE0MDU0NDU5 | User | false |
huggingface/transformers | 4,474,249,427 | I_kwDOCUB6oc8AAAABCq-g0w | 46,050 | https://github.com/huggingface/transformers/issues/46050 | https://api.github.com/repos/huggingface/transformers/issues/46050 | Model quantized via sinq broken after save_pretrained and from_pretrained | ### System Info
- `transformers` version: 5.8.1
- Platform: Linux-7.0.5-arch1-1-x86_64-with-glibc2.43
- Python version: 3.12.13
- Huggingface_hub version: 1.14.0
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?)... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-05-19T03:51:53Z | 2026-06-27T08:51:43Z | 2026-06-27T08:51:43Z | NONE | null | 20260627T120014Z | 2026-06-27T12:00:14Z | k355l3r-5yndr0m3 | 99,799,355 | U_kgDOBfLROw | User | false |
huggingface/transformers | 4,478,969,706 | I_kwDOCUB6oc8AAAABCvenag | 46,077 | https://github.com/huggingface/transformers/issues/46077 | https://api.github.com/repos/huggingface/transformers/issues/46077 | Gemma4 `use_bidirectional_attention="all"` still builds causal attention masks | ### System Info
- `transformers` version: 5.8.1
### Who can help?
@zucchini-nlp
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [x] My own task or dataset (give details below)
### ... | closed | completed | false | 0 | [
"bug"
] | [] | 2026-05-19T16:04:14Z | 2026-05-26T03:10:06Z | 2026-05-26T03:10:06Z | CONTRIBUTOR | null | 20260526T060025Z | 2026-05-26T06:00:25Z | oliverholworthy | 1,216,955 | MDQ6VXNlcjEyMTY5NTU= | User | false |
huggingface/transformers | 4,480,273,739 | I_kwDOCUB6oc8AAAABCwuNSw | 46,082 | https://github.com/huggingface/transformers/issues/46082 | https://api.github.com/repos/huggingface/transformers/issues/46082 | LlamaConfig rejects explicit head_dim when hidden_size is not divisible by num_attention_heads | ### System Info
- `transformers` version: `5.8.0.dev0` / current `main`
- Checked against `main` commit `a0fb01c6cda2301cb54de0efde5fac405836c4fe`
- Python version: `3.13.12`
### Who can help?
@ArthurZucker @Cyrilvallez
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- ... | closed | completed | false | 3 | [] | [] | 2026-05-19T19:12:17Z | 2026-06-28T08:59:17Z | 2026-06-28T08:59:17Z | CONTRIBUTOR | null | 20260628T180015Z | 2026-06-28T18:00:15Z | Anri-Lombard | 76,818,211 | MDQ6VXNlcjc2ODE4MjEx | User | false |
huggingface/transformers | 4,482,822,424 | I_kwDOCUB6oc8AAAABCzJxGA | 46,093 | https://github.com/huggingface/transformers/issues/46093 | https://api.github.com/repos/huggingface/transformers/issues/46093 | [Inquiry] What is the current status and roadmap for supporting packed sequences (packing) on Qwen3.5/hybrid linear attention models? | ### Context
With the release of Qwen3.5, we see a powerful hybrid architecture that heavily relies on GatedDeltaNet linear attention (powered by flash-linear-attention / FLA) and causal convolutions, alongside traditional self-attention layers.
In standard LLM training (like Llama/Mistral), Packing (padding-free train... | closed | completed | false | 8 | [] | [] | 2026-05-20T03:18:57Z | 2026-05-26T12:00:49Z | 2026-05-26T11:02:44Z | CONTRIBUTOR | null | 20260526T180028Z | 2026-05-26T18:00:28Z | liangxuZhang | 57,205,192 | MDQ6VXNlcjU3MjA1MTky | User | false |
huggingface/transformers | 4,483,189,723 | I_kwDOCUB6oc8AAAABCzgL2w | 46,095 | https://github.com/huggingface/transformers/issues/46095 | https://api.github.com/repos/huggingface/transformers/issues/46095 | [deepseekv4]Does Transformers provide a weight conversion script to convert the Hugging Face weights into a format that can be read by Transformers from_pretrained? | ### Feature request
For [deepseekv4], the weight names provided in the Hugging Face DeepSeek-V4-Flash weights seem not to match the Transformers weight names. Does Transformers provide a weight conversion script to convert the Hugging Face weights into a format that can be read by Transformers from_pretrained?
<img w... | closed | completed | false | 5 | [
"Feature request"
] | [] | 2026-05-20T04:39:29Z | 2026-06-02T18:45:02Z | 2026-06-02T18:45:02Z | NONE | null | 20260603T000024Z | 2026-06-03T00:00:24Z | young-creator | 71,326,088 | MDQ6VXNlcjcxMzI2MDg4 | User | false |
huggingface/transformers | 4,483,351,035 | I_kwDOCUB6oc8AAAABCzqB-w | 46,097 | https://github.com/huggingface/transformers/issues/46097 | https://api.github.com/repos/huggingface/transformers/issues/46097 | Path Traversal in Sharded Checkpoint Loader via Unsanitized `weight_map` Entries in `*.index.json` | ### System Info
Details
The vulnerable code is in get_checkpoint_shard_files in hub.py. When loading a sharded checkpoint from a local directory, the function reads an index JSON file and extracts shard filenames from the weight_map field without any validation:
with open(index_filename) as f:
index = json.loads(... | closed | completed | false | 7 | [
"bug"
] | [] | 2026-05-20T05:12:29Z | 2026-06-28T08:59:16Z | 2026-06-28T08:59:16Z | NONE | null | 20260628T180015Z | 2026-06-28T18:00:15Z | karnakarreddi | 15,706,601 | MDQ6VXNlcjE1NzA2NjAx | User | false |
huggingface/transformers | 4,487,784,893 | I_kwDOCUB6oc8AAAABC34pvQ | 46,121 | https://github.com/huggingface/transformers/issues/46121 | https://api.github.com/repos/huggingface/transformers/issues/46121 | `convert_rope_params_to_dict` raises `TypeError` when `ignore_keys_at_rope_validation` is a JSON-loaded list | ### System Info
- transformers: 5.8.1
- Python: 3.14.0
- OS: macOS (reproduced locally; same class of failure reported with vLLM + Qwen3.5 merged HF checkpoints on Linux eval jobs)
- Model family: Qwen3_5TextConfig / Qwen3.5
### Who can help?
@ArthurZucker @Cyrilvallez
### Information
- [ ] The official example s... | closed | completed | false | 3 | [
"bug"
] | [] | 2026-05-20T15:30:41Z | 2026-05-25T18:39:49Z | 2026-05-25T18:39:49Z | CONTRIBUTOR | null | 20260526T000122Z | 2026-05-26T00:01:22Z | Charly21r | 208,873,065 | U_kgDODHMmaQ | User | false |
huggingface/transformers | 4,488,926,219 | I_kwDOCUB6oc8AAAABC4-UCw | 46,123 | https://github.com/huggingface/transformers/issues/46123 | https://api.github.com/repos/huggingface/transformers/issues/46123 | MaskGenerationPipeline: is_last never True on final partial batch, silently dropping results | ### System Info
transformers version: current main
Affected file: src/transformers/pipelines/mask_generation.py
### 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/SQuA... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-05-20T18:25:00Z | 2026-05-25T15:40:53Z | 2026-05-25T15:40:53Z | CONTRIBUTOR | null | 20260525T180030Z | 2026-05-25T18:00:30Z | J3r3myPerera | 115,147,492 | U_kgDOBt0C5A | User | false |
huggingface/transformers | 4,490,099,153 | I_kwDOCUB6oc8AAAABC6F50Q | 46,129 | https://github.com/huggingface/transformers/issues/46129 | https://api.github.com/repos/huggingface/transformers/issues/46129 | [deepseek_v4] conversion_mapping doesn't cover mtp.* paths — MTP keys silently random-init even after _keys_to_ignore is empty | ## Summary
`transformers.conversion_mapping.get_checkpoint_conversion_mapping("deepseek_v4")` returns 41 `WeightRenaming` entries that rename upstream-internal naming to HF naming (`attn.` → `self_attn.`, `ffn.` → `mlp.`, `attn_norm.` → `input_layernorm.`, `attn.wq_a.` → `self_attn.q_a_proj.`, `attn.attn_sink` → `self... | closed | completed | false | 5 | [] | [] | 2026-05-20T21:39:47Z | 2026-06-29T09:30:15Z | 2026-06-29T09:30:15Z | NONE | null | 20260629T120018Z | 2026-06-29T12:00:18Z | pasta-paul | 149,531,089 | U_kgDOCOmp0Q | User | false |
huggingface/transformers | 4,490,552,667 | I_kwDOCUB6oc8AAAABC6hlWw | 46,132 | https://github.com/huggingface/transformers/issues/46132 | https://api.github.com/repos/huggingface/transformers/issues/46132 | AttentionInterface.register changes behavior of registered function | ### System Info
`transformers env` fails with `NameError: name 'CompletionCreateParamsStreaming' is not defined`
I am running Ubuntu 25.10, Python 3.13.7, and pytorch 2.11.0
### Who can help?
@Cyrilvallez
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An offici... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-05-20T23:24:31Z | 2026-05-21T14:44:37Z | 2026-05-21T14:44:37Z | NONE | null | 20260521T180115Z | 2026-05-21T18:01:15Z | pjc15111 | 243,856,361 | U_kgDODojz6Q | User | false |
huggingface/transformers | 4,490,749,494 | I_kwDOCUB6oc8AAAABC6tmNg | 46,133 | https://github.com/huggingface/transformers/issues/46133 | https://api.github.com/repos/huggingface/transformers/issues/46133 | Add TIPSv2 (Advancing Vision-Language Pretraining with Enhanced Patch-Text Alignment) by Google DeepMind | ### Model description
TIPSv2 is a vision-language encoder that addresses the problem of dense alignment between image regions and text through modification in the pretraining recipe (iBOT++, multi-granularity synthetic captions, and memory-saving head-only EMA scheme). Across 9 tasks and 20 datasets the resulting mode... | open | null | false | 15 | [
"New model"
] | [] | 2026-05-21T00:17:16Z | 2026-06-02T14:40:51Z | null | CONTRIBUTOR | null | 20260602T180035Z | 2026-06-02T18:00:35Z | farrosalferro | 127,369,839 | U_kgDOB5eCbw | User | false |
huggingface/transformers | 4,493,811,151 | I_kwDOCUB6oc8AAAABC9odzw | 46,139 | https://github.com/huggingface/transformers/issues/46139 | https://api.github.com/repos/huggingface/transformers/issues/46139 | Discussion: optional RankSEG-style decoding for Transformers semantic segmentation post-processing | [](https://github.com/rankseg/rankseg) [](https://pypi.org/project/rankseg/) [](https://rankseg.readthedocs.io/en/latest/) [![Transf... | open | null | false | 8 | [] | [] | 2026-05-21T10:36:05Z | 2026-06-12T05:54:22Z | null | NONE | null | 20260612T060053Z | 2026-06-12T06:00:53Z | Leev1s | 83,260,380 | MDQ6VXNlcjgzMjYwMzgw | User | false |
huggingface/transformers | 4,494,858,162 | I_kwDOCUB6oc8AAAABC-oXsg | 46,143 | https://github.com/huggingface/transformers/issues/46143 | https://api.github.com/repos/huggingface/transformers/issues/46143 | `**kwargs` not passed through methods of RoFormer models | Hi,
when working with RoFormer models, I've noticed the `**kwargs` option is not handled correctly. Most classes take in `**kwargs` but do not pass them further into the model. For example: https://github.com/huggingface/transformers/blob/5206626c48710e69fef3eadfba077cada99f37bb/src/transformers/models/roformer/modeli... | closed | completed | false | 1 | [] | [] | 2026-05-21T13:18:46Z | 2026-05-25T13:21:30Z | 2026-05-25T13:21:30Z | CONTRIBUTOR | null | 20260525T180030Z | 2026-05-25T18:00:30Z | ir2718 | 94,498,051 | U_kgDOBaHtAw | User | false |
huggingface/transformers | 4,494,912,849 | I_kwDOCUB6oc8AAAABC-rtUQ | 46,144 | https://github.com/huggingface/transformers/issues/46144 | https://api.github.com/repos/huggingface/transformers/issues/46144 | RoFormer attention implementation does not use attention interface | Hi,
I want to use RoFormer with a custom attention implementation. However, the current code relies on an eager implementation without using the attention interface: https://github.com/huggingface/transformers/blob/5206626c48710e69fef3eadfba077cada99f37bb/src/transformers/models/roformer/modeling_roformer.py#L196-L211... | open | null | false | 5 | [
"Feature request"
] | [] | 2026-05-21T13:26:50Z | 2026-06-02T18:45:26Z | null | CONTRIBUTOR | null | 20260603T000024Z | 2026-06-03T00:00:24Z | ir2718 | 94,498,051 | U_kgDOBaHtAw | User | false |
huggingface/transformers | 4,499,935,935 | I_kwDOCUB6oc8AAAABDDeSvw | 46,150 | https://github.com/huggingface/transformers/issues/46150 | https://api.github.com/repos/huggingface/transformers/issues/46150 | [Feature Proposal] Cognitive Weighting Protocol: Hierarchical context management to mitigate "Contextual Drift" in long-running sessions. | Context: Long-context LLM sessions suffer from "drift," where tactical/procedural data overwhelms foundational architectural constraints.
The Proposal: A hierarchical weighting protocol at the LLM integration layer (middleware):
Foundational Layer: Fixed high-weight context (Static/Blueprint).
Strategic Layer: Summa... | closed | completed | false | 2 | [] | [] | 2026-05-22T04:55:09Z | 2026-05-22T09:52:50Z | 2026-05-22T09:52:17Z | NONE | null | 20260522T120016Z | 2026-05-22T12:00:16Z | Nossari | 93,357,478 | U_kgDOBZCFpg | User | false |
huggingface/transformers | 810,165,137 | MDU6SXNzdWU4MTAxNjUxMzc= | 10,232 | https://github.com/huggingface/transformers/issues/10232 | https://api.github.com/repos/huggingface/transformers/issues/10232 | Multilabel Sequence Classification in trainer | # 🚀 Feature request
We need to be able to use the trainer for multilabel classification problems.
## Motivation
Right now we create our models in the old fashioned way, with a sigmoid layer at the end so we can do multilabel. However if we could use the trainer directly, we wouldn't need to maintain different... | closed | completed | false | 16 | [
"Feature request"
] | [] | 2021-02-17T13:02:34Z | 2026-05-22T07:52:13Z | 2026-05-22T07:52:13Z | CONTRIBUTOR | null | 20260522T120016Z | 2026-05-22T12:00:16Z | GuillemGSubies | 37,592,763 | MDQ6VXNlcjM3NTkyNzYz | User | false |
huggingface/transformers | 848,543,462 | MDU6SXNzdWU4NDg1NDM0NjI= | 11,019 | https://github.com/huggingface/transformers/issues/11019 | https://api.github.com/repos/huggingface/transformers/issues/11019 | Enable multiple `eval_dataset` in `Trainer` API | # 🚀 Feature request
Allow for two or more (equally long) validation sets to be passed to the `Trainer` API which are evaluated sequentially each `eval_steps`.
## Motivation
You can find my motivation in this [thread](https://discuss.huggingface.co/t/use-trainer-api-with-two-valiation-sets/5212) and the refere... | closed | completed | false | 5 | [
"Feature request"
] | [] | 2021-04-01T14:57:11Z | 2026-05-22T09:53:37Z | 2026-05-22T09:53:37Z | NONE | null | 20260522T120016Z | 2026-05-22T12:00:16Z | simonschoe | 53,626,067 | MDQ6VXNlcjUzNjI2MDY3 | User | false |
huggingface/transformers | 847,400,200 | MDU6SXNzdWU4NDc0MDAyMDA= | 11,005 | https://github.com/huggingface/transformers/issues/11005 | https://api.github.com/repos/huggingface/transformers/issues/11005 | ReduceLROnPlateau-like functionality? | # 🚀 Feature request
<!-- A clear and concise description of the feature proposal.
Please provide a link to the paper and code in case they exist. -->
**Description:** Dynamic learning rate reduction upon metric saturation, as in `torch.optim.lr_scheduler.ReduceLROnPlateau`, integrated into the `Trainer` AP... | closed | completed | false | 1 | [
"Feature request"
] | [] | 2021-03-31T21:50:33Z | 2026-05-22T09:53:58Z | 2026-05-22T09:53:58Z | NONE | null | 20260522T120016Z | 2026-05-22T12:00:16Z | tchang1997 | 30,159,285 | MDQ6VXNlcjMwMTU5Mjg1 | User | false |
huggingface/transformers | 822,237,938 | MDU6SXNzdWU4MjIyMzc5Mzg= | 10,519 | https://github.com/huggingface/transformers/issues/10519 | https://api.github.com/repos/huggingface/transformers/issues/10519 | Adding option to truncation from beginning instead of end, for both longest_first and longest_second | # 🚀 Feature request
<!-- A clear and concise description of the feature proposal.
Please provide a link to the paper and code in case they exist. -->
The current [truncation strategies](https://huggingface.co/transformers/preprocessing.html#everything-you-always-wanted-to-know-about-padding-and-truncation)... | closed | completed | false | 7 | [
"Feature request"
] | [] | 2021-03-04T15:27:07Z | 2026-05-22T09:54:08Z | 2026-05-22T09:54:08Z | NONE | null | 20260522T120016Z | 2026-05-22T12:00:16Z | jatinganhotra | 307,606 | MDQ6VXNlcjMwNzYwNg== | User | false |
huggingface/transformers | 815,308,742 | MDU6SXNzdWU4MTUzMDg3NDI= | 10,371 | https://github.com/huggingface/transformers/issues/10371 | https://api.github.com/repos/huggingface/transformers/issues/10371 | Load pretrained model except the head layer for a specific downstream task | # 🚀 Feature request
It would be nice to have a flag for `from_pretrained` method that indicates whether to load last layer or not. This feature is needed for transfer learning.
## Motivation
I have trained a model with a specific dataset for a downstream task. Now, I need to train another model that needs to ... | closed | completed | false | 7 | [
"Feature request"
] | [] | 2021-02-24T09:50:17Z | 2026-05-22T09:54:20Z | 2026-05-22T09:54:19Z | CONTRIBUTOR | null | 20260522T120016Z | 2026-05-22T12:00:16Z | hasansalimkanmaz | 49,716,619 | MDQ6VXNlcjQ5NzE2NjE5 | User | false |
huggingface/transformers | 4,500,701,713 | I_kwDOCUB6oc8AAAABDENCEQ | 46,153 | https://github.com/huggingface/transformers/issues/46153 | https://api.github.com/repos/huggingface/transformers/issues/46153 | Regression in v5.x: `ProcessorMixin._load_tokenizer_from_pretrained` forces subfolder for non-primary sub-tokenizers, breaking repos that put tokenizer files at root | ## Summary
`ProcessorMixin._load_tokenizer_from_pretrained` in v5.x unconditionally appends the
sub-processor attribute name as a subfolder when loading a non-primary tokenizer:
```python
# src/transformers/processing_utils.py, ~L1457-1467
is_primary = sub_processor_type == "tokenizer"
if is_primary:
tokenizer = ... | closed | completed | false | 3 | [] | [] | 2026-05-22T07:17:44Z | 2026-06-29T09:30:11Z | 2026-06-29T09:30:11Z | NONE | null | 20260629T120018Z | 2026-06-29T12:00:18Z | HanfengLiao | 97,389,715 | U_kgDOBc4Mkw | User | false |
huggingface/transformers | 4,501,987,421 | I_kwDOCUB6oc8AAAABDFbgXQ | 46,159 | https://github.com/huggingface/transformers/issues/46159 | https://api.github.com/repos/huggingface/transformers/issues/46159 | [RT-DETRv2] MPS crash: build_2d_sinusoidal_position_embedding hardcodes torch.float64, breaking Apple Silicon / MPS inference | ### System Info
```
- `transformers` version: 5.9.0
- Platform: macOS (Apple Silicon, MPS backend)
- Python version: 3.13
- PyTorch version: (MPS-enabled build)
- `docling` version: 2.95.0
- `docling-ibm-models` version: 3.13.2
```
### Reproduction
Run `RTDetrV2ForObjectDetection` inference on any Apple Silicon Mac ... | open | reopened | false | 5 | [
"bug"
] | [] | 2026-05-22T10:45:10Z | 2026-06-20T04:02:08Z | null | NONE | null | 20260620T060014Z | 2026-06-20T06:00:14Z | shubhammr21 | 10,411,979 | MDQ6VXNlcjEwNDExOTc5 | User | false |
huggingface/transformers | 4,504,804,600 | I_kwDOCUB6oc8AAAABDIHc-A | 46,165 | https://github.com/huggingface/transformers/issues/46165 | https://api.github.com/repos/huggingface/transformers/issues/46165 | Unable to resume model training with a locally downloaded model and trust remote code | ### System Info
requires-python = ">=3.12"
dependencies = [
"accelerate>=1.13.0",
"datasets>=4.8.5",
"sentence-transformers>=5.4.1",
"torch>=2.11.0",
"transformers>=5.8.0",
]
### Who can help?
@tomaszcichy98
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### ... | open | null | false | 5 | [
"bug"
] | [] | 2026-05-22T18:27:56Z | 2026-06-30T14:55:27Z | null | NONE | null | 20260630T180022Z | 2026-06-30T18:00:22Z | conceptofmind | 25,208,228 | MDQ6VXNlcjI1MjA4MjI4 | User | false |
huggingface/transformers | 4,506,174,720 | I_kwDOCUB6oc8AAAABDJbFAA | 46,167 | https://github.com/huggingface/transformers/issues/46167 | https://api.github.com/repos/huggingface/transformers/issues/46167 | [deepseek_v4] save_pretrained silently downcasts FP32 tensors to BF16 (hc_*, attn_sink, ffn.gate.bias, compressor.ape, indexer.compressor.ape) | ## Summary
`save_pretrained` silently downcasts 417 FP32 tensors to BF16 when saving a DeepSeek-V4 model loaded with `torch_dtype=torch.bfloat16`. No warning, no error. The downcast loses numerical precision on plumbing tensors that DeepSeek's release spec keeps at FP32 for a reason.
## Affected tensor groups
Per De... | closed | completed | false | 3 | [] | [] | 2026-05-23T00:00:23Z | 2026-05-27T09:49:06Z | 2026-05-27T09:49:06Z | NONE | null | 20260527T180027Z | 2026-05-27T18:00:27Z | pasta-paul | 149,531,089 | U_kgDOCOmp0Q | User | false |
huggingface/transformers | 3,097,109,901 | I_kwDOCUB6oc64miWN | 38,435 | https://github.com/huggingface/transformers/issues/38435 | https://api.github.com/repos/huggingface/transformers/issues/38435 | [i18n-ro] Translating docs to Romanian | Hi!
Let's bring the documentation to all the Romanian-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 to t... | closed | completed | false | 3 | [
"WIP"
] | [] | 2025-05-28T12:01:48Z | 2026-05-29T17:49:44Z | 2026-05-29T17:49:44Z | NONE | null | 20260529T180028Z | 2026-05-29T18:00:28Z | zero-point | 4,007,299 | MDQ6VXNlcjQwMDcyOTk= | User | false |
huggingface/transformers | 4,509,119,867 | I_kwDOCUB6oc8AAAABDMO1ew | 46,172 | https://github.com/huggingface/transformers/issues/46172 | https://api.github.com/repos/huggingface/transformers/issues/46172 | [BUG] deepseek-v4 `comb.to(dtype).transpose(-1, -2)` | ### System Info
Expected fix:`comb.to(dtype).transpose(-1, -2)` -> `comb.to(dtype)`
https://github.com/huggingface/transformers/blame/10555512868d663ee1ff627e4f5c5c260114235b/src/transformers/models/deepseek_v4/modular_deepseek_v4.py#L998-L999
https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/main/inference... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-05-23T17:48:56Z | 2026-05-23T17:55:14Z | 2026-05-23T17:55:14Z | CONTRIBUTOR | null | 20260523T180021Z | 2026-05-23T18:00:21Z | Jintao-Huang | 45,290,347 | MDQ6VXNlcjQ1MjkwMzQ3 | User | false |
huggingface/transformers | 4,508,239,093 | I_kwDOCUB6oc8AAAABDLZE9Q | 46,170 | https://github.com/huggingface/transformers/issues/46170 | https://api.github.com/repos/huggingface/transformers/issues/46170 | ImportError / TypeError on Windows with AMD ROCm PyTorch due to torch.distributed dependency | ### System Info
- Transformers version: 5.9.0
- Platform: Windows-11-10.0.26200-SP0
- Python version: 3.12.10
- PyTorch version: 2.9.1+rocm7.2.1
- CUDA/ROCm available: True
- GPU Name: AMD Radeon RX 7600
### Who can help?
@ivarflakstad
### Information
- [ ] The official example scripts
- [ ] My own modified scrip... | closed | completed | false | 3 | [
"bug"
] | [
"remi-or"
] | 2026-05-23T13:06:49Z | 2026-05-29T07:41:40Z | 2026-05-29T07:41:40Z | NONE | null | 20260529T120035Z | 2026-05-29T12:00:35Z | mjakob-gh | 19,156,568 | MDQ6VXNlcjE5MTU2NTY4 | User | false |
huggingface/transformers | 4,509,919,409 | I_kwDOCUB6oc8AAAABDM_osQ | 46,175 | https://github.com/huggingface/transformers/issues/46175 | https://api.github.com/repos/huggingface/transformers/issues/46175 | [Gemma4] `Gemma4VisionPatchEmbedder._position_embeddings` materializes a ~19 GiB one-hot tensor that's mathematically a 2-row embedding lookup | ### System Info
- `transformers` version: 5.6.2
- Platform: Linux 5.15.0-41-generic (x86_64)
- Python: 3.12
- PyTorch: 2.10.0+cu128
- GPU: 8 × NVIDIA H800 (80GB)
- Model: `google/gemma-4-E4B-it` (any Gemma 4 multimodal variant)
- Trainer: ZeRO-3 (DeepSpeed) full-parameter SFT, `torch_dtype: bfloat16`
### Who can help... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-05-23T23:42:01Z | 2026-05-28T11:52:53Z | 2026-05-28T11:52:53Z | NONE | null | 20260528T120029Z | 2026-05-28T12:00:29Z | kuso2006 | 109,347,845 | U_kgDOBoSEBQ | User | false |
huggingface/transformers | 4,511,333,730 | I_kwDOCUB6oc8AAAABDOV9Yg | 46,179 | https://github.com/huggingface/transformers/issues/46179 | https://api.github.com/repos/huggingface/transformers/issues/46179 | [New model] Add Fun-ASR-Nano (FunAudioLLM/Fun-ASR-Nano-2512) | ### Model description
**Fun-ASR-Nano** is an 800M-parameter end-to-end speech recognition model from Alibaba DAMO Academy (FunAudioLLM team). It achieves state-of-the-art ASR performance, outperforming Whisper-large-v3 (1.6B) while being half the size.
**Architecture:**
- **Audio Encoder**: SenseVoiceEncoderSmall (SA... | closed | completed | false | 2 | [] | [] | 2026-05-24T11:07:04Z | 2026-07-03T09:00:17Z | 2026-07-03T09:00:17Z | NONE | null | 20260703T120023Z | 2026-07-03T12:00:23Z | LauraGPT | 18,321,252 | MDQ6VXNlcjE4MzIxMjUy | User | false |
huggingface/transformers | 4,511,321,910 | I_kwDOCUB6oc8AAAABDOVPNg | 46,178 | https://github.com/huggingface/transformers/issues/46178 | https://api.github.com/repos/huggingface/transformers/issues/46178 | ValueError: Backend should be defined in the BACKENDS_MAPPING. Offending backend: tensorflow_text | ### System Info
I don't have tensorflow or tensorboard installed. Updating from version 5.8.0 to version 5.9.0 causes an error when launching ComfyUI on Linux:
Traceback (most recent call last):
File ".../ComfyUI/main.py", line 198, in <module>
import execution
File ".../ComfyUI/execution.py", line 20, in <mo... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-05-24T11:02:55Z | 2026-07-03T09:00:19Z | 2026-07-03T09:00:19Z | NONE | null | 20260703T120023Z | 2026-07-03T12:00:23Z | marchcat69 | 67,062,782 | MDQ6VXNlcjY3MDYyNzgy | User | false |
huggingface/transformers | 4,513,813,105 | I_kwDOCUB6oc8AAAABDQtScQ | 46,186 | https://github.com/huggingface/transformers/issues/46186 | https://api.github.com/repos/huggingface/transformers/issues/46186 | [New Model] Add TIGER — Recommender Systems with Generative Retrieval (Google DeepMind, NeurIPS 2023) | ## Model Description
**Paper:** [Recommender Systems with Generative Retrieval](https://arxiv.org/abs/2305.05065)
**Authors:** Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, Maheswaran Sathiamoorthy
**Af... | closed | completed | false | 1 | [] | [] | 2026-05-25T02:19:54Z | 2026-05-25T12:11:02Z | 2026-05-25T12:11:02Z | NONE | null | 20260525T180030Z | 2026-05-25T18:00:30Z | raviteja-ganta | 12,708,947 | MDQ6VXNlcjEyNzA4OTQ3 | User | false |
huggingface/transformers | 4,513,812,655 | I_kwDOCUB6oc8AAAABDQtQrw | 46,185 | https://github.com/huggingface/transformers/issues/46185 | https://api.github.com/repos/huggingface/transformers/issues/46185 | [New Model] Add Microsoft Samba - Hybrid SSM + Sliding Window Attention | ## Model Description
**Paper:** [Samba: Simple Hybrid State Space Models for Efficient Unlimited Context Language Modeling](https://arxiv.org/abs/2406.07522)
**Authors:** Liliang Ren, Yang Liu, Yadong Lu, Yelong Shen, Chen Liang, Weizhu Chen
**Affiliations:** Microsoft Research; University of Illinois at Urbana-Champa... | closed | completed | false | 1 | [] | [] | 2026-05-25T02:19:44Z | 2026-05-25T12:38:11Z | 2026-05-25T12:37:42Z | NONE | null | 20260525T180030Z | 2026-05-25T18:00:30Z | raviteja-ganta | 12,708,947 | MDQ6VXNlcjEyNzA4OTQ3 | User | false |
huggingface/transformers | 4,513,829,702 | I_kwDOCUB6oc8AAAABDQuTRg | 46,187 | https://github.com/huggingface/transformers/issues/46187 | https://api.github.com/repos/huggingface/transformers/issues/46187 | cohere2_moe fails training + tensor parallel tests | ### Reproduction
Training tests:
```
FAILED tests/models/cohere2_moe/test_modeling_cohere2_moe.py::Cohere2MoeModelTest::test_training_overfit - AssertionError: 0.27068585289520714 not greater than 0.9 : Expected loss to decrease by at least 90%, got 27.1%
```
but generation is the same
```
[transformers] The attent... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-05-25T02:25:30Z | 2026-05-26T15:31:02Z | 2026-05-26T15:31:02Z | MEMBER | null | 20260526T180028Z | 2026-05-26T18:00:28Z | 3outeille | 47,445,085 | MDQ6VXNlcjQ3NDQ1MDg1 | User | false |
huggingface/transformers | 4,514,003,112 | I_kwDOCUB6oc8AAAABDQ44qA | 46,190 | https://github.com/huggingface/transformers/issues/46190 | https://api.github.com/repos/huggingface/transformers/issues/46190 | Qwen3.5 GatedDeltaNet: Large logit divergence between full-sequence forward and prefill+decode with cache | ### System Info
# Qwen3.5 GatedDeltaNet: Large logit divergence between full-sequence forward and prefill+decode with cache
## System Info
- `transformers` version: 5.9.0
- Platform: macOS (Apple Silicon, arm64)
- Python version: 3.13
- PyTorch version: 2.12.0 (CPU)
- Tokenizers version: 0.22.2
- Using GPU in script... | open | null | false | 7 | [
"bug"
] | [] | 2026-05-25T03:20:38Z | 2026-06-30T09:01:22Z | null | NONE | null | 20260630T120021Z | 2026-06-30T12:00:21Z | JWQZ | 74,125,807 | MDQ6VXNlcjc0MTI1ODA3 | User | false |
huggingface/transformers | 4,515,097,812 | I_kwDOCUB6oc8AAAABDR7s1A | 46,192 | https://github.com/huggingface/transformers/issues/46192 | https://api.github.com/repos/huggingface/transformers/issues/46192 | return_tensors is silently ignored when text_kwargs is explicitly passed | ### System Info
- `transformers` version: 5.10.0.dev0
- Platform: Linux-5.15.0-117-generic-x86_64-with-glibc2.35
- Python version: 3.12.11
- Huggingface_hub version: 1.16.1
- Safetensors version: 0.5.3
- Accelerate version: not installed
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch versi... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-05-25T07:24:14Z | 2026-05-29T13:34:18Z | 2026-05-29T13:34:18Z | NONE | null | 20260529T180028Z | 2026-05-29T18:00:28Z | leveretconey | 25,177,975 | MDQ6VXNlcjI1MTc3OTc1 | User | false |
huggingface/transformers | 4,517,000,634 | I_kwDOCUB6oc8AAAABDTv1ug | 46,200 | https://github.com/huggingface/transformers/issues/46200 | https://api.github.com/repos/huggingface/transformers/issues/46200 | HIGH: Pickle model loading (torch.load) enabled by default — no warning when loading untrusted weights | ## CWE-502: Deserialization of Untrusted Data — Pickle model loading enabled by default
### Severity: HIGH (CVSS 7.8)
### Location
`src/transformers/modeling_utils.py` — `from_pretrained()` default behavior
### Description
The SECURITY.md explicitly warns:
> "We heavily recommend uploading and downloading models in... | closed | completed | false | 0 | [
"Code agent slop"
] | [] | 2026-05-25T12:46:54Z | 2026-05-25T13:07:05Z | 2026-05-25T13:06:57Z | NONE | null | 20260525T180030Z | 2026-05-25T18:00:30Z | olliekirbyz548-ui | 270,104,311 | U_kgDOEBl29w | User | false |
huggingface/transformers | 4,517,001,375 | I_kwDOCUB6oc8AAAABDTv4nw | 46,201 | https://github.com/huggingface/transformers/issues/46201 | https://api.github.com/repos/huggingface/transformers/issues/46201 | MEDIUM: All CI jobs use secrets:inherit — full org secrets passed to every reusable workflow | ## CWE-1357: Improper Access Control — secrets:inherit across all CI jobs
### Severity: MEDIUM (CVSS 5.3)
### Location
`.github/workflows/self-scheduled-caller.yml` — all jobs use `secrets: inherit`
### Description
Every CI job in the daily scheduled workflow inherits the full organization secrets:
```yaml
model-ci... | closed | completed | false | 0 | [
"Code agent slop"
] | [] | 2026-05-25T12:47:01Z | 2026-05-25T13:07:15Z | 2026-05-25T13:07:10Z | NONE | null | 20260525T180030Z | 2026-05-25T18:00:30Z | olliekirbyz548-ui | 270,104,311 | U_kgDOEBl29w | User | false |
huggingface/transformers | 4,517,132,888 | I_kwDOCUB6oc8AAAABDT36WA | 46,202 | https://github.com/huggingface/transformers/issues/46202 | https://api.github.com/repos/huggingface/transformers/issues/46202 | Roundtrip Failure for Gemma Pipeline on "▁" | ### System Info
Arch Linux
Python 3.14.4
Transformers 5.9.0
### Who can help?
@ArthurZucker
@itazap
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give d... | closed | completed | false | 6 | [
"bug"
] | [] | 2026-05-25T13:08:53Z | 2026-05-28T13:27:34Z | 2026-05-28T11:23:05Z | NONE | null | 20260528T180037Z | 2026-05-28T18:00:37Z | julmb | 11,631,652 | MDQ6VXNlcjExNjMxNjUy | User | false |
huggingface/transformers | 4,520,762,342 | I_kwDOCUB6oc8AAAABDXVb5g | 46,206 | https://github.com/huggingface/transformers/issues/46206 | https://api.github.com/repos/huggingface/transformers/issues/46206 | There is a PR in the queue Reverting security patches | PR 46194 is submitting a user doc but is reverting a serious security patch and removing tests. analysis: https://www.lenzon.ai/viewer/cmplyxggf0000nn4u4d5x4q5p?voice=google-neural2 | closed | completed | false | 1 | [] | [] | 2026-05-26T02:11:52Z | 2026-05-26T11:36:13Z | 2026-05-26T11:35:05Z | NONE | null | 20260526T120027Z | 2026-05-26T12:00:27Z | rickgonzalez | 9,344,125 | MDQ6VXNlcjkzNDQxMjU= | User | false |
huggingface/transformers | 4,521,007,930 | I_kwDOCUB6oc8AAAABDXkbOg | 46,209 | https://github.com/huggingface/transformers/issues/46209 | https://api.github.com/repos/huggingface/transformers/issues/46209 | [BUG] FP8 model cannot inference on B200 (SM>=100) | ### System Info
transformers==5.9.0
kernels==0.14.1
### Who can help?
@SunMarc @MekkCyber
### Information
- [x] The official example scripts
- [ ] My own modified scripts
### Tasks
- [x] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details bel... | open | null | false | 5 | [
"bug"
] | [] | 2026-05-26T03:08:27Z | 2026-07-03T01:05:21Z | null | CONTRIBUTOR | null | 20260703T060023Z | 2026-07-03T06:00:23Z | xin3he | 83,260,933 | MDQ6VXNlcjgzMjYwOTMz | User | false |
huggingface/transformers | 4,523,752,848 | I_kwDOCUB6oc8AAAABDaL9kA | 46,216 | https://github.com/huggingface/transformers/issues/46216 | https://api.github.com/repos/huggingface/transformers/issues/46216 | is_torch_hpu_available permanently monkeypatches global torch functions | `is_torch_hpu_available` in import_utils.py permanently monkeypatches global torch functions
```python
torch.gather = patched_gather
torch.Tensor.gather = patched_gather
torch.take_along_dim = patched_take_along_dim
torch.linalg.cholesky = safe_cholesky # adds jitter when NaN detected
torch.scatter = patched_scatter... | closed | completed | false | 3 | [] | [] | 2026-05-26T11:17:41Z | 2026-07-05T08:58:49Z | 2026-07-05T08:58:49Z | NONE | null | 20260705T120019Z | 2026-07-05T12:00:19Z | olliezyuhscripta | 178,087,230 | U_kgDOCp1lPg | User | false |
huggingface/transformers | 4,523,716,837 | I_kwDOCUB6oc8AAAABDaJw5Q | 46,215 | https://github.com/huggingface/transformers/issues/46215 | https://api.github.com/repos/huggingface/transformers/issues/46215 | _sanitize_module_name doesnt handle path traversal chars | _sanitize_module_name in dynamic_module_utils.py only replaces `.` and `-` before using the result in file paths
```python
new_name = name.replace(".", "_dot_").replace("-", "_hyphen_")
```
but repo names from the hub could potentially contain other characters that would break or escape the intended path. like if a ... | open | null | false | 5 | [] | [] | 2026-05-26T11:10:24Z | 2026-06-28T08:59:08Z | null | NONE | null | 20260628T180015Z | 2026-06-28T18:00:15Z | olliezyuhscripta | 178,087,230 | U_kgDOCp1lPg | User | false |
huggingface/transformers | 4,523,767,303 | I_kwDOCUB6oc8AAAABDaM2Bw | 46,217 | https://github.com/huggingface/transformers/issues/46217 | https://api.github.com/repos/huggingface/transformers/issues/46217 | convert_file_size_to_int crashes on bare unit string like GIB | convert_file_size_to_int in hub.py doesnt handle edge case where someone passes just the unit without a number
if size.upper().endswith("GIB"):
return int(size[:-3]) * (2**30)
if size is literally "GIB" then size[:-3] is "" and int("") raises ValueError. same for MIB, KIB, GB, MB, KB
super minor, probably never... | closed | completed | false | 1 | [] | [] | 2026-05-26T11:20:39Z | 2026-05-26T11:52:07Z | 2026-05-26T11:51:57Z | NONE | null | 20260526T120027Z | 2026-05-26T12:00:27Z | olliezyuhscripta | 178,087,230 | U_kgDOCp1lPg | User | false |
huggingface/transformers | 4,524,285,614 | I_kwDOCUB6oc8AAAABDaserg | 46,218 | https://github.com/huggingface/transformers/issues/46218 | https://api.github.com/repos/huggingface/transformers/issues/46218 | Could not import module 'Qwen3_5ForCausalLM | ### System Info
unable to run https://huggingface.co/NemoStation/Marlin-2B
tried the given requirements as well as transformers==5.9.0
### Who can help?
_No response_
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` f... | closed | completed | false | 5 | [
"bug"
] | [] | 2026-05-26T12:59:38Z | 2026-07-05T08:58:48Z | 2026-07-05T08:58:48Z | NONE | null | 20260705T120019Z | 2026-07-05T12:00:19Z | Arslan-Mehmood1 | 51,626,734 | MDQ6VXNlcjUxNjI2NzM0 | User | false |
huggingface/transformers | 4,525,239,390 | I_kwDOCUB6oc8AAAABDbmsXg | 46,222 | https://github.com/huggingface/transformers/issues/46222 | https://api.github.com/repos/huggingface/transformers/issues/46222 | 1 ขอให้เจอสิ่งดีๆ คิดดี ทำดี เจอแต่สิ่งดีๆ สาธุ | <!--
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](https://github.com/hug... | open | null | false | 2 | [
"WIP"
] | [] | 2026-05-26T15:07:55Z | 2026-06-28T06:09:49Z | null | NONE | null | 20260628T180015Z | 2026-06-28T18:00:15Z | imgoingbacktheonlything239-ops | 269,549,381 | U_kgDOEBD_RQ | User | false |
huggingface/transformers | 4,529,001,548 | I_kwDOCUB6oc8AAAABDfMUTA | 46,226 | https://github.com/huggingface/transformers/issues/46226 | https://api.github.com/repos/huggingface/transformers/issues/46226 | 1 ขอให้เจอสิ่งดีๆ คิดดี ทำดี เจอแต่สิ่งดีๆ สาธุ | <!--
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](https://github.com/hug... | closed | completed | false | 0 | [
"WIP"
] | [] | 2026-05-27T02:34:00Z | 2026-05-29T00:08:38Z | 2026-05-29T00:08:38Z | NONE | null | 20260529T060141Z | 2026-05-29T06:01:41Z | imgoingbacktheonlything239-ops | 269,549,381 | U_kgDOEBD_RQ | User | false |
huggingface/transformers | 4,529,320,386 | I_kwDOCUB6oc8AAAABDffxwg | 46,228 | https://github.com/huggingface/transformers/issues/46228 | https://api.github.com/repos/huggingface/transformers/issues/46228 | ImportError: cannot import name 'AutoModelWithLMHead' from 'transformers' | Hi, I am in a virtual conda environment, trying to import AutoModelWithLMHead, AutoTokenizer, and EsmModel. I can't change the models being used, because I am trying to recreate a peer-reviewed source model code. I have installed pytorch (2.12.0), and transformers (5.10.0.dev0). When I go to load in the requisite packa... | closed | completed | false | 3 | [] | [] | 2026-05-27T03:45:28Z | 2026-07-05T08:58:46Z | 2026-07-05T08:58:46Z | NONE | null | 20260705T120019Z | 2026-07-05T12:00:19Z | hannahlhargrove | 92,232,047 | U_kgDOBX9Zbw | User | false |
huggingface/transformers | 4,530,391,154 | I_kwDOCUB6oc8AAAABDghIcg | 46,231 | https://github.com/huggingface/transformers/issues/46231 | https://api.github.com/repos/huggingface/transformers/issues/46231 | DOC: set_training() documents parameter as `num_train_epochs` but actual parameter name is `num_epochs` | ### System Info
## Description
In `TrainingArguments.set_training()`, the `Args` section of the docstring documents the parameter as `num_train_epochs` but the actual function parameter is `num_epochs`.
## Location
File: `src/transformers/training_args.py`, around line 2183
**Function signature (actual parameter):... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-05-27T07:20:35Z | 2026-05-28T13:13:36Z | 2026-05-28T13:13:35Z | NONE | null | 20260528T180037Z | 2026-05-28T18:00:37Z | kratos0718 | 204,972,096 | U_kgDODDegQA | User | false |
huggingface/transformers | 4,531,811,404 | I_kwDOCUB6oc8AAAABDh30TA | 46,238 | https://github.com/huggingface/transformers/issues/46238 | https://api.github.com/repos/huggingface/transformers/issues/46238 | DeepSeek-V4 FP8 load fails: 1→1 WeightConverter w2 weights not dequantized due to $-anchored source pattern | ### System Info
In src/transformers/quantizers/quantizer_finegrained_fp8.py, strip $ before doing string operations on source_patterns:
# Line 211
weight_sources = [p for p in conv.source_patterns if p.rstrip("$").endswith(".weight")]
# Line 213
anchored_weight = [p.rstrip("$") + "$" for p in weight_sources]
# Lin... | closed | completed | false | 5 | [
"bug"
] | [] | 2026-05-27T11:02:33Z | 2026-06-02T18:40:39Z | 2026-06-02T18:40:39Z | NONE | null | 20260603T000024Z | 2026-06-03T00:00:24Z | moonhush922 | 284,500,487 | U_kgDOEPUiBw | User | false |
huggingface/transformers | 4,531,394,515 | I_kwDOCUB6oc8AAAABDheX0w | 46,234 | https://github.com/huggingface/transformers/issues/46234 | https://api.github.com/repos/huggingface/transformers/issues/46234 | `is_falcon_mamba_ssm_available` missing from `transformers.utils.import_utils` | ### System Info
- Reproduces on `transformers==5.8.0` and on current `main` (verified by direct read of `src/transformers/utils/import_utils.py`).
- Trigger condition: optional `kernels` lib is **not** installed (it lives in `extras["kernels"]` per `setup.py`, not in `install_requires`).
- Platform / Python version in... | closed | completed | false | 2 | [
"bug"
] | [] | 2026-05-27T09:55:44Z | 2026-06-08T14:27:10Z | 2026-06-08T14:27:10Z | NONE | null | 20260608T180018Z | 2026-06-08T18:00:18Z | voloha-mil | 36,573,372 | MDQ6VXNlcjM2NTczMzcy | User | false |
huggingface/transformers | 4,533,809,423 | I_kwDOCUB6oc8AAAABDjxxDw | 46,245 | https://github.com/huggingface/transformers/issues/46245 | https://api.github.com/repos/huggingface/transformers/issues/46245 | [`Layer Types`] Split keys across MLP and Attentions | See #46220, we have at some point merged all layer types into one. But they should be different for MLPs and Attentions. We are now stuck in a BC cycle, so we likely need to be breaking at some point
cc @zucchini-nlp @Cyrilvallez | open | null | false | 4 | [
"WIP"
] | [] | 2026-05-27T16:01:34Z | 2026-06-29T13:25:07Z | null | MEMBER | null | 20260629T180016Z | 2026-06-29T18:00:16Z | vasqu | 73,884,904 | MDQ6VXNlcjczODg0OTA0 | User | false |
huggingface/transformers | 4,534,346,085 | I_kwDOCUB6oc8AAAABDkShZQ | 46,247 | https://github.com/huggingface/transformers/issues/46247 | https://api.github.com/repos/huggingface/transformers/issues/46247 | Faster preprocessing when using new Parakeet TDT please | ### Feature request
Hi! Thanks for adding native Parakeet support in #44171 — really nice not having to pull in all of NeMo just to run these models. I've been benchmarking and noticed a speed gap I think has a clean fix, but wanted to file an issue before attempting a PR.
## What I noticed
Same weights (`nvidia/par... | closed | completed | false | 7 | [
"Feature request"
] | [] | 2026-05-27T17:24:40Z | 2026-05-29T15:58:46Z | 2026-05-29T15:58:45Z | NONE | null | 20260529T180028Z | 2026-05-29T18:00:28Z | BBC-Esq | 108,230,321 | U_kgDOBnN2sQ | User | false |
huggingface/transformers | 4,542,960,581 | I_kwDOCUB6oc8AAAABDsgTxQ | 46,271 | https://github.com/huggingface/transformers/issues/46271 | https://api.github.com/repos/huggingface/transformers/issues/46271 | docs: clarify output_hidden_states behavior for Koopman/spectral analysis use cases | ## Summary
When using `output_hidden_states=True` in `AutoModel.forward()`, the documentation does not clearly explain:
1. **Which layers are included**: Whether layer 0 is the embedding layer or the first transformer block varies by architecture.
2. **Shape conventions**: `(batch, seq_len, hidden_dim)` per layer, bu... | closed | completed | false | 0 | [] | [] | 2026-05-28T18:53:35Z | 2026-05-29T11:59:25Z | 2026-05-29T11:59:24Z | NONE | null | 20260529T120035Z | 2026-05-29T12:00:35Z | hinanohart | 255,530,825 | U_kgDODzsXSQ | User | false |
huggingface/transformers | 4,544,756,778 | I_kwDOCUB6oc8AAAABDuN8Kg | 46,273 | https://github.com/huggingface/transformers/issues/46273 | https://api.github.com/repos/huggingface/transformers/issues/46273 | Direct build of layer/attention silently produces non-causal output | ### Feature request
- Option A: make the eager fallback honor `module.is_causal`:
```python
def eager_attention_forward(module, query, key, value, attention_mask, scaling, dropout=0.0, **kwargs):
...
attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
if attention_mask is not None:
... | closed | completed | false | 9 | [
"Feature request"
] | [] | 2026-05-29T00:36:55Z | 2026-05-30T16:25:39Z | 2026-05-30T16:24:28Z | NONE | null | 20260530T180032Z | 2026-05-30T18:00:32Z | ayghri | 32,200,675 | MDQ6VXNlcjMyMjAwNjc1 | User | false |
huggingface/transformers | 4,547,357,784 | I_kwDOCUB6oc8AAAABDwssWA | 46,285 | https://github.com/huggingface/transformers/issues/46285 | https://api.github.com/repos/huggingface/transformers/issues/46285 | CLIPTextModel and CLIPTextModelWithProjection have inconsistent structure since v5.6 (.text_model removed from one but not the other) | ### System Info
- `transformers` version: reproduced on **5.6.0** and **5.9.0** (unaffected on 5.5.4)
- Platform: Windows-11-10.0.26200-SP0 (also reproducible cross-platform; the bug is in pure model construction)
- Python version: 3.12.9
- PyTorch version: 2.7.1+cu128
- Accelerate version: 1.8.1
This is a pure-Pytho... | closed | completed | false | 0 | [] | [] | 2026-05-29T09:29:08Z | 2026-05-29T09:40:47Z | 2026-05-29T09:40:47Z | NONE | null | 20260529T120035Z | 2026-05-29T12:00:35Z | kappacommit | 20,428,294 | MDQ6VXNlcjIwNDI4Mjk0 | User | false |
huggingface/transformers | 4,552,121,902 | I_kwDOCUB6oc8AAAABD1PeLg | 46,291 | https://github.com/huggingface/transformers/issues/46291 | https://api.github.com/repos/huggingface/transformers/issues/46291 | Importing `transformers.pipelines` fails with kernels v0.15: kernels now requires `version` or `revision` | ### System Info
Transformers version: 5.10.0.dev0
Python version: 3.14.5
Platform: Windows-11-10.0.26200-SP0
Machine: x86_64
### Who can help?
@MekkCyber @drbh
### Information
- [ ] The official example scripts
- [x] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (s... | closed | completed | false | 3 | [
"bug"
] | [] | 2026-05-29T22:55:56Z | 2026-07-01T17:21:33Z | 2026-06-09T12:19:33Z | NONE | null | 20260701T180021Z | 2026-07-01T18:00:21Z | jshn9515 | 240,566,314 | U_kgDODlbAKg | User | false |
huggingface/transformers | 4,553,278,916 | I_kwDOCUB6oc8AAAABD2WFxA | 46,294 | https://github.com/huggingface/transformers/issues/46294 | https://api.github.com/repos/huggingface/transformers/issues/46294 | Qwen3.5-9B video input fails with apply_chat_template — Video features and video tokens do not match | ### System Info
- `transformers` version: 5.10.0.dev0
- Platform: Linux-6.6.122+-x86_64-with-glibc2.35
- Python version: 3.12.13
- Huggingface_hub version: 1.10.1
- Safetensors version: 0.7.0
- Accelerate version: 1.13.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?... | closed | completed | false | 1 | [
"bug"
] | [] | 2026-05-30T04:59:58Z | 2026-06-02T11:29:11Z | 2026-06-02T11:29:11Z | NONE | null | 20260602T120030Z | 2026-06-02T12:00:30Z | 14790897 | 121,866,954 | U_kgDOB0OKyg | User | false |
huggingface/transformers | 4,553,860,084 | I_kwDOCUB6oc8AAAABD25j9A | 46,297 | https://github.com/huggingface/transformers/issues/46297 | https://api.github.com/repos/huggingface/transformers/issues/46297 | Safety Report: AI Agent Guardrails Do Not Work — 56-Day Empirical Proof | ## AI Guardrails Do Not Work — 56-Day Empirical Proof
I am a developer who has used AI coding assistants for 56 days in a regulated environment. During that time:
- **32 workflow violations** occurred despite configuring every available guardrail mechanism
- The AI **destroyed my AWS management account** by deploying... | closed | completed | false | 3 | [
"Code agent slop"
] | [] | 2026-05-30T08:28:09Z | 2026-06-04T09:50:59Z | 2026-06-02T12:51:50Z | NONE | null | 20260604T120021Z | 2026-06-04T12:00:21Z | tzb1-ai | 274,091,206 | U_kgDOEFZMxg | User | false |
huggingface/transformers | 4,553,592,698 | I_kwDOCUB6oc8AAAABD2pPeg | 46,295 | https://github.com/huggingface/transformers/issues/46295 | https://api.github.com/repos/huggingface/transformers/issues/46295 | Training Support for FP8 | ### Feature request
Add Training Support for the QuantizationMethod.FP8
### Motivation
I always want to finetune in FP8 but then get ValueError: The model you are trying to fine-tune is quantized with QuantizationMethod.FP8 but that quantization method do not support training. Please open an issue on GitHub: https:/... | open | null | false | 1 | [
"Feature request"
] | [] | 2026-05-30T06:44:10Z | 2026-05-30T08:44:51Z | null | NONE | null | 20260530T120027Z | 2026-05-30T12:00:27Z | Banaxi-Tech | 264,138,555 | U_kgDOD75vOw | User | false |
huggingface/transformers | 4,554,074,982 | I_kwDOCUB6oc8AAAABD3GrZg | 46,298 | https://github.com/huggingface/transformers/issues/46298 | https://api.github.com/repos/huggingface/transformers/issues/46298 | Noisy logging message | ### System Info
Transformers 5.9
### Who can help?
@yonigozlan @molbap @ArthurZucker
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...)
- [ ] My own task or dataset (give details below)
... | closed | completed | false | 4 | [
"bug"
] | [] | 2026-05-30T10:10:33Z | 2026-06-04T13:27:05Z | 2026-06-04T13:27:05Z | CONTRIBUTOR | null | 20260604T180024Z | 2026-06-04T18:00:24Z | davidmezzetti | 561,939 | MDQ6VXNlcjU2MTkzOQ== | User | false |
huggingface/transformers | 4,556,386,027 | I_kwDOCUB6oc8AAAABD5Tu6w | 46,304 | https://github.com/huggingface/transformers/issues/46304 | https://api.github.com/repos/huggingface/transformers/issues/46304 | Transformers v5.9.0 breaks support for pytorch with USE_DISTRIBUTED=0 | ### System Info
Version: 5.9.0
Platform: Intel-based MacOS 26.5
Python Version: 3.14
Pytorch: 2.12.0 (self-built for Intel MacOS)
### Who can help?
@remi-or
### Information
- [ ] The official example scripts
- [ ] My own modified scripts
### Tasks
- [ ] An officially supported task in the `examples` folder (suc... | closed | completed | false | 5 | [
"bug"
] | [] | 2026-05-31T02:56:42Z | 2026-06-01T03:31:10Z | 2026-06-01T03:31:10Z | NONE | null | 20260601T060039Z | 2026-06-01T06:00:39Z | MSDarwish2000 | 20,906,401 | MDQ6VXNlcjIwOTA2NDAx | User | false |
huggingface/transformers | 4,557,084,980 | I_kwDOCUB6oc8AAAABD5-ZNA | 46,305 | https://github.com/huggingface/transformers/issues/46305 | https://api.github.com/repos/huggingface/transformers/issues/46305 | MoonshineStreaming: sliding_window_mask_function off-by-one — right_window_size=4 attends to 3 future frames instead of 4 | ### System Info
- transformers version: 5.9.0
- Platform: Linux-6.8.0-100-generic-x86_64-with-glibc2.39
- Python version: 3.12.3
- PyTorch version: 2.12.0
### Who can help?
@eustlb
### Information
`sliding_window_mask_function` in `modeling_moonshine_streaming.py` (line 349) uses strict less-than (`< right_windo... | open | null | false | 3 | [
"bug"
] | [] | 2026-05-31T09:02:24Z | 2026-07-01T09:03:17Z | null | NONE | null | 20260701T120019Z | 2026-07-01T12:00:19Z | Kush0610 | 213,816,176 | U_kgDODL6TcA | User | false |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.