diff --git "a/data/prs.json" "b/data/prs.json"
--- "a/data/prs.json"
+++ "b/data/prs.json"
@@ -1,24759 +1,24573 @@
 [
   {
-    "additions": 138,
-    "author": "SunMarc",
+    "additions": 1,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR updates the support for response api. I was mainly basing myself on chat completion api but there are minor differences with it e.g `input_image` vs `image_url` for `type` or `input_text` vs `text`, differen\u2026",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? Minor update",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45463",
-    "created_at": "2026-04-15T15:56:45Z",
-    "deletions": 174,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45484",
+    "created_at": "2026-04-16T20:10:45Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45463/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45463",
+    "files_url": "https://github.com/huggingface/transformers/pull/45484/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45484",
     "labels": [],
-    "merged": false,
-    "number": 45463,
-    "review_comments_count": 2,
-    "state": "open",
-    "title": "Fix response api support ",
-    "updated_at": "2026-04-15T20:42:24Z"
+    "merged": true,
+    "number": 45484,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Minor update",
+    "updated_at": "2026-04-16T20:27:48Z"
   },
   {
-    "additions": 118,
-    "author": "tarekziade",
+    "additions": 36,
+    "author": "vasqu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Activated some bandit rules and fixed a few spots",
-    "changed_files": 33,
+    "body_excerpt": "Fixup some conversion mappings --> everything should live directly under `mapping` except we add onto it - Solar change is the same as qwen2 moe - Cohere moved - Ernie moe similar to minimax with one additional rename",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45462",
-    "created_at": "2026-04-15T15:37:15Z",
-    "deletions": 67,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45483",
+    "created_at": "2026-04-16T15:59:26Z",
+    "deletions": 64,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45462/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45462",
+    "files_url": "https://github.com/huggingface/transformers/pull/45483/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45483",
     "labels": [],
     "merged": false,
-    "number": 45462,
+    "number": 45483,
     "review_comments_count": 0,
     "state": "open",
-    "title": "chore(sec): added a handful of security checks",
-    "updated_at": "2026-04-15T16:04:51Z"
+    "title": "[`Conversion Mapping`] Small fixups",
+    "updated_at": "2026-04-16T16:19:28Z"
   },
   {
-    "additions": 0,
-    "author": "JiauZhang",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Remove redundant condition checks in `get_image_size` method",
-    "changed_files": 1,
+    "additions": 184,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": ".",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45461",
-    "created_at": "2026-04-15T14:00:11Z",
-    "deletions": 8,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45481",
+    "created_at": "2026-04-16T15:51:21Z",
+    "deletions": 185,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45461/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45461",
+    "files_url": "https://github.com/huggingface/transformers/pull/45481/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45481",
     "labels": [],
     "merged": false,
-    "number": 45461,
-    "review_comments_count": 2,
+    "number": 45481,
+    "review_comments_count": 1,
     "state": "open",
-    "title": "Remove redundant condition checks in `get_image_size` method",
-    "updated_at": "2026-04-15T15:03:42Z"
+    "title": "Add check-auto in repo-consistency and fix sorting",
+    "updated_at": "2026-04-16T16:15:48Z"
   },
   {
-    "additions": 5,
-    "author": "cloudyun888",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary When protobuf is not installed, is a function call used as an expression. Because it is evaluated lazily when the try block raises, the resulting from the function itself bypasses the RuntimeError and OSError handlers below it,\u2026",
-    "changed_files": 1,
+    "additions": 27,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR fixes quantization tests. A few things were deprecated when compressed-tensors had their latest release, so i'm updating the tests. For fouroversix, it's just that the model was a bit too big for the CI",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45460",
-    "created_at": "2026-04-15T13:18:36Z",
-    "deletions": 5,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45480",
+    "created_at": "2026-04-16T15:23:30Z",
+    "deletions": 86,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45460/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45460",
+    "files_url": "https://github.com/huggingface/transformers/pull/45480/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45480",
     "labels": [],
     "merged": false,
-    "number": 45460,
+    "number": 45480,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix(tokenization): re-raise ImportError to allow RuntimeError/OSError fallback (#45459)",
-    "updated_at": "2026-04-15T13:18:36Z"
+    "title": "Update quants tests ",
+    "updated_at": "2026-04-16T15:34:33Z"
   },
   {
-    "additions": 8,
-    "author": "tomaarsen",
+    "additions": 321,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Currently, for qwen2_5_omni and qwen3_omni_moe, you can only load the 'Talker' variant, i.e. with the audio output. This is a bit like only being able to load a checkpoint with `AutoModelForCausalLM` while `AutoMode\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? As per title, I think this pattern is used quite often and deserves to be a public mask-fn. Used currently in gemma/paligemma family, GIT, PI0 and will be used in two upcoming models (deepseekOcr and Molmo2) Imo it\u2026",
+    "changed_files": 9,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45457",
-    "created_at": "2026-04-15T12:29:47Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45477",
+    "created_at": "2026-04-16T14:12:03Z",
+    "deletions": 593,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45457/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45457",
+    "files_url": "https://github.com/huggingface/transformers/pull/45477/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45477",
     "labels": [],
     "merged": false,
-    "number": 45457,
-    "review_comments_count": 4,
+    "number": 45477,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Allow loading Qwen Thinker 'base' models without generative head",
-    "updated_at": "2026-04-15T16:14:02Z"
+    "title": "[WIP] Add blockwise mask fn as opt arg for all masking functions",
+    "updated_at": "2026-04-16T14:42:38Z"
   },
   {
-    "additions": 2,
-    "author": "tarekziade",
+    "additions": 14,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? we're extending `ty` to more modules and we need stubs from more libs like openai.",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Call CI workflow",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45456",
-    "created_at": "2026-04-15T12:10:18Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45476",
+    "created_at": "2026-04-16T13:49:50Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45456/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45456",
+    "files_url": "https://github.com/huggingface/transformers/pull/45476/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45476",
     "labels": [],
-    "merged": true,
-    "number": 45456,
+    "merged": false,
+    "number": 45476,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "refactor(qa): extend extras so ty can run on server modules",
-    "updated_at": "2026-04-15T16:08:23Z"
+    "state": "open",
+    "title": "[Don't merge] Call CI workflow",
+    "updated_at": "2026-04-16T21:04:16Z"
   },
   {
-    "additions": 3,
-    "author": "tomaarsen",
+    "additions": 28,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? There's 2 changes, one is a definite fix and one is a preference. Some background: there are a lot of models that have finetuned `qwen2_5_omni`, e.g. https://huggingface.co/LCO-Embedding/LCO-Embedding-Omni-3B, and i\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Split out the `mlinter` tool. We want to be able to: - use it from other CI projects - remove the ability to alter the linter from Transformers PRs This change should keep the exact same behavior than before",
+    "changed_files": 27,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45455",
-    "created_at": "2026-04-15T11:38:05Z",
-    "deletions": 25,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45475",
+    "created_at": "2026-04-16T12:01:49Z",
+    "deletions": 2661,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45455/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45455",
+    "files_url": "https://github.com/huggingface/transformers/pull/45475/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45475",
     "labels": [],
     "merged": false,
-    "number": 45455,
-    "review_comments_count": 5,
+    "number": 45475,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "[`fix`] Make Qwen2_5OmniProcessor warning a lot less noisy via warning_once",
-    "updated_at": "2026-04-15T12:23:04Z"
+    "title": "chore(qa): split out mlinter",
+    "updated_at": "2026-04-16T17:09:22Z"
   },
   {
-    "additions": 81,
-    "author": "zucchini-nlp",
+    "additions": 2,
+    "author": "rtrompier",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45200 As per title, this error was actually needed only in PG. Other models don't have such prefix/suffix separation when training",
-    "changed_files": 7,
+    "body_excerpt": "Bump the pinned doc-builder SHA so that main documentation builds also sync to the HF bucket (dual-write).",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 12,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45454",
-    "created_at": "2026-04-15T11:11:34Z",
-    "deletions": 104,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45454/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45454",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45474",
+    "created_at": "2026-04-16T11:59:17Z",
+    "deletions": 2,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45474/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45474",
     "labels": [],
     "merged": false,
-    "number": 45454,
+    "number": 45474,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Gemma4 training with text-only samples",
-    "updated_at": "2026-04-15T15:51:09Z"
+    "title": "chore: bump doc-builder SHA for main doc build workflow",
+    "updated_at": "2026-04-16T12:09:45Z"
   },
   {
-    "additions": 466,
-    "author": "ArthurZucker",
+    "additions": 201,
+    "author": "AmineDiro",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Ai init",
+    "body_excerpt": "While benchmarking Qwen3-30B-A3B SFT training with Expert Parallelism (EP) using TRL, I found three bugs that combine to produce silently wrong results or NaN loss. Every existing test uses `tp_plan=\"auto\"` which bypasses `RouterParallel`\u2026",
     "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45453",
-    "created_at": "2026-04-15T10:22:17Z",
-    "deletions": 50,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45453/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45453",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45473",
+    "created_at": "2026-04-16T10:59:14Z",
+    "deletions": 41,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45473/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45473",
     "labels": [],
     "merged": false,
-    "number": 45453,
+    "number": 45473,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Draft commit",
-    "updated_at": "2026-04-15T11:40:23Z"
+    "title": "Fix EP: RouterParallel shape, tp_plan property, grouped_mm sentinels",
+    "updated_at": "2026-04-16T13:56:10Z"
   },
   {
-    "additions": 3058,
-    "author": "DavidSolanas",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #45306 ## What and why All `models/X/__init__.py` files used `from .module import *` inside the `TYPE_CHECKING` block. This makes it impossible for static analysis tools (pyright, mypy, IDEs) to know which symbols are actually export\u2026",
-    "changed_files": 446,
+    "additions": 2,
+    "author": "kevinmalana",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes a crash in `get_device_properties()` in `testing_utils.py` when CUDA is installed on the system but no GPU device is present (e.g., a CPU-only cloud studio with CUDA libraries installed). The function called\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45452",
-    "created_at": "2026-04-15T08:44:00Z",
-    "deletions": 1305,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45472",
+    "created_at": "2026-04-16T10:03:07Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45452/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45452",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45472/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45472",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 45452,
+    "number": 45472,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "refactor: replace wildcard imports with explicit imports in model __init__.py files",
-    "updated_at": "2026-04-15T12:15:24Z"
+    "state": "closed",
+    "title": "fix(testing_utils): guard get_device_capability with torch.cuda.is_available()",
+    "updated_at": "2026-04-16T10:57:34Z"
   },
   {
-    "additions": 20,
-    "author": "MukundaKatta",
-    "author_association": "NONE",
-    "body_excerpt": "Docstring/comment-only typo fixes across Qwen3-VL, Qwen3.5, GLM4V, GLM4.6V, GLM-OCR and their MoE variants. `seperate` -> `separate`. No behavior changes. I deliberately left `image_seperate.weight` in `convert_mm_grounding_dino_to_hf.py`\u2026",
-    "changed_files": 10,
+    "additions": 3413,
+    "author": "nuxlear",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Add EXAONE 4.5 architecture for the [EXAONE 4.5 model](https://huggingface.co/collections/LGAI-EXAONE/exaone-45) released by LG AI Research. This PR adds the modeling code for EXAONE 4.5, which uses the same LLM arc\u2026",
+    "changed_files": 16,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45451",
-    "created_at": "2026-04-15T08:33:29Z",
-    "deletions": 20,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45471",
+    "created_at": "2026-04-16T08:52:35Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45451/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45451",
+    "files_url": "https://github.com/huggingface/transformers/pull/45471/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45471",
     "labels": [],
     "merged": false,
-    "number": 45451,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix 'seperate' typo in qwen3/glm video-model docstrings",
-    "updated_at": "2026-04-15T10:57:26Z"
+    "number": 45471,
+    "review_comments_count": 15,
+    "state": "open",
+    "title": "Add EXAONE 4.5 implementations",
+    "updated_at": "2026-04-16T14:21:42Z"
   },
   {
-    "additions": 1,
-    "author": "rtrompier",
-    "author_association": "MEMBER",
-    "body_excerpt": "Switch the PR doc upload flow from the legacy dataset push to the new HF bucket.",
+    "additions": 7,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@ydshieh pls help review, thx!",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45450",
-    "created_at": "2026-04-15T08:26:45Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45470",
+    "created_at": "2026-04-16T07:44:20Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45450/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45450",
+    "files_url": "https://github.com/huggingface/transformers/pull/45470/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45470",
     "labels": [],
     "merged": false,
-    "number": 45450,
+    "number": 45470,
     "review_comments_count": 0,
     "state": "open",
-    "title": "chore: bump doc-builder SHA for PR upload workflow",
-    "updated_at": "2026-04-15T10:13:31Z"
+    "title": "skip test_flash_attn_2_can_dispatch_composite_models tests for",
+    "updated_at": "2026-04-16T07:45:33Z"
   },
   {
-    "additions": 1,
-    "author": "hmellor",
-    "author_association": "MEMBER",
-    "body_excerpt": "This model also has the wrong tokenizer class in its config",
-    "changed_files": 1,
+    "additions": 12,
+    "author": "Spectual",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? `PixioPatchEmbeddings.forward` already accepted an `interpolate_pos_encoding` flag (inherited from `ViTPatchEmbeddings`) to skip image-size validation and allow variable-resolution inputs. However, neither `PixioEmb\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45449",
-    "created_at": "2026-04-15T08:26:05Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45469",
+    "created_at": "2026-04-16T06:54:56Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45449/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45449",
+    "files_url": "https://github.com/huggingface/transformers/pull/45469/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45469",
     "labels": [],
-    "merged": true,
-    "number": 45449,
+    "merged": false,
+    "number": 45469,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add `step3_vl` to `MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS`",
-    "updated_at": "2026-04-15T09:28:46Z"
+    "state": "open",
+    "title": "Fix: propagate interpolate_pos_encoding through Pixio model hierarchy",
+    "updated_at": "2026-04-16T06:56:04Z"
   },
   {
-    "additions": 178,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do?",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45448",
-    "created_at": "2026-04-15T08:06:53Z",
-    "deletions": 54,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45448/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45448",
-    "labels": [],
-    "merged": false,
-    "number": 45448,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix clips",
-    "updated_at": "2026-04-15T14:29:41Z"
-  },
-  {
-    "additions": 1,
-    "author": "ZSLsherly",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "This commit corrects the PyTorch version check for importing `AuxRequest` from `torch.nn.attention.flex_attention`(line51). The `AuxRequest` class was actually introduced in PyTorch 2.9.1, not 2.9.0. The current code attempts to import it\u2026",
+    "additions": 17,
+    "author": "Jah-yee",
+    "author_association": "NONE",
+    "body_excerpt": "Good day, ## Problem On Apple Silicon (MPS backend), `torch.nn.functional.scaled_dot_product_attention` produces incorrect output when the value tensor's head dimension differs from the query tensor's head dimension. This affects DeepSeek\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45445",
-    "created_at": "2026-04-15T03:09:38Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45467",
+    "created_at": "2026-04-16T06:44:51Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45445/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45445",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45467/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45467",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 45445,
-    "review_comments_count": 1,
+    "number": 45467,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Update Torch version check for flex attention",
-    "updated_at": "2026-04-15T11:40:34Z"
+    "title": "Fix MPS SDPA output shape when value head dim differs from query head dim",
+    "updated_at": "2026-04-16T10:53:42Z"
   },
   {
-    "additions": 50,
-    "author": "tomaarsen",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Resolves https://github.com/huggingface/sentence-transformers/issues/3724 ## Code Agent Policy - [x] I confirm that this is not a pure code agent PR. ## Before submitting - [ ] This PR fixes a typo or improves the d\u2026",
-    "changed_files": 2,
+    "additions": 3,
+    "author": "Jah-yee",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #45459 - Previously import_protobuf_decode_error() raised ImportError when protobuf wasn't installed even for other exceptions, hiding the real error. Now returns empty tuple () so the actual exception propagates.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45444",
-    "created_at": "2026-04-14T19:28:34Z",
-    "deletions": 20,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45466",
+    "created_at": "2026-04-16T00:55:40Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45444/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45444",
+    "files_url": "https://github.com/huggingface/transformers/pull/45466/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45466",
     "labels": [],
     "merged": false,
-    "number": 45444,
-    "review_comments_count": 4,
-    "state": "open",
-    "title": "[`fix`] Always early return for non-Mistral models in _patch_mistral_regex",
-    "updated_at": "2026-04-15T12:03:55Z"
+    "number": 45466,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix: return empty tuple when protobuf not available",
+    "updated_at": "2026-04-16T03:16:39Z"
   },
   {
-    "additions": 38,
-    "author": "qgallouedec",
+    "additions": 65,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "When `transformers serve` is launched with a positional model argument, the server silently overwrites the `\"model\"` field in every incoming request with the pinned model id. This is surprising: a client that asks for model B receives a re\u2026",
-    "changed_files": 2,
+    "body_excerpt": "refactors the \"Contribute to Transformers\" doc to be more lightweight and an easy entry point that links out to more dedicated guides",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45443",
-    "created_at": "2026-04-14T19:14:10Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45443/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45443",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45465",
+    "created_at": "2026-04-15T23:12:11Z",
+    "deletions": 469,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45465/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45465",
     "labels": [],
     "merged": false,
-    "number": 45443,
+    "number": 45465,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Raise 400 on model mismatch when `transformers serve` is pinned",
-    "updated_at": "2026-04-15T11:42:44Z"
+    "title": "[docs] contributing",
+    "updated_at": "2026-04-15T23:22:12Z"
   },
   {
-    "additions": 4,
-    "author": "paulinebm",
+    "additions": 186,
+    "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? This PR updates the support for response api. I was mainly basing myself on chat completion api but there are minor differences with it e.g `input_image` vs `image_url` for `type` or `input_text` vs `text`, differen\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45442",
-    "created_at": "2026-04-14T18:49:49Z",
-    "deletions": 4,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45463",
+    "created_at": "2026-04-15T15:56:45Z",
+    "deletions": 181,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45442/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45442",
+    "files_url": "https://github.com/huggingface/transformers/pull/45463/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45463",
     "labels": [],
     "merged": true,
-    "number": 45442,
-    "review_comments_count": 0,
+    "number": 45463,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "Update workflow references to new commit hash",
-    "updated_at": "2026-04-14T18:59:01Z"
+    "title": "Fix response api support ",
+    "updated_at": "2026-04-16T20:54:08Z"
   },
   {
-    "additions": 9,
-    "author": "casinca",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Please see fix #45440 for more details Discussed with @vasqu Also fixed via regen/inheritance: - exaone-moe - glm4-moe - glm4v-moe - glm4-moe-lite - glm_moe_dsa - nemotron_h - solar_open &nbsp; > [!WARNING] > > **Ed\u2026",
-    "changed_files": 9,
+    "additions": 118,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Activated some bandit rules and fixed a few spots",
+    "changed_files": 33,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45441",
-    "created_at": "2026-04-14T17:51:08Z",
-    "deletions": 9,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45462",
+    "created_at": "2026-04-15T15:37:15Z",
+    "deletions": 67,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45441/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45441",
+    "files_url": "https://github.com/huggingface/transformers/pull/45462/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45462",
     "labels": [],
     "merged": false,
-    "number": 45441,
+    "number": 45462,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix(DSV3): parity between native `DeepseekV3MoE` and remote official implementation",
-    "updated_at": "2026-04-15T09:27:41Z"
+    "title": "chore(sec): added a handful of security checks",
+    "updated_at": "2026-04-15T16:04:51Z"
   },
   {
-    "additions": 1,
-    "author": "McPatate",
-    "author_association": "MEMBER",
-    "body_excerpt": "Bumping safetensors to its latest (pre)released version! cc @tarekziade",
+    "additions": 0,
+    "author": "JiauZhang",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Remove redundant condition checks in `get_image_size` method",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45439",
-    "created_at": "2026-04-14T17:20:51Z",
-    "deletions": 1,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45461",
+    "created_at": "2026-04-15T14:00:11Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45439/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45439",
+    "files_url": "https://github.com/huggingface/transformers/pull/45461/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45461",
     "labels": [],
     "merged": false,
-    "number": 45439,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "feat: bump min safetensors version to `0.8.0-rc.0`",
-    "updated_at": "2026-04-15T11:54:37Z"
+    "number": 45461,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "Remove redundant condition checks in `get_image_size` method",
+    "updated_at": "2026-04-16T11:38:31Z"
   },
   {
-    "additions": 275,
-    "author": "Charly21r",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #45373 Adds `Gemma4TextForSequenceClassification` and `Gemma4ForSequenceClassification` to `transformers.models.gemma4`, following the same pattern established by Gemma 3. Prior to this change, `AutoModelForSe\u2026",
-    "changed_files": 5,
+    "additions": 5,
+    "author": "cloudyun888",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary When protobuf is not installed, is a function call used as an expression. Because it is evaluated lazily when the try block raises, the resulting from the function itself bypasses the RuntimeError and OSError handlers below it,\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45438",
-    "created_at": "2026-04-14T17:20:36Z",
-    "deletions": 11,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45460",
+    "created_at": "2026-04-15T13:18:36Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45438/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45438",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45460/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45460",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 45438,
-    "review_comments_count": 12,
-    "state": "open",
-    "title": "Add Gemma4ForSequenceClassification",
-    "updated_at": "2026-04-15T16:13:05Z"
+    "number": 45460,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix(tokenization): re-raise ImportError to allow RuntimeError/OSError fallback (#45459)",
+    "updated_at": "2026-04-16T10:48:07Z"
   },
   {
-    "additions": 4,
+    "additions": 2,
     "author": "tomaarsen",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Currently, for qwen2_5_omni and qwen3_omni_moe, you can only load the 'Talker' variant, i.e. with the audio output. This is a bit like only being able to load a checkpoint with `AutoModelForCausalLM` while `AutoMode\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45437",
-    "created_at": "2026-04-14T17:14:24Z",
-    "deletions": 8,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45457",
+    "created_at": "2026-04-15T12:29:47Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45437/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45437",
+    "files_url": "https://github.com/huggingface/transformers/pull/45457/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45457",
     "labels": [],
-    "merged": false,
-    "number": 45437,
-    "review_comments_count": 2,
-    "state": "open",
-    "title": "Fix spurious position_ids warnings for at least 40 architectures",
-    "updated_at": "2026-04-15T11:40:14Z"
+    "merged": true,
+    "number": 45457,
+    "review_comments_count": 4,
+    "state": "closed",
+    "title": "Allow loading Qwen Thinker 'base' models without generative head",
+    "updated_at": "2026-04-16T12:24:48Z"
   },
   {
-    "additions": 19,
-    "author": "AmineDiro",
+    "additions": 2,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "## Summary - Added `base_model_ep_plan` to `Qwen3MoeConfig` enabling expert parallelism via `DistributedConfig(enable_expert_parallel=True)` - Fix two small bugs in `GroupedGemmParallel` that prevented EP from working with 2D device meshes\u2026",
+    "body_excerpt": "# What does this PR do? we're extending `ty` to more modules and we need stubs from more libs like openai.",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45436",
-    "created_at": "2026-04-14T15:35:56Z",
-    "deletions": 5,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45456",
+    "created_at": "2026-04-15T12:10:18Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45436/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45436",
+    "files_url": "https://github.com/huggingface/transformers/pull/45456/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45456",
     "labels": [],
-    "merged": false,
-    "number": 45436,
+    "merged": true,
+    "number": 45456,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add expert parallelism (EP) support for Qwen3 MoE + fix GroupedGemmParallel for 2D meshes",
-    "updated_at": "2026-04-15T14:57:41Z"
+    "state": "closed",
+    "title": "refactor(qa): extend extras so ty can run on server modules",
+    "updated_at": "2026-04-15T16:08:23Z"
   },
   {
-    "additions": 30,
-    "author": "itazap",
+    "additions": 3,
+    "author": "tomaarsen",
     "author_association": "MEMBER",
-    "body_excerpt": "fixed https://github.com/huggingface/transformers/issues/44869 add check to not index past decoded chars with special tokens",
+    "body_excerpt": "# What does this PR do? There's 2 changes, one is a definite fix and one is a preference. Some background: there are a lot of models that have finetuned `qwen2_5_omni`, e.g. https://huggingface.co/LCO-Embedding/LCO-Embedding-Omni-3B, and i\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45435",
-    "created_at": "2026-04-14T15:28:20Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45455",
+    "created_at": "2026-04-15T11:38:05Z",
+    "deletions": 25,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45435/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45435",
+    "files_url": "https://github.com/huggingface/transformers/pull/45455/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45455",
     "labels": [],
-    "merged": false,
-    "number": 45435,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "do not index past decoded chars with special tokens",
-    "updated_at": "2026-04-15T13:30:22Z"
+    "merged": true,
+    "number": 45455,
+    "review_comments_count": 5,
+    "state": "closed",
+    "title": "[`fix`] Make Qwen2_5OmniProcessor warning a lot less noisy via warning_once",
+    "updated_at": "2026-04-16T12:10:43Z"
   },
   {
-    "additions": 95,
-    "author": "SunMarc",
+    "additions": 81,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR updates the grad acc tests so that we are able to better catch issues. We check the grad norm also in addition to the loss. We are remove some redundant tests. As promised @vasqu. Just read the new test dire\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45200 As per title, this error was actually needed only in PG. Other models don't have such prefix/suffix separation when training",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45434",
-    "created_at": "2026-04-14T15:17:03Z",
-    "deletions": 282,
+    "comments_count": 13,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45454",
+    "created_at": "2026-04-15T11:11:34Z",
+    "deletions": 104,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45434/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45434",
+    "files_url": "https://github.com/huggingface/transformers/pull/45454/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45454",
     "labels": [],
-    "merged": true,
-    "number": 45434,
-    "review_comments_count": 4,
-    "state": "closed",
-    "title": "better grad acc tests",
-    "updated_at": "2026-04-15T13:10:22Z"
+    "merged": false,
+    "number": 45454,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Gemma4 training with text-only samples",
+    "updated_at": "2026-04-16T04:05:06Z"
   },
   {
-    "additions": 115,
-    "author": "IlyasMoutawwakil",
+    "additions": 917,
+    "author": "ArthurZucker",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Ai init",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45433",
-    "created_at": "2026-04-14T14:33:39Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45453",
+    "created_at": "2026-04-15T10:22:17Z",
+    "deletions": 67,
     "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45433/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45433",
+    "files_url": "https://github.com/huggingface/transformers/pull/45453/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45453",
     "labels": [],
     "merged": false,
-    "number": 45433,
+    "number": 45453,
     "review_comments_count": 0,
     "state": "open",
-    "title": "SonicMoe",
-    "updated_at": "2026-04-14T14:47:50Z"
+    "title": "Draft commit",
+    "updated_at": "2026-04-16T20:19:24Z"
   },
   {
-    "additions": 298,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This patch adds the pipeline in our type checker, and also adds mccabe complexity in our Ruff linter with a 75 threshold for now. To reduce complexity, logical blocks are now splitted into private functions to impro\u2026",
-    "changed_files": 5,
+    "additions": 3058,
+    "author": "DavidSolanas",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #45306 ## What and why All `models/X/__init__.py` files used `from .module import *` inside the `TYPE_CHECKING` block. This makes it impossible for static analysis tools (pyright, mypy, IDEs) to know which symbols are actually export\u2026",
+    "changed_files": 446,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45432",
-    "created_at": "2026-04-14T14:04:07Z",
-    "deletions": 193,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45452",
+    "created_at": "2026-04-15T08:44:00Z",
+    "deletions": 1305,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45432/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45432",
+    "files_url": "https://github.com/huggingface/transformers/pull/45452/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45452",
     "labels": [],
     "merged": false,
-    "number": 45432,
+    "number": 45452,
     "review_comments_count": 0,
     "state": "open",
-    "title": "chore(qa): split pipeline and add type checking",
-    "updated_at": "2026-04-15T12:38:45Z"
+    "title": "refactor: replace wildcard imports with explicit imports in model __init__.py files",
+    "updated_at": "2026-04-15T12:15:24Z"
   },
   {
-    "additions": 3,
-    "author": "ambroiseodt",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fix checkpoint path in Dinov2 model_docs. # What does this PR do? Fixes the wrong checkpoint path in Dinov2 model_docs #45431 Changes made: the previous checkpoint (google/dinov2-base-patch16-224) does not exist and has been changed to the\u2026",
-    "changed_files": 1,
+    "additions": 20,
+    "author": "MukundaKatta",
+    "author_association": "NONE",
+    "body_excerpt": "Docstring/comment-only typo fixes across Qwen3-VL, Qwen3.5, GLM4V, GLM4.6V, GLM-OCR and their MoE variants. `seperate` -> `separate`. No behavior changes. I deliberately left `image_seperate.weight` in `convert_mm_grounding_dino_to_hf.py`\u2026",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45430",
-    "created_at": "2026-04-14T13:45:51Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45451",
+    "created_at": "2026-04-15T08:33:29Z",
+    "deletions": 20,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45430/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45430",
+    "files_url": "https://github.com/huggingface/transformers/pull/45451/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45451",
     "labels": [],
-    "merged": true,
-    "number": 45430,
+    "merged": false,
+    "number": 45451,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Doc] Correct checkpoint path in Dinov2 model_docs ",
-    "updated_at": "2026-04-15T09:00:57Z"
+    "title": "Fix 'seperate' typo in qwen3/glm video-model docstrings",
+    "updated_at": "2026-04-15T10:57:26Z"
   },
   {
-    "additions": 5,
-    "author": "ydshieh",
+    "additions": 1,
+    "author": "rtrompier",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Improve workflow file",
+    "body_excerpt": "Switch the PR doc upload flow from the legacy dataset push to the new HF bucket.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45429",
-    "created_at": "2026-04-14T09:44:36Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45450",
+    "created_at": "2026-04-15T08:26:45Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45429/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45429",
+    "files_url": "https://github.com/huggingface/transformers/pull/45450/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45450",
     "labels": [],
     "merged": false,
-    "number": 45429,
+    "number": 45450,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Improve workflow file",
-    "updated_at": "2026-04-14T09:49:38Z"
+    "state": "open",
+    "title": "chore: bump doc-builder SHA for PR upload workflow",
+    "updated_at": "2026-04-15T10:13:31Z"
   },
   {
-    "additions": 75,
-    "author": "tomaarsen",
+    "additions": 1,
+    "author": "hmellor",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 2,
+    "body_excerpt": "This model also has the wrong tokenizer class in its config",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45428",
-    "created_at": "2026-04-14T09:39:32Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45449",
+    "created_at": "2026-04-15T08:26:05Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45428/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45428",
+    "files_url": "https://github.com/huggingface/transformers/pull/45449/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45449",
     "labels": [],
     "merged": true,
-    "number": 45428,
+    "number": 45449,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[`fix`] PEFT integration fixes preventing save/load & integration",
-    "updated_at": "2026-04-14T11:27:26Z"
+    "title": "Add `step3_vl` to `MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS`",
+    "updated_at": "2026-04-15T09:28:46Z"
   },
   {
-    "additions": 1,
-    "author": "Aftabbs",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes #45341. `get_device_properties()` in `testing_utils.py` calls `torch.cuda.get_device_capability()` whenever `IS_CUDA_SYSTEM or IS_ROCM_SYSTEM` is `True`. This raises a `RuntimeError` on environments where CUD\u2026",
-    "changed_files": 1,
+    "additions": 438,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title. ## The issue The problem is that transforms that want to remove a full part of a model name (such as a prefix, e.g. the `model.` start) are non bijective in general, i.e. we completely lose the inf\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45427",
-    "created_at": "2026-04-14T08:56:45Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45448",
+    "created_at": "2026-04-15T08:06:53Z",
+    "deletions": 69,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45427/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45427",
+    "files_url": "https://github.com/huggingface/transformers/pull/45448/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45448",
     "labels": [],
     "merged": false,
-    "number": 45427,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(testing_utils): guard get_device_capability() with torch.cuda.is_available()",
-    "updated_at": "2026-04-15T10:38:19Z"
+    "number": 45448,
+    "review_comments_count": 24,
+    "state": "open",
+    "title": "[loading] Clean way to add/remove full parts in checkpoint names",
+    "updated_at": "2026-04-16T19:57:25Z"
   },
   {
-    "additions": 1333,
-    "author": "kmswin1",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Add A.X K1 model architecture What does this PR do? This PR adds support for A.X K1, a large-scale Mixture-of-Experts (MoE) language model developed by [SK Telecom](https://huggingface.co/skt). A.X K1 contains 519B total parameters with 33\u2026",
-    "changed_files": 8,
+    "additions": 1,
+    "author": "ZSLsherly",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "This commit corrects the PyTorch version check for importing `AuxRequest` from `torch.nn.attention.flex_attention`(line51). The `AuxRequest` class was actually introduced in PyTorch 2.9.1, not 2.9.0. The current code attempts to import it\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45426",
-    "created_at": "2026-04-14T07:23:05Z",
-    "deletions": 0,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45445",
+    "created_at": "2026-04-15T03:09:38Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45426/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45426",
+    "files_url": "https://github.com/huggingface/transformers/pull/45445/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45445",
     "labels": [],
     "merged": false,
-    "number": 45426,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Feature/add axk1",
-    "updated_at": "2026-04-14T07:38:58Z"
+    "number": 45445,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "Update Torch version check for flex attention",
+    "updated_at": "2026-04-15T11:40:34Z"
   },
   {
-    "additions": 129,
-    "author": "tarekziade",
+    "additions": 50,
+    "author": "tomaarsen",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Adds type checking to `modeling_utils.py`",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Resolves https://github.com/huggingface/sentence-transformers/issues/3724 ## Code Agent Policy - [x] I confirm that this is not a pure code agent PR. ## Before submitting - [ ] This PR fixes a typo or improves the d\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45425",
-    "created_at": "2026-04-14T06:56:00Z",
-    "deletions": 69,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45444",
+    "created_at": "2026-04-14T19:28:34Z",
+    "deletions": 20,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45425/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45425",
+    "files_url": "https://github.com/huggingface/transformers/pull/45444/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45444",
     "labels": [],
-    "merged": false,
-    "number": 45425,
-    "review_comments_count": 12,
-    "state": "open",
-    "title": "chore(typing): added modeling_utils to ty",
-    "updated_at": "2026-04-15T15:15:03Z"
+    "merged": true,
+    "number": 45444,
+    "review_comments_count": 4,
+    "state": "closed",
+    "title": "[`fix`] Always early return for non-Mistral models in _patch_mistral_regex",
+    "updated_at": "2026-04-16T12:19:46Z"
   },
   {
-    "additions": 189,
-    "author": "louzongzhi",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR implements **IndexCache** support for GLM5's DeepSeek Sparse Attention (DSA), enabling cross-layer index reuse to accelerate long-context inference. IndexCache accelerates sparse attention by reusing top-k t\u2026",
-    "changed_files": 3,
+    "additions": 38,
+    "author": "qgallouedec",
+    "author_association": "MEMBER",
+    "body_excerpt": "When `transformers serve` is launched with a positional model argument, the server silently overwrites the `\"model\"` field in every incoming request with the pinned model id. This is surprising: a client that asks for model B receives a re\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45424",
-    "created_at": "2026-04-14T05:14:08Z",
-    "deletions": 46,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45443",
+    "created_at": "2026-04-14T19:14:10Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45424/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45424",
+    "files_url": "https://github.com/huggingface/transformers/pull/45443/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45443",
     "labels": [],
     "merged": false,
-    "number": 45424,
-    "review_comments_count": 10,
+    "number": 45443,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Add IndexCache support for GLM5 DSA",
-    "updated_at": "2026-04-15T13:30:13Z"
+    "title": "Raise 400 on model mismatch when `transformers serve` is pinned",
+    "updated_at": "2026-04-15T11:42:44Z"
   },
   {
-    "additions": 68,
-    "author": "Yanis01682",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - preserve existing void/ignore label 255 when educe_label is applied - only reduce non-ignore class ids by 1 for both torch and PIL image processors - add a SegFormer regression test covering",
-    "changed_files": 11,
+    "additions": 4,
+    "author": "paulinebm",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45423",
-    "created_at": "2026-04-14T01:47:54Z",
-    "deletions": 29,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45442",
+    "created_at": "2026-04-14T18:49:49Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45423/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45423",
+    "files_url": "https://github.com/huggingface/transformers/pull/45442/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45442",
     "labels": [],
-    "merged": false,
-    "number": 45423,
+    "merged": true,
+    "number": 45442,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix void segmentation map label reduction",
-    "updated_at": "2026-04-15T10:32:47Z"
+    "state": "closed",
+    "title": "Update workflow references to new commit hash",
+    "updated_at": "2026-04-14T18:59:01Z"
   },
   {
-    "additions": 58,
-    "author": "qgallouedec",
-    "author_association": "MEMBER",
-    "body_excerpt": "In `apply_chat_template`, drop the `content` key from messages when its value is `None` before passing to the Jinja template. ## Why this is a bug fix, not a breaking change `content=None` means \"there is no content\", it is semantically id\u2026",
-    "changed_files": 4,
+    "additions": 9,
+    "author": "casinca",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Please see fix #45440 for more details Discussed with @vasqu Also fixed via regen/inheritance: - exaone-moe - glm4-moe - glm4v-moe - glm4-moe-lite - glm_moe_dsa - nemotron_h - solar_open &nbsp; > [!WARNING] > > **Ed\u2026",
+    "changed_files": 9,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45422",
-    "created_at": "2026-04-14T00:53:00Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45441",
+    "created_at": "2026-04-14T17:51:08Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45422/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45422",
+    "files_url": "https://github.com/huggingface/transformers/pull/45441/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45441",
     "labels": [],
     "merged": false,
-    "number": 45422,
+    "number": 45441,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Drop `content=None` from messages in `apply_chat_template`",
-    "updated_at": "2026-04-14T16:08:31Z"
+    "title": "fix(DSV3): parity between native `DeepseekV3MoE` and remote official implementation",
+    "updated_at": "2026-04-15T09:27:41Z"
   },
   {
-    "additions": 178,
-    "author": "yonigozlan",
+    "additions": 1,
+    "author": "McPatate",
     "author_association": "MEMBER",
-    "body_excerpt": "Split off from the RF-DETR PR (#36895). These are changes to `core_model_loading.py` and `test_modeling_common.py` that are needed to load and save RF-DETR checkpoints correctly and in the same format as the original roboflow checkpoints,\u2026",
-    "changed_files": 6,
+    "body_excerpt": "Bumping safetensors to its latest (pre)released version! cc @tarekziade",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45421",
-    "created_at": "2026-04-13T23:51:08Z",
-    "deletions": 197,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45439",
+    "created_at": "2026-04-14T17:20:51Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45421/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45421",
+    "files_url": "https://github.com/huggingface/transformers/pull/45439/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45439",
     "labels": [],
     "merged": false,
-    "number": 45421,
-    "review_comments_count": 2,
-    "state": "open",
-    "title": "Improve nested `base_model_prefix` handling in weight conversion and loading",
-    "updated_at": "2026-04-15T16:32:23Z"
+    "number": 45439,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "feat: bump min safetensors version to `0.8.0-rc.0`",
+    "updated_at": "2026-04-15T11:54:37Z"
   },
   {
-    "additions": 40,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "As per title, we do not want to have proper `nn.Module`s to be registered for kernels exchanged functions - they are not proper modules (and they are never called as such)! They act as exchange format for kernels but functionally they shou\u2026",
-    "changed_files": 2,
+    "additions": 275,
+    "author": "Charly21r",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #45373 Adds `Gemma4TextForSequenceClassification` and `Gemma4ForSequenceClassification` to `transformers.models.gemma4`, following the same pattern established by Gemma 3. Prior to this change, `AutoModelForSe\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45420",
-    "created_at": "2026-04-13T23:36:14Z",
-    "deletions": 20,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45420/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45420",
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45438",
+    "created_at": "2026-04-14T17:20:36Z",
+    "deletions": 11,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45438/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45438",
     "labels": [],
     "merged": false,
-    "number": 45420,
-    "review_comments_count": 0,
+    "number": 45438,
+    "review_comments_count": 12,
     "state": "open",
-    "title": "[`Kernels`] Fix kernel function registration",
-    "updated_at": "2026-04-13T23:49:46Z"
+    "title": "Add Gemma4ForSequenceClassification",
+    "updated_at": "2026-04-16T05:29:26Z"
   },
   {
-    "additions": 43,
-    "author": "qgallouedec",
+    "additions": 4,
+    "author": "tomaarsen",
     "author_association": "MEMBER",
-    "body_excerpt": "- `get_processor_inputs_from_messages` builds a new dict per message with only `role` and `content`, dropping `tool_calls` and `tool_call_id`. This means `apply_chat_template` can't properly render multi-turn tool-use conversations. - Forw\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45418",
-    "created_at": "2026-04-13T19:44:03Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45437",
+    "created_at": "2026-04-14T17:14:24Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45418/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45418",
+    "files_url": "https://github.com/huggingface/transformers/pull/45437/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45437",
     "labels": [],
     "merged": true,
-    "number": 45418,
-    "review_comments_count": 3,
+    "number": 45437,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "[serve] Forward `tool_calls`/`tool_call_id` in processor inputs",
-    "updated_at": "2026-04-15T20:57:20Z"
+    "title": "Fix spurious position_ids warnings for at least 40 architectures",
+    "updated_at": "2026-04-16T12:19:46Z"
   },
   {
-    "additions": 746,
-    "author": "tarekziade",
+    "additions": 19,
+    "author": "AmineDiro",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Adds type checking to `src/transformers/*.py`",
-    "changed_files": 42,
+    "body_excerpt": "## Summary - Added `base_model_ep_plan` to `Qwen3MoeConfig` enabling expert parallelism via `DistributedConfig(enable_expert_parallel=True)` - Fix two small bugs in `GroupedGemmParallel` that prevented EP from working with 2D device meshes\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45415",
-    "created_at": "2026-04-13T17:09:40Z",
-    "deletions": 517,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45436",
+    "created_at": "2026-04-14T15:35:56Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45415/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45415",
+    "files_url": "https://github.com/huggingface/transformers/pull/45436/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45436",
     "labels": [],
     "merged": false,
-    "number": 45415,
+    "number": 45436,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Adds type checking to `src/transformers/*py`",
-    "updated_at": "2026-04-13T18:09:59Z"
+    "title": "Add expert parallelism (EP) support for Qwen3 MoE + fix GroupedGemmParallel for 2D meshes",
+    "updated_at": "2026-04-16T10:29:58Z"
   },
   {
-    "additions": 12,
-    "author": "ArthurZucker",
+    "additions": 30,
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": "## Summary Fixes #45137. Re-opened from #45395 on a same-repo branch so CI can run. Since #41147, attention layers are decorated with `@use_kernelized_func(apply_rotary_pos_emb)` which attaches a `rotary_fn` child `nn.Module` at init when\u2026",
-    "changed_files": 4,
+    "body_excerpt": "fixed https://github.com/huggingface/transformers/issues/44869 add check to not index past decoded chars with special tokens",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45414",
-    "created_at": "2026-04-13T16:19:15Z",
-    "deletions": 3,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45435",
+    "created_at": "2026-04-14T15:28:20Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45414/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45414",
-    "labels": [
-      "for patch"
-    ],
-    "merged": true,
-    "number": 45414,
+    "files_url": "https://github.com/huggingface/transformers/pull/45435/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45435",
+    "labels": [],
+    "merged": false,
+    "number": 45435,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix IndexError with DeepSpeed ZeRO-3 when kernels rotary is active",
-    "updated_at": "2026-04-13T16:38:53Z"
+    "state": "open",
+    "title": "do not index past decoded chars with special tokens",
+    "updated_at": "2026-04-15T13:30:22Z"
   },
   {
-    "additions": 12,
-    "author": "ezylopx5",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "I ran into an edge case in eta sampling where `EtaLogitsWarper` crashes if a row is fully masked (`scores == -inf` for all tokens). The previous entropy computation used `Categorical(logits=scores).entropy()`, which fails on that input. Th\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45413",
-    "created_at": "2026-04-13T15:54:45Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45413/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45413",
-    "labels": [],
-    "merged": false,
-    "number": 45413,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix EtaLogitsWarper on fully masked logits",
-    "updated_at": "2026-04-15T13:36:39Z"
-  },
-  {
-    "additions": 12,
-    "author": "Rocketknight1",
+    "additions": 95,
+    "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "This PR fixes the response schema used in the Gemma4 conversion script. The change has already been made in the Hub repos themselves, so I'm just copying the updated schema into the codebase!",
+    "body_excerpt": "# What does this PR do? This PR updates the grad acc tests so that we are able to better catch issues. We check the grad norm also in addition to the loss. We are remove some redundant tests. As promised @vasqu. Just read the new test dire\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45411",
-    "created_at": "2026-04-13T15:31:33Z",
-    "deletions": 6,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45434",
+    "created_at": "2026-04-14T15:17:03Z",
+    "deletions": 282,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45411/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45411",
+    "files_url": "https://github.com/huggingface/transformers/pull/45434/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45434",
     "labels": [],
     "merged": true,
-    "number": 45411,
-    "review_comments_count": 0,
+    "number": 45434,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Fix the response schema for the gemma4 converter",
-    "updated_at": "2026-04-14T10:21:48Z"
+    "title": "better grad acc tests",
+    "updated_at": "2026-04-15T13:10:22Z"
   },
   {
-    "additions": 2,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": null,
-    "changed_files": 1,
+    "additions": 115,
+    "author": "IlyasMoutawwakil",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45410",
-    "created_at": "2026-04-13T14:36:01Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45433",
+    "created_at": "2026-04-14T14:33:39Z",
+    "deletions": 0,
     "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45410/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45410",
+    "files_url": "https://github.com/huggingface/transformers/pull/45433/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45433",
     "labels": [],
     "merged": false,
-    "number": 45410,
+    "number": 45433,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(altclip): fix failing tests",
-    "updated_at": "2026-04-14T05:39:18Z"
+    "state": "open",
+    "title": "SonicMoe",
+    "updated_at": "2026-04-14T14:47:50Z"
   },
   {
-    "additions": 310,
-    "author": "3outeille",
+    "additions": 298,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "## Summary - Full `distributed_config` integration in `from_pretrained()` \u2014 mesh creation, apply TP + FSDP, attach `model.device_mesh` - `gather_full_state_dict()` for streaming DTensor\u2192full tensor saving (rank 0 only) - `convert_strided_t\u2026",
+    "body_excerpt": "# What does this PR do? This patch adds the pipeline in our type checker, and also adds mccabe complexity in our Ruff linter with a 75 threshold for now. To reduce complexity, logical blocks are now splitted into private functions to impro\u2026",
     "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45409",
-    "created_at": "2026-04-13T14:27:38Z",
-    "deletions": 49,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45432",
+    "created_at": "2026-04-14T14:04:07Z",
+    "deletions": 193,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45409/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45409",
+    "files_url": "https://github.com/huggingface/transformers/pull/45432/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45432",
     "labels": [],
-    "merged": true,
-    "number": 45409,
-    "review_comments_count": 4,
-    "state": "closed",
-    "title": "from_pretrained orchestration + distributed save/load",
-    "updated_at": "2026-04-14T16:18:11Z"
+    "merged": false,
+    "number": 45432,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "chore(qa): split pipeline and add type checking",
+    "updated_at": "2026-04-15T12:38:45Z"
   },
   {
-    "additions": 697,
-    "author": "3outeille",
-    "author_association": "MEMBER",
-    "body_excerpt": "## Summary - Extends the TPStyle API (from #45028) with MoE expert parallelism and sequence parallelism support - Adds `PackedColwiseParallel`, `MoEExpertsParallel`, `PrepareModuleInputOutput`, `_AllReduceBackward` custom ParallelStyle sub\u2026",
-    "changed_files": 22,
+    "additions": 3,
+    "author": "ambroiseodt",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fix checkpoint path in Dinov2 model_docs. # What does this PR do? Fixes the wrong checkpoint path in Dinov2 model_docs #45431 Changes made: the previous checkpoint (google/dinov2-base-patch16-224) does not exist and has been changed to the\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45408",
-    "created_at": "2026-04-13T14:25:08Z",
-    "deletions": 421,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45430",
+    "created_at": "2026-04-14T13:45:51Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45408/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45408",
+    "files_url": "https://github.com/huggingface/transformers/pull/45430/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45430",
     "labels": [],
     "merged": true,
-    "number": 45408,
+    "number": 45430,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "MoE expert parallelism + sequence parallelism",
-    "updated_at": "2026-04-14T16:25:22Z"
+    "title": "[Doc] Correct checkpoint path in Dinov2 model_docs ",
+    "updated_at": "2026-04-15T09:00:57Z"
   },
   {
-    "additions": 6,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "trainer: @SunMarc , pls help review, thx!",
+    "additions": 5,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Improve workflow file",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45407",
-    "created_at": "2026-04-13T14:18:37Z",
-    "deletions": 2,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45429",
+    "created_at": "2026-04-14T09:44:36Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45407/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45407",
+    "files_url": "https://github.com/huggingface/transformers/pull/45429/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45429",
     "labels": [],
-    "merged": true,
-    "number": 45407,
+    "merged": false,
+    "number": 45429,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "avoid wrap 4bit-quantized model into DP",
-    "updated_at": "2026-04-15T16:57:03Z"
+    "title": "Improve workflow file",
+    "updated_at": "2026-04-14T09:49:38Z"
   },
   {
-    "additions": 2,
-    "author": "vasqu",
+    "additions": 75,
+    "author": "tomaarsen",
     "author_association": "MEMBER",
-    "body_excerpt": "As per title, it is likely this model will be deprecated but I want to have a quick fix here by moving the instruct tokenizer to the internal testing repo",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45404",
-    "created_at": "2026-04-13T12:41:36Z",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45428",
+    "created_at": "2026-04-14T09:39:32Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45404/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45404",
+    "files_url": "https://github.com/huggingface/transformers/pull/45428/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45428",
     "labels": [],
     "merged": true,
-    "number": 45404,
+    "number": 45428,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[`Tokenizers`] Move gpt sw3 tokenizer out",
-    "updated_at": "2026-04-13T12:51:38Z"
+    "title": "[`fix`] PEFT integration fixes preventing save/load & integration",
+    "updated_at": "2026-04-14T11:27:26Z"
   },
   {
-    "additions": 14,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": null,
+    "additions": 1,
+    "author": "Aftabbs",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes #45341. `get_device_properties()` in `testing_utils.py` calls `torch.cuda.get_device_capability()` whenever `IS_CUDA_SYSTEM or IS_ROCM_SYSTEM` is `True`. This raises a `RuntimeError` on environments where CUD\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45403",
-    "created_at": "2026-04-13T12:25:48Z",
-    "deletions": 28,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45427",
+    "created_at": "2026-04-14T08:56:45Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45403/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45403",
+    "files_url": "https://github.com/huggingface/transformers/pull/45427/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45427",
     "labels": [],
-    "merged": true,
-    "number": 45403,
-    "review_comments_count": 1,
+    "merged": false,
+    "number": 45427,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(clipseg): fix 2 failing tests",
-    "updated_at": "2026-04-15T15:59:29Z"
+    "title": "fix(testing_utils): guard get_device_capability() with torch.cuda.is_available()",
+    "updated_at": "2026-04-16T05:45:31Z"
   },
   {
-    "additions": 44,
-    "author": "saslifat-gif",
+    "additions": 1333,
+    "author": "kmswin1",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #45397 ## What does this PR do? Fixes #45397 **Root cause:** `_load_state_dict_into_zero3_model` in `src/transformers/integrations/deepspeed.py` only iterates over `named_parameters` \u2014 never `named_buffers`. Buffers registered via `r\u2026",
-    "changed_files": 2,
+    "body_excerpt": "Add A.X K1 model architecture What does this PR do? This PR adds support for A.X K1, a large-scale Mixture-of-Experts (MoE) language model developed by [SK Telecom](https://huggingface.co/skt). A.X K1 contains 519B total parameters with 33\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45402",
-    "created_at": "2026-04-13T12:21:02Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45426",
+    "created_at": "2026-04-14T07:23:05Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45402/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45402",
+    "files_url": "https://github.com/huggingface/transformers/pull/45426/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45426",
     "labels": [],
     "merged": false,
-    "number": 45402,
-    "review_comments_count": 2,
+    "number": 45426,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Fix ZeRO-3 from_pretrained: load registered buffers in _load_state_dict_into_zero3_model",
-    "updated_at": "2026-04-15T06:11:00Z"
+    "title": "Feature/add axk1",
+    "updated_at": "2026-04-14T07:38:58Z"
   },
   {
-    "additions": 3157,
-    "author": "sachinkumarsingh092",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 16,
+    "additions": 124,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Adds type checking to `modeling_utils.py`",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45401",
-    "created_at": "2026-04-13T11:26:48Z",
-    "deletions": 1,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45401/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45401",
-    "labels": [
-      "New model",
-      "Audio"
-    ],
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45425",
+    "created_at": "2026-04-14T06:56:00Z",
+    "deletions": 64,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45425/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45425",
+    "labels": [],
     "merged": false,
-    "number": 45401,
-    "review_comments_count": 0,
+    "number": 45425,
+    "review_comments_count": 13,
     "state": "open",
-    "title": "Add support for Voxtral-4B-TTS-2603 to transformers",
-    "updated_at": "2026-04-13T13:38:15Z"
+    "title": "chore(typing): added modeling_utils to ty",
+    "updated_at": "2026-04-16T09:17:06Z"
   },
   {
-    "additions": 407,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45381 but it is weird, I remember checking position ids by value as well in qwen2.5 to verify that time-interval works \ud83e\udd14 update: i know why, the integration t\u2026",
-    "changed_files": 18,
+    "additions": 189,
+    "author": "louzongzhi",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR implements **IndexCache** support for GLM5's DeepSeek Sparse Attention (DSA), enabling cross-layer index reuse to accelerate long-context inference. IndexCache accelerates sparse attention by reusing top-k t\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45400",
-    "created_at": "2026-04-13T11:01:34Z",
-    "deletions": 375,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45424",
+    "created_at": "2026-04-14T05:14:08Z",
+    "deletions": 46,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45400/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45400",
-    "labels": [
-      "for patch"
-    ],
-    "merged": true,
-    "number": 45400,
-    "review_comments_count": 14,
-    "state": "closed",
-    "title": "Fix Qwen2.5VL temporal grid positions",
-    "updated_at": "2026-04-14T13:03:42Z"
+    "files_url": "https://github.com/huggingface/transformers/pull/45424/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45424",
+    "labels": [],
+    "merged": false,
+    "number": 45424,
+    "review_comments_count": 10,
+    "state": "open",
+    "title": "Add IndexCache support for GLM5 DSA",
+    "updated_at": "2026-04-16T14:16:08Z"
   },
   {
-    "additions": 73,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Credits to @xenova , adds an example code since it is more complicated that text LLMs. Beginner users might not know and an example is pretty much useful for them",
-    "changed_files": 2,
+    "additions": 68,
+    "author": "Yanis01682",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary - preserve existing void/ignore label 255 when educe_label is applied - only reduce non-ignore class ids by 1 for both torch and PIL image processors - add a SegFormer regression test covering",
+    "changed_files": 11,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45398",
-    "created_at": "2026-04-13T09:28:34Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45423",
+    "created_at": "2026-04-14T01:47:54Z",
+    "deletions": 29,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45398/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45398",
+    "files_url": "https://github.com/huggingface/transformers/pull/45423/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45423",
     "labels": [],
-    "merged": true,
-    "number": 45398,
-    "review_comments_count": 3,
-    "state": "closed",
-    "title": "Add example for iterative chatting with MLLMs",
-    "updated_at": "2026-04-15T11:45:19Z"
+    "merged": false,
+    "number": 45423,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix void segmentation map label reduction",
+    "updated_at": "2026-04-15T10:32:47Z"
   },
   {
-    "additions": 2508,
-    "author": "IlyasMoutawwakil",
+    "additions": 58,
+    "author": "qgallouedec",
     "author_association": "MEMBER",
-    "body_excerpt": "needed both claude and copilot's help on this one \ud83d\ude05 The idea is to make the vlms and their visual/audio encders compileable / exportable. here's a demo of the full model forward being compileable with these precomputed tensors. ```python \"\u2026",
-    "changed_files": 45,
+    "body_excerpt": "In `apply_chat_template`, drop the `content` key from messages when its value is `None` before passing to the Jinja template. ## Why this is a bug fix, not a breaking change `content=None` means \"there is no content\", it is semantically id\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45396",
-    "created_at": "2026-04-13T08:46:10Z",
-    "deletions": 2054,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45422",
+    "created_at": "2026-04-14T00:53:00Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45396/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45396",
+    "files_url": "https://github.com/huggingface/transformers/pull/45422/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45422",
     "labels": [],
     "merged": false,
-    "number": 45396,
-    "review_comments_count": 24,
+    "number": 45422,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Extract dynamic vision/audio tensors into standalone pure functions",
-    "updated_at": "2026-04-15T14:22:15Z"
+    "title": "Drop `content=None` from messages in `apply_chat_template`",
+    "updated_at": "2026-04-16T13:25:49Z"
   },
   {
-    "additions": 12,
-    "author": "ArthurZucker",
+    "additions": 178,
+    "author": "yonigozlan",
     "author_association": "MEMBER",
-    "body_excerpt": "## Summary Fixes #45137. Since #41147, attention layers are decorated with `@use_kernelized_func(apply_rotary_pos_emb)` which attaches a `rotary_fn` child `nn.Module` at init when the `kernels` library is available. DeepSpeed ZeRO-3's para\u2026",
-    "changed_files": 4,
+    "body_excerpt": "Split off from the RF-DETR PR (#36895). These are changes to `core_model_loading.py` and `test_modeling_common.py` that are needed to load and save RF-DETR checkpoints correctly and in the same format as the original roboflow checkpoints,\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45395",
-    "created_at": "2026-04-13T08:35:06Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45421",
+    "created_at": "2026-04-13T23:51:08Z",
+    "deletions": 197,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45395/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45395",
+    "files_url": "https://github.com/huggingface/transformers/pull/45421/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45421",
     "labels": [],
     "merged": false,
-    "number": 45395,
-    "review_comments_count": 4,
-    "state": "closed",
-    "title": "Fix IndexError with DeepSpeed ZeRO-3 when kernels rotary is active",
-    "updated_at": "2026-04-13T16:18:40Z"
+    "number": 45421,
+    "review_comments_count": 2,
+    "state": "open",
+    "title": "Improve nested `base_model_prefix` handling in weight conversion and loading",
+    "updated_at": "2026-04-15T16:32:23Z"
   },
   {
-    "additions": 47,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": null,
+    "additions": 40,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "As per title, we do not want to have proper `nn.Module`s to be registered for kernels exchanged functions - they are not proper modules (and they are never called as such)! They act as exchange format for kernels but functionally they shou\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45394",
-    "created_at": "2026-04-13T08:28:21Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45394/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45394",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45420",
+    "created_at": "2026-04-13T23:36:14Z",
+    "deletions": 20,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45420/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45420",
     "labels": [],
     "merged": false,
-    "number": 45394,
+    "number": 45420,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix(x_clip): fix 8 failed test cases",
-    "updated_at": "2026-04-14T07:44:33Z"
+    "title": "[`Kernels`] Fix kernel function registration",
+    "updated_at": "2026-04-13T23:49:46Z"
   },
   {
-    "additions": 0,
-    "author": "tarekziade",
+    "additions": 43,
+    "author": "qgallouedec",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Remove accidental addition of an ignored file",
-    "changed_files": 1,
+    "body_excerpt": "- `get_processor_inputs_from_messages` builds a new dict per message with only `role` and `content`, dropping `tool_calls` and `tool_call_id`. This means `apply_chat_template` can't properly render multi-turn tool-use conversations. - Forw\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45392",
-    "created_at": "2026-04-13T07:14:44Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45418",
+    "created_at": "2026-04-13T19:44:03Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45392/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45392",
+    "files_url": "https://github.com/huggingface/transformers/pull/45418/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45418",
     "labels": [],
     "merged": true,
-    "number": 45392,
-    "review_comments_count": 0,
+    "number": 45418,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "remove cache file from tree",
-    "updated_at": "2026-04-13T08:24:27Z"
+    "title": "[serve] Forward `tool_calls`/`tool_call_id` in processor inputs",
+    "updated_at": "2026-04-15T20:57:20Z"
   },
   {
-    "additions": 461,
+    "additions": 746,
     "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Similarly to the VLM tester, this patch introduces a audio tester class, used in - Qwen2Audio - AudioFlamingo3 - GraniteSpeech Adding a new audio-language model using this will require ~8-20 lines for the tester (vs\u2026",
-    "changed_files": 5,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45391",
-    "created_at": "2026-04-13T06:32:49Z",
-    "deletions": 471,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45391/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45391",
-    "labels": [],
-    "merged": false,
-    "number": 45391,
-    "review_comments_count": 9,
-    "state": "open",
-    "title": "audio tester class",
-    "updated_at": "2026-04-13T12:38:16Z"
-  },
-  {
-    "additions": 32,
-    "author": "ruben-aghayan",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR fails loudly when using repetition penalty on input_embed without input_ids args. Previously, users were able to call repetition penalty on generate calls with input_embeds args. Since they don't actually ha\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Adds type checking to `src/transformers/*.py`",
+    "changed_files": 42,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45389",
-    "created_at": "2026-04-13T03:57:04Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45415",
+    "created_at": "2026-04-13T17:09:40Z",
+    "deletions": 517,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45389/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45389",
+    "files_url": "https://github.com/huggingface/transformers/pull/45415/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45415",
     "labels": [],
     "merged": false,
-    "number": 45389,
+    "number": 45415,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Require input_ids for repetition penalty",
-    "updated_at": "2026-04-14T10:22:01Z"
+    "title": "Adds type checking to `src/transformers/*py`",
+    "updated_at": "2026-04-13T18:09:59Z"
   },
   {
-    "additions": 45,
-    "author": "albertorkive",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Makes `Gemma4ClippableLinear` inherit from `nn.Linear` instead of wrapping one via composition, enabling PEFT/LoRA to discover and target vision/audio encoder layers. **Problem:** PEFT's LoRA module discovery uses `\u2026",
+    "additions": 12,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "## Summary Fixes #45137. Re-opened from #45395 on a same-repo branch so CI can run. Since #41147, attention layers are decorated with `@use_kernelized_func(apply_rotary_pos_emb)` which attaches a `rotary_fn` child `nn.Module` at init when\u2026",
     "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45388",
-    "created_at": "2026-04-12T17:02:39Z",
-    "deletions": 31,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45388/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45388",
-    "labels": [],
-    "merged": false,
-    "number": 45388,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "Make Gemma4ClippableLinear inherit from nn.Linear for PEFT/LoRA compatibility",
-    "updated_at": "2026-04-14T15:36:20Z"
-  },
-  {
-    "additions": 35,
-    "author": "albertorkive",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes `attn_implementation=\"flash_attention_3\"` which is currently broken for the most common FA3 install method \u2014 the hopper wheel built from `flash-attention/hopper/`. **Three issues fixed:** 1. **`is_flash_attn_3\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45387",
-    "created_at": "2026-04-12T17:02:14Z",
-    "deletions": 15,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45387/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45387",
-    "labels": [],
-    "merged": false,
-    "number": 45387,
-    "review_comments_count": 4,
-    "state": "closed",
-    "title": "Fix flash_attention_3 detection and import for hopper wheel installs",
-    "updated_at": "2026-04-13T16:14:21Z"
-  },
-  {
-    "additions": 23,
-    "author": "UsamaKenway",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Optimizes memory usage when loading GGUF models by performing dtype casting immediately after dequantization. While I was adding the support for Gemma4 in this PR #45296, i noticed this issue that the GGUF tensors are dequantized to `float\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45386",
-    "created_at": "2026-04-12T13:17:17Z",
-    "deletions": 11,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45386/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45386",
-    "labels": [],
-    "merged": false,
-    "number": 45386,
-    "review_comments_count": 4,
-    "state": "open",
-    "title": "[GGUF] Reduce peak RAM usage by casting dequantized tensors early during load",
-    "updated_at": "2026-04-15T11:58:05Z"
-  },
-  {
-    "additions": 15,
-    "author": "songyuc",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Loading `openai/clip-vit-base-patch32` currently reports the following keys as unexpected: - `text_model.embeddings.position_ids` - `vision_model.embeddings.position_ids` In the current CLIP implementation, these b\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45385",
-    "created_at": "2026-04-12T12:53:49Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45385/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45385",
-    "labels": [],
-    "merged": false,
-    "number": 45385,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Ignore CLIP position_ids in unexpected key loading report",
-    "updated_at": "2026-04-12T15:05:10Z"
-  },
-  {
-    "additions": 25,
-    "author": "GitGlimpse895",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? `StoppingCriteriaList.__call__` previously evaluated every registered criterion unconditionally on every generation step, even after `is_done` was already `True` for all sequences in the batch. This adds a single `i\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45384",
-    "created_at": "2026-04-12T10:14:58Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45384/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45384",
-    "labels": [],
-    "merged": false,
-    "number": 45384,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "generation/stopping_criteria: short-circuit StoppingCriteriaList when all sequences are done",
-    "updated_at": "2026-04-14T03:23:23Z"
-  },
-  {
-    "additions": 6,
-    "author": "Aftabbs",
-    "author_association": "NONE",
-    "body_excerpt": "## Description Fixes #45290. `apply_chat_template(tokenize=True)` raises `KeyError: 'content'` when a conversation contains an assistant message that has `tool_calls` but no `content` key: ```python processor.apply_chat_template( [[ {\"role\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45383",
-    "created_at": "2026-04-12T08:48:26Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45414",
+    "created_at": "2026-04-13T16:19:15Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45383/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45383",
+    "files_url": "https://github.com/huggingface/transformers/pull/45414/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45414",
     "labels": [
-      "Code agent slop"
+      "for patch"
     ],
-    "merged": false,
-    "number": 45383,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(processing): guard message content access in apply_chat_template",
-    "updated_at": "2026-04-13T09:34:50Z"
-  },
-  {
-    "additions": 334,
-    "author": "zFlux",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Adds conversion from `facebook/audiogen-medium` (AudioCraft Hub layout: `state_dict.bin` + `compression_state_dict.bin`) to `MusicgenForConditionalGeneration`. - `convert_audiogen_transformers.py` \u2014 reuses `rename_state_dict` fr\u2026",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45382",
-    "created_at": "2026-04-12T03:38:01Z",
-    "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45382/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45382",
-    "labels": [],
-    "merged": false,
-    "number": 45382,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Add AudioGen (AudioCraft) to MusicGen conversion scripts",
-    "updated_at": "2026-04-12T03:39:00Z"
-  },
-  {
-    "additions": 11,
-    "author": "Brianzhengca",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45380",
-    "created_at": "2026-04-11T21:41:32Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45380/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45380",
-    "labels": [],
-    "merged": false,
-    "number": 45380,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix Qwen3_5MoeVisionConfig deepstack_visual_indexes silently dropped by @strict (Issue: https://github.com/huggingface/transformers/issues/45375)",
-    "updated_at": "2026-04-11T22:02:50Z"
-  },
-  {
-    "additions": 9,
-    "author": "hijingsong",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Problem The `@strict` decorator on `Qwen3_5MoeVisionConfig` silently drops the `deepstack_visual_indexes` field during config loading because it is not declared as a class attribute. Every Qwen3.5 MoE model on HuggingFace ships with thi\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45379",
-    "created_at": "2026-04-11T19:40:19Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45379/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45379",
-    "labels": [],
-    "merged": false,
-    "number": 45379,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "fix(config): add deepstack_visual_indexes to Qwen3_5MoeVisionConfig",
-    "updated_at": "2026-04-13T15:34:33Z"
-  },
-  {
-    "additions": 5,
-    "author": "hijingsong",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Fix: Guard ReasoningEffort import for older mistral_common versions Fixes #45372 ### Problem `ReasoningEffort` was added in `mistral-common>=1.10.0`, but the import in `tokenization_mistral_common.py` was unconditional within the `is_mi\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45378",
-    "created_at": "2026-04-11T18:26:14Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45378/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45378",
-    "labels": [],
-    "merged": false,
-    "number": 45378,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "fix(mistral): guard ReasoningEffort import for older mistral_common versions",
-    "updated_at": "2026-04-13T12:18:54Z"
-  },
-  {
-    "additions": 53,
-    "author": "Dippp10",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45377",
-    "created_at": "2026-04-11T14:17:32Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45377/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45377",
-    "labels": [],
-    "merged": false,
-    "number": 45377,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Enhance README documentation",
-    "updated_at": "2026-04-11T14:17:32Z"
-  },
-  {
-    "additions": 625,
-    "author": "mohamad-tohidi",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? i added an example for hierarchical text classification ## Before submitting - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). - [x] Did you read the [contributor\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45374",
-    "created_at": "2026-04-11T10:00:07Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45374/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45374",
-    "labels": [],
-    "merged": false,
-    "number": 45374,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Adding hierarchical classification example",
-    "updated_at": "2026-04-13T13:05:08Z"
-  },
-  {
-    "additions": 3,
-    "author": "HelloAnner",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #45341 When CUDA is installed but no GPU is available, `get_device_properties()` calls `torch.cuda.get_device_capability()` which fails because there is no CUDA device. The fix moves `import torch` to the top of the function and adds\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45371",
-    "created_at": "2026-04-11T06:22:52Z",
-    "deletions": 3,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45371/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45371",
-    "labels": [],
-    "merged": false,
-    "number": 45371,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: check CUDA availability before calling get_device_capability",
-    "updated_at": "2026-04-13T06:40:16Z"
-  },
-  {
-    "additions": 12,
-    "author": "RudrenduPaul",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes five documentation errors in the `Gemma3nTextConfig` docstring in `modular_gemma3n.py` (and the generated `configuration_gemma3n.py`): 1. **Typo**: `\"emebeddings\"` \u2192 `\"embeddings\"` in `hidden_size_per_layer_in\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45370",
-    "created_at": "2026-04-11T06:15:05Z",
-    "deletions": 12,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45370/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45370",
-    "labels": [],
-    "merged": true,
-    "number": 45370,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "docs: fix 5 docstring errors in Gemma3nTextConfig (typos, grammar, formatting)",
-    "updated_at": "2026-04-13T13:14:39Z"
-  },
-  {
-    "additions": 10,
-    "author": "sharziki",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #45245 \u2014 `torch.multinomial` crashes with `RuntimeError: number of categories cannot exceed 2^24` when `num_beams * vocab_size > 16,777,216` during beam search with `do_sample=True`. **Root cause:** In `_get_top_k_continua\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45369",
-    "created_at": "2026-04-11T02:42:07Z",
-    "deletions": 3,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45369/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45369",
-    "labels": [],
-    "merged": false,
-    "number": 45369,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(generation): handle CUDA multinomial limit in beam search sampling",
-    "updated_at": "2026-04-13T12:49:42Z"
-  },
-  {
-    "additions": 6,
-    "author": "sharziki",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #45362 \u2014 `transformers chat` crashes with `AttributeError: 'Qwen3VLProcessor' object has no attribute '_tokenizer'` when streaming responses from Qwen models. **Root cause:** `GenerateManager.generate_streaming()` and `CBG\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45368",
-    "created_at": "2026-04-11T02:34:32Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45368/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45368",
-    "labels": [],
-    "merged": true,
-    "number": 45368,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(serving): resolve rust tokenizer from ProcessorMixin in streaming generation",
-    "updated_at": "2026-04-13T15:01:51Z"
-  },
-  {
-    "additions": 15,
-    "author": "jackcook",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Update to the latest Four Over Six API by adding options to specify the data type of activations, weights, and gradients individually cc @SunMarc",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45367",
-    "created_at": "2026-04-11T01:47:51Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45367/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45367",
-    "labels": [],
-    "merged": false,
-    "number": 45367,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Add dtype config options for Four Over Six",
-    "updated_at": "2026-04-13T13:29:07Z"
-  },
-  {
-    "additions": 64,
-    "author": "owwll",
-    "author_association": "NONE",
-    "body_excerpt": "This PR addresses two separate issues: 1. **Fixes a bug in `Mistral4` RoPE dimension calculation.** The `Mistral4RotaryEmbedding` was incorrectly using the full `head_dim` to calculate the rotary dimension, instead of respecting the `parti\u2026",
-    "changed_files": 5,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45366",
-    "created_at": "2026-04-10T21:13:32Z",
-    "deletions": 11,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45366/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45366",
-    "labels": [],
-    "merged": false,
-    "number": 45366,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix OLMoE routing and Mistral4 RoPE dimensions",
-    "updated_at": "2026-04-10T21:40:40Z"
-  },
-  {
-    "additions": 46,
-    "author": "burtenshaw",
-    "author_association": "MEMBER",
-    "body_excerpt": "Applied the overlapping GPT-J refactor from the staged PRs: added `_can_record_outputs`, moved `GPTJModel.forward` to decorator-based output capture, switched wrapper forwards to `@can_return_tuple`, and removed manual hidden-state/attenti\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45365",
-    "created_at": "2026-04-10T19:37:53Z",
-    "deletions": 101,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45365/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45365",
-    "labels": [],
-    "merged": false,
-    "number": 45365,
+    "merged": true,
+    "number": 45414,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor GPT-J output tracing to use standardized decorators",
-    "updated_at": "2026-04-10T19:53:25Z"
+    "title": "Fix IndexError with DeepSpeed ZeRO-3 when kernels rotary is active",
+    "updated_at": "2026-04-13T16:38:53Z"
   },
   {
-    "additions": 815,
-    "author": "caiovicentino",
-    "author_association": "NONE",
-    "body_excerpt": "### Summary Adds a third backend to `QuantizedCache`: `polarquant`. Joins the existing `quanto` and `hqq` options and implements a Walsh-Hadamard rotation plus Lloyd-Max scalar quantization scheme tuned for KV cache compression. Pure PyTor\u2026",
-    "changed_files": 5,
+    "additions": 12,
+    "author": "ezylopx5",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "I ran into an edge case in eta sampling where `EtaLogitsWarper` crashes if a row is fully masked (`scores == -inf` for all tokens). The previous entropy computation used `Categorical(logits=scores).entropy()`, which fails on that input. Th\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45364",
-    "created_at": "2026-04-10T19:02:16Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45413",
+    "created_at": "2026-04-13T15:54:45Z",
     "deletions": 2,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45364/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45364",
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45413/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45413",
     "labels": [],
     "merged": false,
-    "number": 45364,
+    "number": 45413,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add PolarQuant backend to QuantizedCache (Hadamard-rotated Lloyd-Max)",
-    "updated_at": "2026-04-13T13:31:36Z"
-  },
-  {
-    "additions": 301,
-    "author": "michaelbenayoun",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds support for fusing multiple modules into a single kernel \u2014 the motivating case being fused RMSNorm+MLP kernels, but the API is generic. ## What changed - `FusedModuleBase`, `fuse_modules`, `unfuse_modul\u2026",
-    "changed_files": 5,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45363",
-    "created_at": "2026-04-10T18:53:22Z",
-    "deletions": 5,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45363/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45363",
-    "labels": [],
-    "merged": false,
-    "number": 45363,
-    "review_comments_count": 3,
     "state": "open",
-    "title": "n-to-1 kernel fusion via `KernelConfig`",
-    "updated_at": "2026-04-14T13:14:57Z"
+    "title": "Fix EtaLogitsWarper on fully masked logits",
+    "updated_at": "2026-04-15T13:36:39Z"
   },
   {
-    "additions": 98,
-    "author": "zucchini-nlp",
+    "additions": 12,
+    "author": "Rocketknight1",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/trl/issues/5497, also fixes https://github.com/huggingface/transformers/issues/45390 TL;DR; the base model prefix is never appended if it is part of a bigger VLM, which was true\u2026",
-    "changed_files": 4,
+    "body_excerpt": "This PR fixes the response schema used in the Gemma4 conversion script. The change has already been made in the Hub repos themselves, so I'm just copying the updated schema into the codebase!",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45361",
-    "created_at": "2026-04-10T13:49:28Z",
-    "deletions": 13,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45411",
+    "created_at": "2026-04-13T15:31:33Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45361/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45361",
+    "files_url": "https://github.com/huggingface/transformers/pull/45411/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45411",
     "labels": [],
-    "merged": false,
-    "number": 45361,
+    "merged": true,
+    "number": 45411,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add CLIP-like models in conversion to VLMs",
-    "updated_at": "2026-04-13T12:16:59Z"
+    "state": "closed",
+    "title": "Fix the response schema for the gemma4 converter",
+    "updated_at": "2026-04-14T10:21:48Z"
   },
   {
-    "additions": 3,
-    "author": "hanouticelina",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? `huggingface-cli` is deprecated and no longer maintained. This PR updates the remaining references with `hf`",
-    "changed_files": 3,
+    "additions": 2,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": null,
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45360",
-    "created_at": "2026-04-10T11:59:37Z",
-    "deletions": 3,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45360/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45360",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45410",
+    "created_at": "2026-04-13T14:36:01Z",
+    "deletions": 2,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45410/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45410",
     "labels": [],
     "merged": false,
-    "number": 45360,
+    "number": 45410,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Replace deprecated `huggingface-cli` references with `hf`",
-    "updated_at": "2026-04-10T12:13:18Z"
+    "state": "closed",
+    "title": "fix(altclip): fix failing tests",
+    "updated_at": "2026-04-14T05:39:18Z"
   },
   {
-    "additions": 3,
-    "author": "ArthurZucker",
+    "additions": 310,
+    "author": "3outeille",
     "author_association": "MEMBER",
-    "body_excerpt": "Fixes #45356 ## Summary - Remove `kimi_k25` from `MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS`: its remote `TikTokenTokenizer` is the only correct backend \u2014 the model has no `tokenizer.json`, and its `added_tokens_decoder` has non-sequential\u2026",
-    "changed_files": 2,
+    "body_excerpt": "## Summary - Full `distributed_config` integration in `from_pretrained()` \u2014 mesh creation, apply TP + FSDP, attach `model.device_mesh` - `gather_full_state_dict()` for streaming DTensor\u2192full tensor saving (rank 0 only) - `convert_strided_t\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45359",
-    "created_at": "2026-04-10T10:42:32Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45409",
+    "created_at": "2026-04-13T14:27:38Z",
+    "deletions": 49,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45359/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45359",
-    "labels": [
-      "for patch"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/45409/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45409",
+    "labels": [],
     "merged": true,
-    "number": 45359,
-    "review_comments_count": 0,
+    "number": 45409,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Fix Kimi-K2.5 tokenizer regression and _patch_mistral_regex AttributeError",
-    "updated_at": "2026-04-13T15:16:26Z"
+    "title": "from_pretrained orchestration + distributed save/load",
+    "updated_at": "2026-04-14T16:18:11Z"
   },
   {
-    "additions": 111,
-    "author": "Cyrilvallez",
+    "additions": 697,
+    "author": "3outeille",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/transformers/issues/45357 finally. This was not catched in the previous fix, as the model can be reloaded correctly by `from_pretrained`, but keys are still wrongly serialized! Aft\u2026",
-    "changed_files": 24,
+    "body_excerpt": "## Summary - Extends the TPStyle API (from #45028) with MoE expert parallelism and sequence parallelism support - Adds `PackedColwiseParallel`, `MoEExpertsParallel`, `PrepareModuleInputOutput`, `_AllReduceBackward` custom ParallelStyle sub\u2026",
+    "changed_files": 22,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45358",
-    "created_at": "2026-04-10T10:19:42Z",
-    "deletions": 42,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45408",
+    "created_at": "2026-04-13T14:25:08Z",
+    "deletions": 421,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45358/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45358",
-    "labels": [
-      "for patch"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/45408/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45408",
+    "labels": [],
     "merged": true,
-    "number": 45358,
-    "review_comments_count": 2,
-    "state": "closed",
-    "title": "Fix vlm weight mappings",
-    "updated_at": "2026-04-10T15:41:47Z"
-  },
-  {
-    "additions": 2297,
-    "author": "Shikhar-S",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR introduces [PhoneticXeus](https://arxiv.org/abs/2603.29042), which is the state-of-the-art universal phone recognizer trained on 70+ languages and evaluated on ~100 languages. The model should have high util\u2026",
-    "changed_files": 16,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45355",
-    "created_at": "2026-04-10T04:26:05Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45355/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45355",
-    "labels": [
-      "New model",
-      "Audio"
-    ],
-    "merged": false,
-    "number": 45355,
+    "number": 45408,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add universal phone recognition model - PhoneticXeus",
-    "updated_at": "2026-04-14T15:02:02Z"
+    "state": "closed",
+    "title": "MoE expert parallelism + sequence parallelism",
+    "updated_at": "2026-04-14T16:25:22Z"
   },
   {
-    "additions": 8,
-    "author": "winglian",
-    "author_association": "COLLABORATOR",
-    "body_excerpt": "# What does this PR do? Gemma 4 was calculating the CE loss incorrectly and not handling gradient accumulation steps properly, leading to losses scaled up by the value of the gradient accumulation steps rather than letting the built in HF\u2026",
-    "changed_files": 4,
+    "additions": 6,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "trainer: @SunMarc , pls help review, thx!",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45354",
-    "created_at": "2026-04-10T02:54:56Z",
-    "deletions": 76,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45407",
+    "created_at": "2026-04-13T14:18:37Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45354/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45354",
+    "files_url": "https://github.com/huggingface/transformers/pull/45407/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45407",
     "labels": [],
     "merged": true,
-    "number": 45354,
+    "number": 45407,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix gemma4 gradient accumulation loss and last token incorrect labels",
-    "updated_at": "2026-04-10T10:08:12Z"
+    "title": "avoid wrap 4bit-quantized model into DP",
+    "updated_at": "2026-04-15T16:57:03Z"
   },
   {
-    "additions": 30,
-    "author": "wilnn",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Add `**kwargs` to all methods in the `CallbackHandler` class. Previously, only the `CallbackHandler.on_push_begin()` method accepted `**kwargs`, while all other methods did not. This forces users who want to customi\u2026",
+    "additions": 2,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "As per title, it is likely this model will be deprecated but I want to have a quick fix here by moving the instruct tokenizer to the internal testing repo",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45353",
-    "created_at": "2026-04-09T23:14:20Z",
-    "deletions": 30,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45353/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45353",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45404",
+    "created_at": "2026-04-13T12:41:36Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45404/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45404",
     "labels": [],
     "merged": true,
-    "number": 45353,
+    "number": 45404,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "add kwargs to all methods in the CallbackHandler class",
-    "updated_at": "2026-04-10T14:16:46Z"
+    "title": "[`Tokenizers`] Move gpt sw3 tokenizer out",
+    "updated_at": "2026-04-13T12:51:38Z"
   },
   {
-    "additions": 4,
-    "author": "RudrenduPaul",
+    "additions": 14,
+    "author": "kaixuanliu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Corrects an incorrect return type annotation on `Qwen3MoeSparseMoeBlock.forward`. The method is annotated as returning `tuple[torch.Tensor, torch.Tensor]` but actually returns a single reshaped `torch.Tensor` (see\u2026",
-    "changed_files": 4,
+    "body_excerpt": null,
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45352",
-    "created_at": "2026-04-09T21:53:31Z",
-    "deletions": 4,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45403",
+    "created_at": "2026-04-13T12:25:48Z",
+    "deletions": 28,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45352/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45352",
+    "files_url": "https://github.com/huggingface/transformers/pull/45403/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45403",
     "labels": [],
     "merged": true,
-    "number": 45352,
-    "review_comments_count": 0,
+    "number": 45403,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "fix(qwen3_moe): correct return type annotation on Qwen3MoeSparseMoeBlock.forward",
-    "updated_at": "2026-04-13T14:07:30Z"
+    "title": "fix(clipseg): fix 2 failing tests",
+    "updated_at": "2026-04-15T15:59:29Z"
   },
   {
-    "additions": 2,
-    "author": "RudrenduPaul",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes a crash in `get_device_properties()` in `testing_utils.py` when CUDA is installed on the system but no GPU device is present (e.g., a CPU-only cloud studio with CUDA libraries installed). The function called\u2026",
-    "changed_files": 1,
+    "additions": 64,
+    "author": "saslifat-gif",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #45397 ## What does this PR do? Fixes #45397 **Root cause:** `_load_state_dict_into_zero3_model` in `src/transformers/integrations/deepspeed.py` only iterates over `named_parameters` \u2014 never `named_buffers`. Buffers registered via `r\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45351",
-    "created_at": "2026-04-09T21:51:42Z",
-    "deletions": 0,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45402",
+    "created_at": "2026-04-13T12:21:02Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45351/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45351",
+    "files_url": "https://github.com/huggingface/transformers/pull/45402/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45402",
     "labels": [],
     "merged": false,
-    "number": 45351,
-    "review_comments_count": 0,
+    "number": 45402,
+    "review_comments_count": 2,
     "state": "open",
-    "title": "fix(testing_utils): guard get_device_capability with torch.cuda.is_available()",
-    "updated_at": "2026-04-14T00:49:18Z"
+    "title": "Fix ZeRO-3 from_pretrained: load registered buffers in _load_state_dict_into_zero3_model",
+    "updated_at": "2026-04-16T22:11:13Z"
   },
   {
-    "additions": 1090,
-    "author": "mrutkows",
+    "additions": 3157,
+    "author": "sachinkumarsingh092",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Support new Granite 4 vision arch. <!-- Remove if not applicable --> Fixes # (issue) ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by c\u2026",
-    "changed_files": 8,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 16,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45350",
-    "created_at": "2026-04-09T17:46:37Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45401",
+    "created_at": "2026-04-13T11:26:48Z",
+    "deletions": 1,
     "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45350/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45350",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45401/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45401",
+    "labels": [
+      "New model",
+      "Audio"
+    ],
     "merged": false,
-    "number": 45350,
+    "number": 45401,
     "review_comments_count": 0,
     "state": "open",
-    "title": "WIP: Add support for Granite4VisionForConditionalGeneration",
-    "updated_at": "2026-04-10T12:34:50Z"
+    "title": "Add support for Voxtral-4B-TTS-2603 to transformers",
+    "updated_at": "2026-04-13T13:38:15Z"
   },
   {
-    "additions": 90,
-    "author": "florian6973",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #45305 Add a regression test in `TrainerGradientAccumulationTest` to avoid passing the GAS value to Accelerate by mistake Description: I force the value of the `num_steps` parameter to be 1, and the regression\u2026",
-    "changed_files": 2,
+    "additions": 407,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45381 but it is weird, I remember checking position ids by value as well in qwen2.5 to verify that time-interval works \ud83e\udd14 update: i know why, the integration t\u2026",
+    "changed_files": 18,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45349",
-    "created_at": "2026-04-09T17:24:39Z",
-    "deletions": 2,
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45400",
+    "created_at": "2026-04-13T11:01:34Z",
+    "deletions": 375,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45349/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45349",
+    "files_url": "https://github.com/huggingface/transformers/pull/45400/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45400",
     "labels": [
       "for patch"
     ],
     "merged": true,
-    "number": 45349,
-    "review_comments_count": 6,
+    "number": 45400,
+    "review_comments_count": 14,
     "state": "closed",
-    "title": "Fix #45305 + add regression test GAS",
-    "updated_at": "2026-04-13T14:41:43Z"
+    "title": "Fix Qwen2.5VL temporal grid positions",
+    "updated_at": "2026-04-14T13:03:42Z"
   },
   {
-    "additions": 50,
-    "author": "qgallouedec",
+    "additions": 73,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes #45290 ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by our ability to review and r\u2026",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? Credits to @xenova , adds an example code since it is more complicated that text LLMs. Beginner users might not know and an example is pretty much useful for them",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45348",
-    "created_at": "2026-04-09T15:59:07Z",
-    "deletions": 19,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45398",
+    "created_at": "2026-04-13T09:28:34Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45348/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45348",
+    "files_url": "https://github.com/huggingface/transformers/pull/45398/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45398",
     "labels": [],
     "merged": true,
-    "number": 45348,
-    "review_comments_count": 7,
+    "number": 45398,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Fix `apply_chat_template` crash on `tool_call` messages without content",
-    "updated_at": "2026-04-13T19:44:38Z"
+    "title": "Add example for iterative chatting with MLLMs",
+    "updated_at": "2026-04-15T11:45:19Z"
   },
   {
-    "additions": 35,
-    "author": "Cyrilvallez",
+    "additions": 2508,
+    "author": "IlyasMoutawwakil",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. `accelerate` destroys the dict otherwise, if it's not BOTH passed as kwarg AND part of `_skip_keys_device_placement`.......... `per_layer_input` needs to stay as a positional arg, for gradient chec\u2026",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45347",
-    "created_at": "2026-04-09T15:31:34Z",
-    "deletions": 6,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45347/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45347",
-    "labels": [],
-    "merged": true,
-    "number": 45347,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[gemma4] Fix device map auto",
-    "updated_at": "2026-04-09T15:45:15Z"
-  },
-  {
-    "additions": 16,
-    "author": "ionut-anghelina",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": null,
-    "changed_files": 2,
+    "body_excerpt": "needed both claude and copilot's help on this one \ud83d\ude05 The idea is to make the vlms and their visual/audio encders compileable / exportable. here's a demo of the full model forward being compileable with these precomputed tensors. ```python \"\u2026",
+    "changed_files": 45,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45346",
-    "created_at": "2026-04-09T14:48:28Z",
-    "deletions": 0,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45396",
+    "created_at": "2026-04-13T08:46:10Z",
+    "deletions": 2054,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45346/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45346",
+    "files_url": "https://github.com/huggingface/transformers/pull/45396/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45396",
     "labels": [],
     "merged": false,
-    "number": 45346,
-    "review_comments_count": 1,
+    "number": 45396,
+    "review_comments_count": 46,
     "state": "open",
-    "title": "Fix Double Application of Softmax for Router Logits in MoE models",
-    "updated_at": "2026-04-13T12:40:28Z"
-  },
-  {
-    "additions": 30,
-    "author": "ansley",
-    "author_association": "NONE",
-    "body_excerpt": "The `transformers` V5 \"rm slow tokenizers\" refactor (\\#40936) aliased `LlamaTokenizerFast` to `LlamaTokenizer`, whose `__init__` unconditionally installs a SentencePiece Metaspace pre-tokenizer. This is correct for classic Llama/Llama-2 mo\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45345",
-    "created_at": "2026-04-09T14:31:40Z",
-    "deletions": 14,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45345/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45345",
-    "labels": [],
-    "merged": false,
-    "number": 45345,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix ByteLevel-BPE tokenizers silently breaking in `LlamaTokenizer`",
-    "updated_at": "2026-04-10T12:45:24Z"
+    "title": "Extract dynamic vision/audio tensors into standalone pure functions",
+    "updated_at": "2026-04-16T13:56:44Z"
   },
   {
-    "additions": 6,
-    "author": "tarekziade",
+    "additions": 12,
+    "author": "ArthurZucker",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Simple hook to display test duration. This will append inline duration per test during the run, example: ``` tests/utils/test_configuration_utils.py::ConfigPushToHubTester::test_push_to_hub [gw1] [ 90%] PASSED tests\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary Fixes #45137. Since #41147, attention layers are decorated with `@use_kernelized_func(apply_rotary_pos_emb)` which attaches a `rotary_fn` child `nn.Module` at init when the `kernels` library is available. DeepSpeed ZeRO-3's para\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45344",
-    "created_at": "2026-04-09T14:22:46Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45395",
+    "created_at": "2026-04-13T08:35:06Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45344/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45344",
+    "files_url": "https://github.com/huggingface/transformers/pull/45395/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45395",
     "labels": [],
-    "merged": true,
-    "number": 45344,
-    "review_comments_count": 0,
+    "merged": false,
+    "number": 45395,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "refactor: display test duration",
-    "updated_at": "2026-04-09T15:19:26Z"
+    "title": "Fix IndexError with DeepSpeed ZeRO-3 when kernels rotary is active",
+    "updated_at": "2026-04-13T16:18:40Z"
   },
   {
-    "additions": 8,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
+    "additions": 59,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
     "body_excerpt": null,
-    "changed_files": 1,
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45342",
-    "created_at": "2026-04-09T14:13:15Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45394",
+    "created_at": "2026-04-13T08:28:21Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45342/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45342",
+    "files_url": "https://github.com/huggingface/transformers/pull/45394/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45394",
     "labels": [],
     "merged": false,
-    "number": 45342,
+    "number": 45394,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Use `_keys_to_ignore_on_load_unexpected/missing` recursively from children",
-    "updated_at": "2026-04-09T14:23:31Z"
+    "title": "fix(x_clip): fix 8 failed test cases",
+    "updated_at": "2026-04-16T03:41:07Z"
   },
   {
-    "additions": 17,
-    "author": "Cyrilvallez",
+    "additions": 0,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Supersedes https://github.com/huggingface/transformers/pull/45314 with a better fix. Fixes https://github.com/huggingface/transformers/issues/45216 and https://github.com/huggingface/transformers/issues/45310 and ht\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Remove accidental addition of an ignored file",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45340",
-    "created_at": "2026-04-09T12:02:14Z",
-    "deletions": 14,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45392",
+    "created_at": "2026-04-13T07:14:44Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45340/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45340",
+    "files_url": "https://github.com/huggingface/transformers/pull/45392/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45392",
     "labels": [],
     "merged": true,
-    "number": 45340,
+    "number": 45392,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix conversion mappings for vlms",
-    "updated_at": "2026-04-10T10:18:09Z"
+    "title": "remove cache file from tree",
+    "updated_at": "2026-04-13T08:24:27Z"
   },
   {
-    "additions": 156,
+    "additions": 522,
     "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The CircleCI config file is not ruff formatted, leading to unwanted changes when it's opened in an editor that follows our repository ruff configuration. This patch adds it and runs `make style` to update it",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Similarly to the VLM tester, this patch introduces a audio tester class, used in - Qwen2Audio - AudioFlamingo3 - GraniteSpeech Adding a new audio-language model using this will require ~8-20 lines for the tester (vs\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45339",
-    "created_at": "2026-04-09T09:44:16Z",
-    "deletions": 58,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45391",
+    "created_at": "2026-04-13T06:32:49Z",
+    "deletions": 471,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45339/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45339",
+    "files_url": "https://github.com/huggingface/transformers/pull/45391/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45391",
     "labels": [],
-    "merged": true,
-    "number": 45339,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "chore: added circleci python script to ruff and ty checkers",
-    "updated_at": "2026-04-09T12:00:08Z"
+    "merged": false,
+    "number": 45391,
+    "review_comments_count": 10,
+    "state": "open",
+    "title": "audio tester class",
+    "updated_at": "2026-04-16T09:19:31Z"
   },
   {
-    "additions": 37,
-    "author": "RudrenduPaul",
+    "additions": 32,
+    "author": "ruben-aghayan",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Closes #45162 ## What this PR does Expands the docstrings of `_can_set_attn_implementation` and `_can_set_experts_implementation` in `modeling_utils.py` to explicitly document the known limitations of their source-inspection heuristic. **C\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? This PR fails loudly when using repetition penalty on input_embed without input_ids args. Previously, users were able to call repetition penalty on generate calls with input_embeds args. Since they don't actually ha\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45338",
-    "created_at": "2026-04-09T09:35:52Z",
-    "deletions": 4,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45389",
+    "created_at": "2026-04-13T03:57:04Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45338/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45338",
+    "files_url": "https://github.com/huggingface/transformers/pull/45389/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45389",
     "labels": [],
     "merged": false,
-    "number": 45338,
+    "number": 45389,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "docs: document known limitations of _can_set_attn/experts_implementation source inspection",
-    "updated_at": "2026-04-09T13:43:04Z"
+    "state": "open",
+    "title": "Require input_ids for repetition penalty",
+    "updated_at": "2026-04-14T10:22:01Z"
   },
   {
-    "additions": 13,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Removing test_hub from CI for now",
-    "changed_files": 3,
+    "additions": 45,
+    "author": "albertorkive",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Makes `Gemma4ClippableLinear` inherit from `nn.Linear` instead of wrapping one via composition, enabling PEFT/LoRA to discover and target vision/audio encoder layers. **Problem:** PEFT's LoRA module discovery uses `\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45337",
-    "created_at": "2026-04-09T08:54:45Z",
-    "deletions": 30,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45388",
+    "created_at": "2026-04-12T17:02:39Z",
+    "deletions": 31,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45337/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45337",
+    "files_url": "https://github.com/huggingface/transformers/pull/45388/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45388",
     "labels": [],
-    "merged": true,
-    "number": 45337,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "chore: remove test_hub for now",
-    "updated_at": "2026-04-09T09:28:52Z"
+    "merged": false,
+    "number": 45388,
+    "review_comments_count": 3,
+    "state": "open",
+    "title": "Make Gemma4ClippableLinear inherit from nn.Linear for PEFT/LoRA compatibility",
+    "updated_at": "2026-04-14T15:36:20Z"
   },
   {
-    "additions": 84,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. Follow-up of https://github.com/huggingface/transformers/pull/45312. This removes the unnecessary weights, and silently skip them during loading, so that the checkpoints on the hub do not have to b\u2026",
-    "changed_files": 3,
+    "additions": 35,
+    "author": "albertorkive",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes `attn_implementation=\"flash_attention_3\"` which is currently broken for the most common FA3 install method \u2014 the hopper wheel built from `flash-attention/hopper/`. **Three issues fixed:** 1. **`is_flash_attn_3\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45336",
-    "created_at": "2026-04-09T08:43:55Z",
-    "deletions": 26,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45387",
+    "created_at": "2026-04-12T17:02:14Z",
+    "deletions": 15,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45336/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45336",
+    "files_url": "https://github.com/huggingface/transformers/pull/45387/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45387",
     "labels": [],
-    "merged": true,
-    "number": 45336,
-    "review_comments_count": 7,
+    "merged": false,
+    "number": 45387,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "[gemma4] Remove all shared weights, and silently skip them during loading",
-    "updated_at": "2026-04-09T13:23:33Z"
+    "title": "Fix flash_attention_3 detection and import for hopper wheel installs",
+    "updated_at": "2026-04-13T16:14:21Z"
   },
   {
-    "additions": 1333,
-    "author": "kmswin1",
+    "additions": 23,
+    "author": "UsamaKenway",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Add A.X K1 model architecture What does this PR do? This PR adds support for A.X K1, a large-scale Mixture-of-Experts (MoE) language model developed by [SK Telecom](https://huggingface.co/skt). A.X K1 contains 519B total parameters with 33\u2026",
-    "changed_files": 8,
+    "body_excerpt": "Optimizes memory usage when loading GGUF models by performing dtype casting immediately after dequantization. While I was adding the support for Gemma4 in this PR #45296, i noticed this issue that the GGUF tensors are dequantized to `float\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45334",
-    "created_at": "2026-04-09T06:21:43Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45386",
+    "created_at": "2026-04-12T13:17:17Z",
+    "deletions": 11,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45334/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45334",
+    "files_url": "https://github.com/huggingface/transformers/pull/45386/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45386",
     "labels": [],
     "merged": false,
-    "number": 45334,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Feature/add axk1",
-    "updated_at": "2026-04-14T07:23:08Z"
+    "number": 45386,
+    "review_comments_count": 4,
+    "state": "open",
+    "title": "[GGUF] Reduce peak RAM usage by casting dequantized tensors early during load",
+    "updated_at": "2026-04-15T11:58:05Z"
   },
   {
-    "additions": 471,
-    "author": "eladsegal",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds heterogeneous model support - the ability for individual layers to differ from the global config (e.g., different `intermediate_size`, `num_key_value_heads`) and to skip sub-modules entirely (MLP, attention, et\u2026",
-    "changed_files": 5,
+    "additions": 15,
+    "author": "songyuc",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Loading `openai/clip-vit-base-patch32` currently reports the following keys as unexpected: - `text_model.embeddings.position_ids` - `vision_model.embeddings.position_ids` In the current CLIP implementation, these b\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45333",
-    "created_at": "2026-04-09T06:18:11Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45385",
+    "created_at": "2026-04-12T12:53:49Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45333/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45333",
+    "files_url": "https://github.com/huggingface/transformers/pull/45385/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45385",
     "labels": [],
     "merged": false,
-    "number": 45333,
+    "number": 45385,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add heterogeneous config support (per-layer configuration)",
-    "updated_at": "2026-04-14T14:08:07Z"
+    "title": "Ignore CLIP position_ids in unexpected key loading report",
+    "updated_at": "2026-04-12T15:05:10Z"
   },
   {
-    "additions": 2152,
-    "author": "eladsegal",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds heterogeneous model support - the ability for individual layers to differ from the global config (e.g., different `intermediate_size`, `num_key_value_heads`) and to skip sub-modules entirely (MLP, attention, et\u2026",
-    "changed_files": 14,
+    "additions": 34,
+    "author": "GitGlimpse895",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? `StoppingCriteriaList.__call__` previously evaluated every registered criterion unconditionally on every generation step, even after `is_done` was already `True` for all sequences in the batch. This adds a single `i\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45332",
-    "created_at": "2026-04-09T05:56:31Z",
-    "deletions": 40,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45384",
+    "created_at": "2026-04-12T10:14:58Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45332/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45332",
+    "files_url": "https://github.com/huggingface/transformers/pull/45384/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45384",
     "labels": [],
     "merged": false,
-    "number": 45332,
+    "number": 45384,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add heterogeneous model support (per-layer config and modeling)",
-    "updated_at": "2026-04-15T04:50:09Z"
+    "title": "generation/stopping_criteria: short-circuit StoppingCriteriaList when all sequences are done",
+    "updated_at": "2026-04-16T10:37:20Z"
   },
-  {
-    "additions": 12,
-    "author": "Kash6",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "get_rope_index unconditionally applies tokens_per_second temporal scaling to both images and videos. For still images (modality_type == 1), this shifts the temporal position origin to start_position * tokens_per_second instead of start_pos\u2026",
-    "changed_files": 2,
+  {
+    "additions": 6,
+    "author": "Aftabbs",
+    "author_association": "NONE",
+    "body_excerpt": "## Description Fixes #45290. `apply_chat_template(tokenize=True)` raises `KeyError: 'content'` when a conversation contains an assistant message that has `tool_calls` but no `content` key: ```python processor.apply_chat_template( [[ {\"role\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45330",
-    "created_at": "2026-04-08T23:51:52Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45383",
+    "created_at": "2026-04-12T08:48:26Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45330/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45330",
+    "files_url": "https://github.com/huggingface/transformers/pull/45383/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45383",
     "labels": [
-      "for patch"
+      "Code agent slop"
     ],
-    "merged": true,
-    "number": 45330,
+    "merged": false,
+    "number": 45383,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Qwen2.5-VL temporal RoPE scaling applied to still images",
-    "updated_at": "2026-04-14T03:21:32Z"
+    "title": "fix(processing): guard message content access in apply_chat_template",
+    "updated_at": "2026-04-13T09:34:50Z"
   },
   {
-    "additions": 152,
-    "author": "abidlabs",
-    "author_association": "MEMBER",
-    "body_excerpt": "Updates `TrackioCallback` and `TrainingArguments` for the latest version of Trackio using HF Buckets as the backend, and control over creating a static Space for the Trackio dashboard during or at the end of training. These are now the `Tr\u2026",
-    "changed_files": 2,
+    "additions": 334,
+    "author": "zFlux",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Adds conversion from `facebook/audiogen-medium` (AudioCraft Hub layout: `state_dict.bin` + `compression_state_dict.bin`) to `MusicgenForConditionalGeneration`. - `convert_audiogen_transformers.py` \u2014 reuses `rename_state_dict` fr\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45329",
-    "created_at": "2026-04-08T22:36:08Z",
-    "deletions": 57,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45329/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45329",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45382",
+    "created_at": "2026-04-12T03:38:01Z",
+    "deletions": 0,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45382/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45382",
     "labels": [],
-    "merged": true,
-    "number": 45329,
-    "review_comments_count": 21,
-    "state": "closed",
-    "title": "Update `trackio` integration to use Buckets and \"freeze\" Space after training",
-    "updated_at": "2026-04-13T14:30:27Z"
+    "merged": false,
+    "number": 45382,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Add AudioGen (AudioCraft) to MusicGen conversion scripts",
+    "updated_at": "2026-04-12T03:39:00Z"
   },
   {
-    "additions": 9,
-    "author": "RyanMullins",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #45242 * Drops `k_proj`, `k_norm`, and `v_proj` weights for `Gemma4TextAttention` modules from the checkpoint if the layer shares KV cache values. These changes can also be adapted to Gemma 3n if that's desira\u2026",
-    "changed_files": 1,
+    "additions": 11,
+    "author": "Brianzhengca",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45328",
-    "created_at": "2026-04-08T20:43:42Z",
-    "deletions": 6,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45380",
+    "created_at": "2026-04-11T21:41:32Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45328/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45328",
+    "files_url": "https://github.com/huggingface/transformers/pull/45380/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45380",
     "labels": [],
     "merged": false,
-    "number": 45328,
+    "number": 45380,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Drop unused Gemma4TextAttention weights when sharing KV Cache",
-    "updated_at": "2026-04-09T18:31:13Z"
+    "state": "closed",
+    "title": "fix Qwen3_5MoeVisionConfig deepstack_visual_indexes silently dropped by @strict (Issue: https://github.com/huggingface/transformers/issues/45375)",
+    "updated_at": "2026-04-11T22:02:50Z"
   },
   {
-    "additions": 337,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "refactors the how to add a model with modular transformers doc: - structure: - flipped the order so you learn how to write the modular file first before generating it - remove the motivator examples with BERT/RoBERTa - merge the two `super\u2026",
+    "additions": 9,
+    "author": "hijingsong",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Problem The `@strict` decorator on `Qwen3_5MoeVisionConfig` silently drops the `deepstack_visual_indexes` field during config loading because it is not declared as a class attribute. Every Qwen3.5 MoE model on HuggingFace ships with thi\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45327",
-    "created_at": "2026-04-08T20:23:28Z",
-    "deletions": 408,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45327/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45327",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45379",
+    "created_at": "2026-04-11T19:40:19Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45379/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45379",
     "labels": [],
     "merged": false,
-    "number": 45327,
-    "review_comments_count": 1,
+    "number": 45379,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "[docs] modular transformers",
-    "updated_at": "2026-04-15T18:15:34Z"
+    "title": "fix(config): add deepstack_visual_indexes to Qwen3_5MoeVisionConfig",
+    "updated_at": "2026-04-13T15:34:33Z"
   },
   {
-    "additions": 19,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? \u2192 This PR introduces compat fixes across several audio models to ensure they can be loaded and used by a companion vLLM PR. <ins>These changes are deliberate and are blocking</ins> [this vLLM PR](https://github.co\u2026",
-    "changed_files": 11,
+    "additions": 5,
+    "author": "hijingsong",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Fix: Guard ReasoningEffort import for older mistral_common versions Fixes #45372 ### Problem `ReasoningEffort` was added in `mistral-common>=1.10.0`, but the import in `tokenization_mistral_common.py` was unconditional within the `is_mi\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45326",
-    "created_at": "2026-04-08T18:28:35Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45378",
+    "created_at": "2026-04-11T18:26:14Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45326/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45326",
+    "files_url": "https://github.com/huggingface/transformers/pull/45378/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45378",
     "labels": [],
     "merged": false,
-    "number": 45326,
+    "number": 45378,
     "review_comments_count": 0,
     "state": "open",
-    "title": "feat[vLLM \u00d7 v5]: Add vLLM compatibility for audio models",
-    "updated_at": "2026-04-14T15:45:32Z"
+    "title": "fix(mistral): guard ReasoningEffort import for older mistral_common versions",
+    "updated_at": "2026-04-13T12:18:54Z"
   },
   {
-    "additions": 236,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45276 and https://github.com/huggingface/transformers/issues/45335 In gemma4 per-layer inputs have to be resized as long as they aren't part of soft multimoda\u2026",
-    "changed_files": 16,
+    "additions": 53,
+    "author": "Dippp10",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45324",
-    "created_at": "2026-04-08T17:06:26Z",
-    "deletions": 53,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45377",
+    "created_at": "2026-04-11T14:17:32Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45324/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45324",
+    "files_url": "https://github.com/huggingface/transformers/pull/45377/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45377",
     "labels": [],
-    "merged": true,
-    "number": 45324,
-    "review_comments_count": 5,
-    "state": "closed",
-    "title": "Gemma4 resizing per layer inputs",
-    "updated_at": "2026-04-15T11:15:23Z"
+    "merged": false,
+    "number": 45377,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Enhance README documentation",
+    "updated_at": "2026-04-11T14:17:32Z"
   },
   {
-    "additions": 226,
-    "author": "remi-or",
-    "author_association": "MEMBER",
-    "body_excerpt": "# Summary This PR fixes the issue raised in https://github.com/huggingface/transformers/pull/45274 . CUDA graph reuse in continuous batching used (num_q_tokens, max_kv_read) as the graph cache key. However, FlashAttention varlen kernels al\u2026",
-    "changed_files": 7,
+    "additions": 625,
+    "author": "mohamad-tohidi",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? i added an example for hierarchical text classification ## Before submitting - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). - [x] Did you read the [contributor\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45323",
-    "created_at": "2026-04-08T16:30:18Z",
-    "deletions": 126,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45374",
+    "created_at": "2026-04-11T10:00:07Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45323/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45323",
+    "files_url": "https://github.com/huggingface/transformers/pull/45374/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45374",
     "labels": [],
     "merged": false,
-    "number": 45323,
+    "number": 45374,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[CB] Fix capture of max_seqlen",
-    "updated_at": "2026-04-15T01:41:00Z"
+    "state": "closed",
+    "title": "Adding hierarchical classification example",
+    "updated_at": "2026-04-13T13:05:08Z"
   },
   {
-    "additions": 20,
-    "author": "andrewor14",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "**Summary:** TorchAO recently deprecated AffineQuantizedTensor and related classes (pytorch/ao#2752). These will be removed in the next release. We should remove references of these classes in transformers before then. **Test Plan:** ``` p\u2026",
-    "changed_files": 2,
+    "additions": 3,
+    "author": "HelloAnner",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #45341 When CUDA is installed but no GPU is available, `get_device_properties()` calls `torch.cuda.get_device_capability()` which fails because there is no CUDA device. The fix moves `import torch` to the top of the function and adds\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45321",
-    "created_at": "2026-04-08T15:42:16Z",
-    "deletions": 29,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45371",
+    "created_at": "2026-04-11T06:22:52Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45321/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45321",
+    "files_url": "https://github.com/huggingface/transformers/pull/45371/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45371",
     "labels": [],
     "merged": false,
-    "number": 45321,
+    "number": 45371,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Remove references to torchao's AffineQuantizedTensor",
-    "updated_at": "2026-04-09T12:21:03Z"
+    "state": "closed",
+    "title": "fix: check CUDA availability before calling get_device_capability",
+    "updated_at": "2026-04-13T06:40:16Z"
   },
   {
-    "additions": 5,
-    "author": "Regata3010",
+    "additions": 12,
+    "author": "RudrenduPaul",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes a crash in assisted generation when using model pairs with different vocabulary sizes but the same tokenizer family (e.g., Qwen2.5-7B + Qwen2.5-0.5B). `map_input_embeddings` is only initialized when `len(self\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Fixes five documentation errors in the `Gemma3nTextConfig` docstring in `modular_gemma3n.py` (and the generated `configuration_gemma3n.py`): 1. **Typo**: `\"emebeddings\"` \u2192 `\"embeddings\"` in `hidden_size_per_layer_in\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45320",
-    "created_at": "2026-04-08T15:30:16Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45370",
+    "created_at": "2026-04-11T06:15:05Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45320/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45320",
+    "files_url": "https://github.com/huggingface/transformers/pull/45370/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45370",
     "labels": [],
     "merged": true,
-    "number": 45320,
+    "number": 45370,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix AttributeError in AssistantToTargetTranslator.unmap_input_ids with cross-vocab models",
-    "updated_at": "2026-04-10T17:46:37Z"
+    "title": "docs: fix 5 docstring errors in Gemma3nTextConfig (typos, grammar, formatting)",
+    "updated_at": "2026-04-13T13:14:39Z"
   },
   {
-    "additions": 78,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - Removes `HUGGINGFACE_CO_STAGING` when downloading artifacts - adds a retry mechanism for external URLs (with partial file cleanup)",
-    "changed_files": 6,
+    "additions": 10,
+    "author": "sharziki",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes #45245 \u2014 `torch.multinomial` crashes with `RuntimeError: number of categories cannot exceed 2^24` when `num_beams * vocab_size > 16,777,216` during beam search with `do_sample=True`. **Root cause:** In `_get_top_k_continua\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45319",
-    "created_at": "2026-04-08T14:51:48Z",
-    "deletions": 32,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45369",
+    "created_at": "2026-04-11T02:42:07Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45319/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45319",
+    "files_url": "https://github.com/huggingface/transformers/pull/45369/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45369",
     "labels": [],
-    "merged": true,
-    "number": 45319,
-    "review_comments_count": 3,
+    "merged": false,
+    "number": 45369,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: dont download artifacts from the test hub",
-    "updated_at": "2026-04-15T16:52:10Z"
+    "title": "fix(generation): handle CUDA multinomial limit in beam search sampling",
+    "updated_at": "2026-04-13T12:49:42Z"
   },
   {
-    "additions": 5,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? AutoTokenizer.register() adds classes to the global `REGISTERED_TOKENIZER_CLASSES` dict and some tests did not clean up behind them, leading to leaky state between tests",
+    "additions": 6,
+    "author": "sharziki",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes #45362 \u2014 `transformers chat` crashes with `AttributeError: 'Qwen3VLProcessor' object has no attribute '_tokenizer'` when streaming responses from Qwen models. **Root cause:** `GenerateManager.generate_streaming()` and `CBG\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45318",
-    "created_at": "2026-04-08T13:46:47Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45368",
+    "created_at": "2026-04-11T02:34:32Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45318/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45318",
+    "files_url": "https://github.com/huggingface/transformers/pull/45368/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45368",
     "labels": [],
     "merged": true,
-    "number": 45318,
+    "number": 45368,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: leak in tokenizer registry for `test_processors`",
-    "updated_at": "2026-04-09T10:12:46Z"
+    "title": "fix(serving): resolve rust tokenizer from ProcessorMixin in streaming generation",
+    "updated_at": "2026-04-13T15:01:51Z"
   },
   {
-    "additions": 24,
-    "author": "mohdfaour03",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #45081 ## Problem Loading a Mistral tokenizer with `fix_mistral_regex=True` crashes because `_patch_mistral_regex` receives a raw `tokenizers.Tokenizer` but tries to access `.backend_tokenizer.pre_tokenizer` on it \u2014 that attribute on\u2026",
+    "additions": 15,
+    "author": "jackcook",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Update to the latest Four Over Six API by adding options to specify the data type of activations, weights, and gradients individually cc @SunMarc",
     "changed_files": 2,
-    "cluster_id": "cluster-45081-3",
-    "cluster_ids": [
-      "cluster-45081-3"
-    ],
-    "cluster_role": "canonical",
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45317",
-    "created_at": "2026-04-08T13:38:46Z",
-    "deletions": 3,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45317/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45317",
-    "labels": [],
-    "merged": false,
-    "number": 45317,
-    "review_comments_count": 1,
-    "state": "open",
-    "title": "Fix AttributeError in _patch_mistral_regex when fix_mistral_regex=True  ",
-    "updated_at": "2026-04-09T13:52:30Z"
-  },
-  {
-    "additions": 9,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per title and seems like there are no objections. Also added some colors in verbose logging cc @tarekziade @tomaarsen @yonigozlan if you have better ideas to style this (just tagging since you reacted \u2795 ) This is\u2026",
-    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45316",
-    "created_at": "2026-04-08T13:01:15Z",
-    "deletions": 2,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45367",
+    "created_at": "2026-04-11T01:47:51Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45316/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45316",
+    "files_url": "https://github.com/huggingface/transformers/pull/45367/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45367",
     "labels": [],
-    "merged": true,
-    "number": 45316,
-    "review_comments_count": 1,
-    "state": "closed",
-    "title": "Logger has `[transformers]` prefix in non-verbose mode",
-    "updated_at": "2026-04-14T14:08:04Z"
+    "merged": false,
+    "number": 45367,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Add dtype config options for Four Over Six",
+    "updated_at": "2026-04-13T13:29:07Z"
   },
   {
-    "additions": 46,
-    "author": "Rocketknight1",
-    "author_association": "MEMBER",
-    "body_excerpt": "Reusing a variable name meant that we returned a softmaxed value instead of the original logits in some MoE routers. This generally did not affect inference, but could affect the auxiliary loss on MoE logits in training when the coefficien\u2026",
-    "changed_files": 15,
+    "additions": 64,
+    "author": "owwll",
+    "author_association": "NONE",
+    "body_excerpt": "This PR addresses two separate issues: 1. **Fixes a bug in `Mistral4` RoPE dimension calculation.** The `Mistral4RotaryEmbedding` was incorrectly using the full `head_dim` to calculate the rotary dimension, instead of respecting the `parti\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45315",
-    "created_at": "2026-04-08T12:54:52Z",
-    "deletions": 30,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45366",
+    "created_at": "2026-04-10T21:13:32Z",
+    "deletions": 11,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45315/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45315",
+    "files_url": "https://github.com/huggingface/transformers/pull/45366/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45366",
     "labels": [],
     "merged": false,
-    "number": 45315,
+    "number": 45366,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix softmaxing router logits",
-    "updated_at": "2026-04-10T13:25:20Z"
+    "title": "Fix OLMoE routing and Mistral4 RoPE dimensions",
+    "updated_at": "2026-04-10T21:40:40Z"
   },
   {
-    "additions": 18,
-    "author": "zucchini-nlp",
+    "additions": 46,
+    "author": "burtenshaw",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? fixes https://github.com/huggingface/transformers/issues/45216 and https://github.com/huggingface/transformers/issues/45310 and https://github.com/huggingface/transformers/issues/45313 TBH load-save-load works for t\u2026",
-    "changed_files": 10,
+    "body_excerpt": "Applied the overlapping GPT-J refactor from the staged PRs: added `_can_record_outputs`, moved `GPTJModel.forward` to decorator-based output capture, switched wrapper forwards to `@can_return_tuple`, and removed manual hidden-state/attenti\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45314",
-    "created_at": "2026-04-08T11:54:53Z",
-    "deletions": 27,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45365",
+    "created_at": "2026-04-10T19:37:53Z",
+    "deletions": 101,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45314/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45314",
+    "files_url": "https://github.com/huggingface/transformers/pull/45365/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45365",
     "labels": [],
     "merged": false,
-    "number": 45314,
+    "number": 45365,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Conversion for LLM class loading with VLM ckpt ",
-    "updated_at": "2026-04-10T09:18:26Z"
+    "title": "Refactor GPT-J output tracing to use standardized decorators",
+    "updated_at": "2026-04-10T19:53:25Z"
   },
   {
-    "additions": 61,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. It was confirmed that the weight matrices of shared layers are NEVER used, and that kv states should ALWAYS be shared, even during training or inference without Cache. I will fully remove them on a\u2026",
-    "changed_files": 3,
+    "additions": 815,
+    "author": "caiovicentino",
+    "author_association": "NONE",
+    "body_excerpt": "### Summary Adds a third backend to `QuantizedCache`: `polarquant`. Joins the existing `quanto` and `hqq` options and implements a Walsh-Hadamard rotation plus Lloyd-Max scalar quantization scheme tuned for KV cache compression. Pure PyTor\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45312",
-    "created_at": "2026-04-08T11:33:33Z",
-    "deletions": 24,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45312/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45312",
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45364",
+    "created_at": "2026-04-10T19:02:16Z",
+    "deletions": 2,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45364/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45364",
     "labels": [],
-    "merged": true,
-    "number": 45312,
+    "merged": false,
+    "number": 45364,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[gemma4] Dissociate kv states sharing from the Cache",
-    "updated_at": "2026-04-09T08:08:07Z"
+    "title": "Add PolarQuant backend to QuantizedCache (Hadamard-rotated Lloyd-Max)",
+    "updated_at": "2026-04-13T13:31:36Z"
   },
   {
-    "additions": 2,
-    "author": "KoichiYasuoka",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #45292 (seems to come from #41580) ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by\u2026",
-    "changed_files": 1,
+    "additions": 301,
+    "author": "michaelbenayoun",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR adds support for fusing multiple modules into a single kernel \u2014 the motivating case being fused RMSNorm+MLP kernels, but the API is generic. ## What changed - `FusedModuleBase`, `fuse_modules`, `unfuse_modul\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45311",
-    "created_at": "2026-04-08T10:38:34Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45311/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45311",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45363",
+    "created_at": "2026-04-10T18:53:22Z",
+    "deletions": 5,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45363/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45363",
     "labels": [],
     "merged": false,
-    "number": 45311,
-    "review_comments_count": 0,
+    "number": 45363,
+    "review_comments_count": 3,
     "state": "open",
-    "title": "resize_token_embeddings does not effect to output_embeddings",
-    "updated_at": "2026-04-13T13:55:03Z"
+    "title": "n-to-1 kernel fusion via `KernelConfig`",
+    "updated_at": "2026-04-14T13:14:57Z"
   },
   {
-    "additions": 301,
-    "author": "agentspan",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #45290. `ProcessorMixin.apply_chat_template` and several related code paths assumed every message in a conversation has a `content` key. Assistant messages with `tool_calls` and no textual content (a valid shape per the Op\u2026",
-    "changed_files": 9,
+    "additions": 98,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/trl/issues/5497, also fixes https://github.com/huggingface/transformers/issues/45390 TL;DR; the base model prefix is never appended if it is part of a bigger VLM, which was true\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45309",
-    "created_at": "2026-04-08T08:40:08Z",
-    "deletions": 23,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45361",
+    "created_at": "2026-04-10T13:49:28Z",
+    "deletions": 13,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45309/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45309",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/45361/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45361",
+    "labels": [],
     "merged": false,
-    "number": 45309,
+    "number": 45361,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix KeyError in apply_chat_template when message has no content (#45290)",
-    "updated_at": "2026-04-08T11:30:37Z"
+    "state": "open",
+    "title": "Add CLIP-like models in conversion to VLMs",
+    "updated_at": "2026-04-13T12:16:59Z"
   },
   {
-    "additions": 10,
-    "author": "juliabush",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes #29942 Flash Attention 2 inference equivalence tests for Whisper can fail due to higher numerical variance compared to the eager attention implementation. This PR increases the tolerance (`atol`, `rtol`) spec\u2026",
-    "changed_files": 1,
+    "additions": 3,
+    "author": "hanouticelina",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? `huggingface-cli` is deprecated and no longer maintained. This PR updates the remaining references with `hf`",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45303",
-    "created_at": "2026-04-07T21:37:00Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45360",
+    "created_at": "2026-04-10T11:59:37Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45303/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45303",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/45360/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45360",
+    "labels": [],
     "merged": false,
-    "number": 45303,
+    "number": 45360,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix FA2 inference equivalence failures for Whisper (closes #29942)",
-    "updated_at": "2026-04-08T14:42:36Z"
+    "state": "open",
+    "title": "Replace deprecated `huggingface-cli` references with `hf`",
+    "updated_at": "2026-04-10T12:13:18Z"
   },
   {
-    "additions": 7,
-    "author": "jagwar",
+    "additions": 3,
+    "author": "ArthurZucker",
     "author_association": "MEMBER",
-    "body_excerpt": "## Security Fix Fixes a trust check bypass in `trl-ci-bot.yml` that allowed any GitHub user to trigger TRL CI on self-hosted GPU runners by commenting `/trl-ci` on any PR. ### The bug The \"Ignore untrusted commenter\" step used `exit 0`, wh\u2026",
-    "changed_files": 1,
+    "body_excerpt": "Fixes #45356 ## Summary - Remove `kimi_k25` from `MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS`: its remote `TikTokenTokenizer` is the only correct backend \u2014 the model has no `tokenizer.json`, and its `added_tokens_decoder` has non-sequential\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45302",
-    "created_at": "2026-04-07T21:35:38Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45359",
+    "created_at": "2026-04-10T10:42:32Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45302/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45302",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45359/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45359",
+    "labels": [
+      "for patch"
+    ],
     "merged": true,
-    "number": 45302,
+    "number": 45359,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(security): prevent untrusted users from triggering TRL CI dispatch",
-    "updated_at": "2026-04-07T21:59:38Z"
+    "title": "Fix Kimi-K2.5 tokenizer regression and _patch_mistral_regex AttributeError",
+    "updated_at": "2026-04-13T15:16:26Z"
   },
   {
-    "additions": 0,
-    "author": "sahildando",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 0,
+    "additions": 111,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/transformers/issues/45357 finally. This was not catched in the previous fix, as the model can be reloaded correctly by `from_pretrained`, but keys are still wrongly serialized! Aft\u2026",
+    "changed_files": 24,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45301",
-    "created_at": "2026-04-07T21:12:29Z",
-    "deletions": 0,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45358",
+    "created_at": "2026-04-10T10:19:42Z",
+    "deletions": 42,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45301/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45301",
-    "labels": [],
-    "merged": false,
-    "number": 45301,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/45358/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45358",
+    "labels": [
+      "for patch"
+    ],
+    "merged": true,
+    "number": 45358,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": " docs maintenance for transformers repository 979e8",
-    "updated_at": "2026-04-09T13:28:27Z"
+    "title": "Fix vlm weight mappings",
+    "updated_at": "2026-04-10T15:41:47Z"
   },
   {
-    "additions": 136,
-    "author": "w4nderlust",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Nemotron-H models use standalone MLP layers in their `hybrid_override_pattern` (the `-` character), but the config parser, validators, and modeling code only know about `mamba`/`attention`/`moe`. This means every Ne\u2026",
-    "changed_files": 5,
+    "additions": 2297,
+    "author": "Shikhar-S",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR introduces [PhoneticXeus](https://arxiv.org/abs/2603.29042), which is the state-of-the-art universal phone recognizer trained on 70+ languages and evaluated on ~100 languages. The model should have high util\u2026",
+    "changed_files": 16,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45300",
-    "created_at": "2026-04-07T20:57:45Z",
-    "deletions": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45355",
+    "created_at": "2026-04-10T04:26:05Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45300/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45300",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45355/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45355",
+    "labels": [
+      "New model",
+      "Audio"
+    ],
     "merged": false,
-    "number": 45300,
+    "number": 45355,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Fix Nemotron-H: add mlp layer type support",
-    "updated_at": "2026-04-09T16:16:30Z"
+    "title": "Add universal phone recognition model - PhoneticXeus",
+    "updated_at": "2026-04-14T15:02:02Z"
   },
   {
-    "additions": 3,
-    "author": "partacc",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Some CI security tests... I will contact you privately if there is anything worth reporting. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear i\u2026",
-    "changed_files": 1,
+    "additions": 8,
+    "author": "winglian",
+    "author_association": "COLLABORATOR",
+    "body_excerpt": "# What does this PR do? Gemma 4 was calculating the CE loss incorrectly and not handling gradient accumulation steps properly, leading to losses scaled up by the value of the gradient accumulation steps rather than letting the built in HF\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45299",
-    "created_at": "2026-04-07T20:07:02Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45354",
+    "created_at": "2026-04-10T02:54:56Z",
+    "deletions": 76,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45299/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45299",
+    "files_url": "https://github.com/huggingface/transformers/pull/45354/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45354",
     "labels": [],
-    "merged": false,
-    "number": 45299,
+    "merged": true,
+    "number": 45354,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Please ignore] CI Test PR",
-    "updated_at": "2026-04-07T20:23:28Z"
+    "title": "fix gemma4 gradient accumulation loss and last token incorrect labels",
+    "updated_at": "2026-04-10T10:08:12Z"
   },
   {
-    "additions": 2958,
-    "author": "DatLe203",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 13,
+    "additions": 30,
+    "author": "wilnn",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Add `**kwargs` to all methods in the `CallbackHandler` class. Previously, only the `CallbackHandler.on_push_begin()` method accepted `**kwargs`, while all other methods did not. This forces users who want to customi\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45298",
-    "created_at": "2026-04-07T19:39:50Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45353",
+    "created_at": "2026-04-09T23:14:20Z",
+    "deletions": 30,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45298/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45298",
+    "files_url": "https://github.com/huggingface/transformers/pull/45353/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45353",
     "labels": [],
-    "merged": false,
-    "number": 45298,
+    "merged": true,
+    "number": 45353,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add new qwen2 5 vl",
-    "updated_at": "2026-04-07T20:00:43Z"
+    "title": "add kwargs to all methods in the CallbackHandler class",
+    "updated_at": "2026-04-10T14:16:46Z"
   },
   {
-    "additions": 21,
-    "author": "EhteshamSid",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes mutable default argument bugs in four quantization config `__init__` methods inside `quantization_config.py`. In Python, mutable objects used as default argument values (e.g. `=[]`, `={}`) are created once at\u2026",
-    "changed_files": 1,
+    "additions": 4,
+    "author": "RudrenduPaul",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Corrects an incorrect return type annotation on `Qwen3MoeSparseMoeBlock.forward`. The method is annotated as returning `tuple[torch.Tensor, torch.Tensor]` but actually returns a single reshaped `torch.Tensor` (see\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45297",
-    "created_at": "2026-04-07T18:40:05Z",
-    "deletions": 7,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45352",
+    "created_at": "2026-04-09T21:53:31Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45297/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45297",
+    "files_url": "https://github.com/huggingface/transformers/pull/45352/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45352",
     "labels": [],
-    "merged": false,
-    "number": 45297,
+    "merged": true,
+    "number": 45352,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix mutable default arguments in quantization config classes",
-    "updated_at": "2026-04-09T14:00:25Z"
+    "title": "fix(qwen3_moe): correct return type annotation on Qwen3MoeSparseMoeBlock.forward",
+    "updated_at": "2026-04-13T14:07:30Z"
   },
   {
-    "additions": 191,
-    "author": "UsamaKenway",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 3,
+    "additions": 15,
+    "author": "RudrenduPaul",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes a crash in `get_device_properties()` in `testing_utils.py` when CUDA is installed on the system but no GPU device is present (e.g., a CPU-only cloud studio with CUDA libraries installed). The function called\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45296",
-    "created_at": "2026-04-07T18:39:33Z",
-    "deletions": 0,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45351",
+    "created_at": "2026-04-09T21:51:42Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45296/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45296",
+    "files_url": "https://github.com/huggingface/transformers/pull/45351/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45351",
     "labels": [],
     "merged": false,
-    "number": 45296,
-    "review_comments_count": 2,
+    "number": 45351,
+    "review_comments_count": 1,
     "state": "open",
-    "title": "Add GGUF support to Gemma4 (31B & 26B-A4B) text ",
-    "updated_at": "2026-04-12T10:53:54Z"
+    "title": "fix(testing_utils): guard get_device_capability with torch.cuda.is_available()",
+    "updated_at": "2026-04-16T09:20:10Z"
   },
   {
-    "additions": 116,
-    "author": "jesperschlegel",
+    "additions": 1090,
+    "author": "mrutkows",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Adds the Gemma4ForSequenceClassification class which supports finetuning Gemma 4 models on sequence classification tasks.",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? Support new Granite 4 vision arch. <!-- Remove if not applicable --> Fixes # (issue) ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by c\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45294",
-    "created_at": "2026-04-07T17:51:38Z",
-    "deletions": 4,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45294/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45294",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45350",
+    "created_at": "2026-04-09T17:46:37Z",
+    "deletions": 0,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45350/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45350",
     "labels": [],
     "merged": false,
-    "number": 45294,
+    "number": 45350,
     "review_comments_count": 0,
     "state": "open",
-    "title": "feat: add Gemma4ForSequenceClassification",
-    "updated_at": "2026-04-13T13:05:41Z"
+    "title": "WIP: Add support for Granite4VisionForConditionalGeneration",
+    "updated_at": "2026-04-10T12:34:50Z"
   },
   {
-    "additions": 4,
-    "author": "yonigozlan",
-    "author_association": "MEMBER",
-    "body_excerpt": "## Fix global state leak in `AutoTokenizer.register` causing test failures ### Problem `test_from_pretrained_dynamic_processor` was failing when run as part of the full test class with: ``` AttributeError: NewTokenizer has no attribute spe\u2026",
+    "additions": 90,
+    "author": "florian6973",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #45305 Add a regression test in `TrainerGradientAccumulationTest` to avoid passing the GAS value to Accelerate by mistake Description: I force the value of the `num_steps` parameter to be 1, and the regression\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45293",
-    "created_at": "2026-04-07T16:29:25Z",
-    "deletions": 16,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45349",
+    "created_at": "2026-04-09T17:24:39Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45293/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45293",
-    "labels": [],
-    "merged": false,
-    "number": 45293,
-    "review_comments_count": 1,
-    "state": "open",
-    "title": "Fix \"AttributeError: NewTokenizer has no attribute special_attribute_present\" (Remove `REGISTERED_FAST_ALIASES`)",
-    "updated_at": "2026-04-09T12:04:21Z"
+    "files_url": "https://github.com/huggingface/transformers/pull/45349/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45349",
+    "labels": [
+      "for patch"
+    ],
+    "merged": true,
+    "number": 45349,
+    "review_comments_count": 6,
+    "state": "closed",
+    "title": "Fix #45305 + add regression test GAS",
+    "updated_at": "2026-04-13T14:41:43Z"
   },
   {
-    "additions": 2958,
-    "author": "DatLe203",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 11,
+    "additions": 50,
+    "author": "qgallouedec",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes #45290 ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by our ability to review and r\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45291",
-    "created_at": "2026-04-07T14:30:38Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45348",
+    "created_at": "2026-04-09T15:59:07Z",
+    "deletions": 19,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45291/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45291",
+    "files_url": "https://github.com/huggingface/transformers/pull/45348/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45348",
     "labels": [],
-    "merged": false,
-    "number": 45291,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 45348,
+    "review_comments_count": 7,
     "state": "closed",
-    "title": "First pull request",
-    "updated_at": "2026-04-07T14:53:52Z"
+    "title": "Fix `apply_chat_template` crash on `tool_call` messages without content",
+    "updated_at": "2026-04-13T19:44:38Z"
   },
   {
-    "additions": 24,
-    "author": "zucchini-nlp",
+    "additions": 35,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45265 and doesn't warn if the value is an int.",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? As per the title. `accelerate` destroys the dict otherwise, if it's not BOTH passed as kwarg AND part of `_skip_keys_device_placement`.......... `per_layer_input` needs to stay as a positional arg, for gradient chec\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45289",
-    "created_at": "2026-04-07T13:00:52Z",
-    "deletions": 73,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45347",
+    "created_at": "2026-04-09T15:31:34Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45289/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45289",
+    "files_url": "https://github.com/huggingface/transformers/pull/45347/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45347",
     "labels": [],
     "merged": true,
-    "number": 45289,
+    "number": 45347,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Less unnecessary RoPE warnings",
-    "updated_at": "2026-04-13T13:14:41Z"
+    "title": "[gemma4] Fix device map auto",
+    "updated_at": "2026-04-09T15:45:15Z"
   },
   {
-    "additions": 35,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
+    "additions": 16,
+    "author": "ionut-anghelina",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
     "body_excerpt": null,
-    "changed_files": 1,
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45288",
-    "created_at": "2026-04-07T10:24:53Z",
-    "deletions": 11,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45288/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45288",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45346",
+    "created_at": "2026-04-09T14:48:28Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45346/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45346",
     "labels": [],
     "merged": false,
-    "number": 45288,
+    "number": 45346,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "Fix Double Application of Softmax for Router Logits in MoE models",
+    "updated_at": "2026-04-13T12:40:28Z"
+  },
+  {
+    "additions": 30,
+    "author": "ansley",
+    "author_association": "NONE",
+    "body_excerpt": "The `transformers` V5 \"rm slow tokenizers\" refactor (\\#40936) aliased `LlamaTokenizerFast` to `LlamaTokenizer`, whose `__init__` unconditionally installs a SentencePiece Metaspace pre-tokenizer. This is correct for classic Llama/Llama-2 mo\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45345",
+    "created_at": "2026-04-09T14:31:40Z",
+    "deletions": 14,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45345/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45345",
+    "labels": [],
+    "merged": false,
+    "number": 45345,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(cohere_asr): auto-fix failing tests",
-    "updated_at": "2026-04-10T10:41:27Z"
+    "title": "Fix ByteLevel-BPE tokenizers silently breaking in `LlamaTokenizer`",
+    "updated_at": "2026-04-10T12:45:24Z"
   },
   {
-    "additions": 102,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": null,
+    "additions": 6,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Simple hook to display test duration. This will append inline duration per test during the run, example: ``` tests/utils/test_configuration_utils.py::ConfigPushToHubTester::test_push_to_hub [gw1] [ 90%] PASSED tests\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45287",
-    "created_at": "2026-04-07T10:23:45Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45344",
+    "created_at": "2026-04-09T14:22:46Z",
     "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45287/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45287",
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45344/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45344",
     "labels": [],
-    "merged": false,
-    "number": 45287,
+    "merged": true,
+    "number": 45344,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(videomt): auto-fix failing tests",
-    "updated_at": "2026-04-10T10:41:30Z"
+    "title": "refactor: display test duration",
+    "updated_at": "2026-04-09T15:19:26Z"
   },
   {
-    "additions": 29,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
+    "additions": 8,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
     "body_excerpt": null,
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45286",
-    "created_at": "2026-04-07T10:22:51Z",
-    "deletions": 1,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45286/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45286",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45342",
+    "created_at": "2026-04-09T14:13:15Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45342/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45342",
     "labels": [],
     "merged": false,
-    "number": 45286,
+    "number": 45342,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(nomic_bert): auto-fix failing tests",
-    "updated_at": "2026-04-10T10:41:28Z"
+    "state": "open",
+    "title": "Use `_keys_to_ignore_on_load_unexpected/missing` recursively from children",
+    "updated_at": "2026-04-09T14:23:31Z"
   },
   {
-    "additions": 153,
+    "additions": 17,
     "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Supersedes https://github.com/huggingface/transformers/pull/45314 with a better fix. Fixes https://github.com/huggingface/transformers/issues/45216 and https://github.com/huggingface/transformers/issues/45310 and ht\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45285",
-    "created_at": "2026-04-07T09:18:11Z",
-    "deletions": 329,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45340",
+    "created_at": "2026-04-09T12:02:14Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45285/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45285",
+    "files_url": "https://github.com/huggingface/transformers/pull/45340/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45340",
     "labels": [],
     "merged": true,
-    "number": 45285,
-    "review_comments_count": 6,
-    "state": "closed",
-    "title": "Fix export for gemma4 and add Integration tests",
-    "updated_at": "2026-04-08T13:07:40Z"
+    "number": 45340,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix conversion mappings for vlms",
+    "updated_at": "2026-04-10T10:18:09Z"
   },
   {
-    "additions": 4,
-    "author": "Abdennacer-Badaoui",
+    "additions": 156,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "Following this PR : https://github.com/huggingface/transformers/pull/45268. This PR fix Qwen2 expectations for AMD.",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? The CircleCI config file is not ruff formatted, leading to unwanted changes when it's opened in an editor that follows our repository ruff configuration. This patch adds it and runs `make style` to update it",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45284",
-    "created_at": "2026-04-07T09:12:01Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45339",
+    "created_at": "2026-04-09T09:44:16Z",
+    "deletions": 58,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45284/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45284",
+    "files_url": "https://github.com/huggingface/transformers/pull/45339/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45339",
     "labels": [],
     "merged": true,
-    "number": 45284,
+    "number": 45339,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[AMD CI] Fix Qwen2 expectations",
-    "updated_at": "2026-04-07T12:01:58Z"
+    "title": "chore: added circleci python script to ruff and ty checkers",
+    "updated_at": "2026-04-09T12:00:08Z"
   },
   {
-    "additions": 107,
-    "author": "jIab-b",
+    "additions": 37,
+    "author": "RudrenduPaul",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 2,
+    "body_excerpt": "Closes #45162 ## What this PR does Expands the docstrings of `_can_set_attn_implementation` and `_can_set_experts_implementation` in `modeling_utils.py` to explicitly document the known limitations of their source-inspection heuristic. **C\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45283",
-    "created_at": "2026-04-07T09:11:14Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45338",
+    "created_at": "2026-04-09T09:35:52Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45283/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45283",
+    "files_url": "https://github.com/huggingface/transformers/pull/45338/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45338",
     "labels": [],
     "merged": false,
-    "number": 45283,
+    "number": 45338,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add Qwen3.5 GGUF loading support",
-    "updated_at": "2026-04-09T14:49:22Z"
+    "state": "closed",
+    "title": "docs: document known limitations of _can_set_attn/experts_implementation source inspection",
+    "updated_at": "2026-04-09T13:43:04Z"
   },
   {
-    "additions": 30,
-    "author": "Abdennacer-Badaoui",
+    "additions": 13,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "`_register_model_output_pytree_node` was calling set.__contains__ during TorchDynamo tracing, which is unsupported in PyTorch 2.8.0 (ROCm). Added an early return when `torch.compiler.is_compiling()` is True, since pytree nodes are already\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Removing test_hub from CI for now",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45282",
-    "created_at": "2026-04-07T08:50:54Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45337",
+    "created_at": "2026-04-09T08:54:45Z",
+    "deletions": 30,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45282/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45282",
+    "files_url": "https://github.com/huggingface/transformers/pull/45337/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45337",
     "labels": [],
     "merged": true,
-    "number": 45282,
-    "review_comments_count": 3,
+    "number": 45337,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[AMD CI] Fix torch.compile/export failures on AMD CI due to untraceable set.__contains__ ",
-    "updated_at": "2026-04-13T15:59:55Z"
+    "title": "chore: remove test_hub for now",
+    "updated_at": "2026-04-09T09:28:52Z"
   },
   {
-    "additions": 25,
-    "author": "zhang-prog",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### Description This PR fixes a boundary issue in the `_extract_polygon_points_by_masks` method of PP-DocLayoutV3. When running inference with a low confidence threshold, or due to coordinate clipping during scaling, the extracted `cropped\u2026",
+    "additions": 84,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title. Follow-up of https://github.com/huggingface/transformers/pull/45312. This removes the unnecessary weights, and silently skip them during loading, so that the checkpoints on the hub do not have to b\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45281",
-    "created_at": "2026-04-07T08:49:24Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45336",
+    "created_at": "2026-04-09T08:43:55Z",
+    "deletions": 26,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45281/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45281",
+    "files_url": "https://github.com/huggingface/transformers/pull/45336/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45336",
     "labels": [],
     "merged": true,
-    "number": 45281,
-    "review_comments_count": 0,
+    "number": 45336,
+    "review_comments_count": 7,
     "state": "closed",
-    "title": "Fix resize failure caused by zero-sized masks in PP-DocLayoutV3",
-    "updated_at": "2026-04-09T13:06:44Z"
+    "title": "[gemma4] Remove all shared weights, and silently skip them during loading",
+    "updated_at": "2026-04-09T13:23:33Z"
   },
   {
-    "additions": 2502,
-    "author": "marvinzh",
+    "additions": 1333,
+    "author": "kmswin1",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? add Qianfan-OCR model definition - `QianfanOCRForConditionalGeneration` - image-text to text model definition - `QianfanOCRModel` - backbone of image-text to text model without lm heads - `QianfanOCRProcessor` - tex\u2026",
-    "changed_files": 18,
+    "body_excerpt": "Add A.X K1 model architecture What does this PR do? This PR adds support for A.X K1, a large-scale Mixture-of-Experts (MoE) language model developed by [SK Telecom](https://huggingface.co/skt). A.X K1 contains 519B total parameters with 33\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45280",
-    "created_at": "2026-04-07T06:49:34Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45334",
+    "created_at": "2026-04-09T06:21:43Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45280/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45280",
+    "files_url": "https://github.com/huggingface/transformers/pull/45334/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45334",
     "labels": [],
     "merged": false,
-    "number": 45280,
-    "review_comments_count": 69,
-    "state": "open",
-    "title": "add Qianfan-OCR model definition",
-    "updated_at": "2026-04-15T11:30:07Z"
+    "number": 45334,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Feature/add axk1",
+    "updated_at": "2026-04-14T07:23:08Z"
   },
   {
-    "additions": 43,
-    "author": "sywangyi",
+    "additions": 471,
+    "author": "eladsegal",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Adds heterogeneous model support - the ability for individual layers to differ from the global config (e.g., different `intermediate_size`, `num_key_value_heads`) and to skip sub-modules entirely (MLP, attention, et\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45279",
-    "created_at": "2026-04-07T06:40:35Z",
-    "deletions": 6,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45333",
+    "created_at": "2026-04-09T06:18:11Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45279/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45279",
+    "files_url": "https://github.com/huggingface/transformers/pull/45333/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45333",
     "labels": [],
     "merged": false,
-    "number": 45279,
-    "review_comments_count": 4,
+    "number": 45333,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "add expert parallelism for gemma-4-26B-A4B-it",
-    "updated_at": "2026-04-14T13:32:10Z"
+    "title": "Add heterogeneous config support (per-layer configuration)",
+    "updated_at": "2026-04-14T14:08:07Z"
   },
   {
-    "additions": 4,
-    "author": "kamalrajkannan78",
+    "additions": 2152,
+    "author": "eladsegal",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? - Gemma3ForConditionalGeneration.forward & Gemma3ForSequenceClassification.forward calls self.model() without return_dict=True, so @can_return_tuple silently converts the output to a plain tuple, causing outputs.pas\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45277",
-    "created_at": "2026-04-07T05:24:43Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45277/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45277",
-    "labels": [],
-    "merged": true,
-    "number": 45277,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix AttributeError in Gemma3ForConditionalGeneration and Gemma3ForSequenceClassification when config.return_dict=False",
-    "updated_at": "2026-04-10T10:37:16Z"
-  },
-  {
-    "additions": 9,
-    "author": "avarga1",
-    "author_association": "NONE",
-    "body_excerpt": "## Problem `AutoConfig.from_pretrained(\"baidu/ERNIE-4.5-VL-28B-A3B-Paddle\", trust_remote_code=True)` raises errors that prevent the model from loading at all. Three separate bugs compound each other: ### Bug 1 \u2014 `model_type` mismatch (KeyE\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Adds heterogeneous model support - the ability for individual layers to differ from the global config (e.g., different `intermediate_size`, `num_key_value_heads`) and to skip sub-modules entirely (MLP, attention, et\u2026",
+    "changed_files": 14,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45275",
-    "created_at": "2026-04-07T02:22:26Z",
-    "deletions": 5,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45332",
+    "created_at": "2026-04-09T05:56:31Z",
+    "deletions": 40,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45275/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45275",
+    "files_url": "https://github.com/huggingface/transformers/pull/45332/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45332",
     "labels": [],
     "merged": false,
-    "number": 45275,
-    "review_comments_count": 6,
-    "state": "closed",
-    "title": "fix(ernie4_5_vl_moe): resolve three config loading failures for ERNIE-4.5-VL MoE models",
-    "updated_at": "2026-04-09T16:57:22Z"
+    "number": 45332,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Add heterogeneous model support (per-layer config and modeling)",
+    "updated_at": "2026-04-15T04:50:09Z"
   },
   {
-    "additions": 215,
-    "author": "Qubitium",
+    "additions": 12,
+    "author": "Kash6",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? 1. Fix: CUDA graph reuse for FA2 continuous batching was wrongly keyed causing quality collapse for specific configuration CUDA graph reuse used the wrong key: replay reuse depended on padded tensor sizes, but FA va\u2026",
-    "changed_files": 5,
+    "body_excerpt": "get_rope_index unconditionally applies tokens_per_second temporal scaling to both images and videos. For still images (modality_type == 1), this shifts the temporal position origin to start_position * tokens_per_second instead of start_pos\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45274",
-    "created_at": "2026-04-07T01:43:12Z",
-    "deletions": 36,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45330",
+    "created_at": "2026-04-08T23:51:52Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45274/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45274",
-    "labels": [],
-    "merged": false,
-    "number": 45274,
+    "files_url": "https://github.com/huggingface/transformers/pull/45330/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45330",
+    "labels": [
+      "for patch"
+    ],
+    "merged": true,
+    "number": 45330,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix CB Accuracy Regression under FA2",
-    "updated_at": "2026-04-09T03:47:47Z"
+    "state": "closed",
+    "title": "Fix Qwen2.5-VL temporal RoPE scaling applied to still images",
+    "updated_at": "2026-04-14T03:21:32Z"
   },
   {
-    "additions": 2,
-    "author": "excepshenal",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Liger kernel unnecessarily materializes logits in VRAM during eval with `prediction_loss_only=True`, causing OOM. We explicitly tell Liger to `skip_logits`. <!-- Congratulations! You've made it this far! You're not\u2026",
-    "changed_files": 1,
+    "additions": 152,
+    "author": "abidlabs",
+    "author_association": "MEMBER",
+    "body_excerpt": "Updates `TrackioCallback` and `TrainingArguments` for the latest version of Trackio using HF Buckets as the backend, and control over creating a static Space for the Trackio dashboard during or at the end of training. These are now the `Tr\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45273",
-    "created_at": "2026-04-06T21:11:21Z",
-    "deletions": 0,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45329",
+    "created_at": "2026-04-08T22:36:08Z",
+    "deletions": 57,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45273/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45273",
+    "files_url": "https://github.com/huggingface/transformers/pull/45329/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45329",
     "labels": [],
-    "merged": false,
-    "number": 45273,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "fix: liger unnecessarily materializes logits in VRAM during eval, causing OOM",
-    "updated_at": "2026-04-09T15:32:29Z"
+    "merged": true,
+    "number": 45329,
+    "review_comments_count": 21,
+    "state": "closed",
+    "title": "Update `trackio` integration to use Buckets and \"freeze\" Space after training",
+    "updated_at": "2026-04-13T14:30:27Z"
   },
   {
-    "additions": 6,
-    "author": "yonigozlan",
-    "author_association": "MEMBER",
-    "body_excerpt": "Cc @ydshieh",
+    "additions": 9,
+    "author": "RyanMullins",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #45242 * Drops `k_proj`, `k_norm`, and `v_proj` weights for `Gemma4TextAttention` modules from the checkpoint if the layer shares KV cache values. These changes can also be adapted to Gemma 3n if that's desira\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45272",
-    "created_at": "2026-04-06T20:39:29Z",
-    "deletions": 20,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45328",
+    "created_at": "2026-04-08T20:43:42Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45272/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45272",
+    "files_url": "https://github.com/huggingface/transformers/pull/45328/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45328",
     "labels": [],
     "merged": false,
-    "number": 45272,
+    "number": 45328,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Fix redundant logic in video processing SmolVLM",
-    "updated_at": "2026-04-06T20:49:13Z"
+    "title": "Drop unused Gemma4TextAttention weights when sharing KV Cache",
+    "updated_at": "2026-04-09T18:31:13Z"
   },
   {
-    "additions": 94,
+    "additions": 337,
     "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "adds a separate vlm contribution doc for more visibility instead of being hidden in the Contribute to Transformers doc, and integration tests are covered in #45152",
+    "body_excerpt": "refactors the how to add a model with modular transformers doc: - structure: - flipped the order so you learn how to write the modular file first before generating it - remove the motivator examples with BERT/RoBERTa - merge the two `super\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45271",
-    "created_at": "2026-04-06T18:56:42Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45271/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45271",
-    "labels": [],
-    "merged": true,
-    "number": 45271,
-    "review_comments_count": 8,
-    "state": "closed",
-    "title": "[docs] vlm addition",
-    "updated_at": "2026-04-15T17:02:43Z"
-  },
-  {
-    "additions": 162,
-    "author": "madhav1k",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Introduce logging of individual loss components when models return a dict of losses. - Add TrainingArguments.logging_loss_components flag to enable/disable this behavior. - Track per-component running sums with _tr_loss_components and aggr\u2026",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45270",
-    "created_at": "2026-04-06T18:38:51Z",
-    "deletions": 7,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45327",
+    "created_at": "2026-04-08T20:23:28Z",
+    "deletions": 408,
     "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45270/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45270",
+    "files_url": "https://github.com/huggingface/transformers/pull/45327/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45327",
     "labels": [],
     "merged": false,
-    "number": 45270,
-    "review_comments_count": 0,
+    "number": 45327,
+    "review_comments_count": 1,
     "state": "open",
-    "title": "[Trainer] Support multi-loss component logging",
-    "updated_at": "2026-04-13T08:30:40Z"
+    "title": "[docs] modular transformers",
+    "updated_at": "2026-04-15T18:15:34Z"
   },
   {
-    "additions": 3,
-    "author": "ryota-komatsu",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fix minor typos in `src/transformers/utils/output_capturing.py`: - `install_output_capuring_hook` \u2192 `install_output_capturing_hook` (lines 98, 147) - `Tis` \u2192 `This` (line 152) ## Before submitting - [x] This PR fixe\u2026",
-    "changed_files": 1,
+    "additions": 19,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? \u2192 This PR introduces compat fixes across several audio models to ensure they can be loaded and used by a companion vLLM PR. <ins>These changes are deliberate and are blocking</ins> [this vLLM PR](https://github.co\u2026",
+    "changed_files": 11,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45269",
-    "created_at": "2026-04-06T18:01:42Z",
-    "deletions": 3,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45326",
+    "created_at": "2026-04-08T18:28:35Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45269/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45269",
+    "files_url": "https://github.com/huggingface/transformers/pull/45326/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45326",
     "labels": [],
     "merged": false,
-    "number": 45269,
+    "number": 45326,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Fix typos in src/transformers/utils/output_capturing.py",
-    "updated_at": "2026-04-06T20:49:06Z"
+    "title": "feat[vLLM \u00d7 v5]: Add vLLM compatibility for audio models",
+    "updated_at": "2026-04-14T15:45:32Z"
   },
   {
-    "additions": 3,
-    "author": "ydshieh",
+    "additions": 236,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix `Qwen2IntegrationTest`: - `test_speculative_generation`: - `0c89522f`: #43794 changed seed, so the actual output changed, but the expected output is not updated (cc @tarekziade more attention next time \ud83d\ude04 ) - `af\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45276 and https://github.com/huggingface/transformers/issues/45335 In gemma4 per-layer inputs have to be resized as long as they aren't part of soft multimoda\u2026",
+    "changed_files": 16,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45268",
-    "created_at": "2026-04-06T18:01:38Z",
-    "deletions": 3,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45324",
+    "created_at": "2026-04-08T17:06:26Z",
+    "deletions": 53,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45268/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45268",
+    "files_url": "https://github.com/huggingface/transformers/pull/45324/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45324",
     "labels": [],
     "merged": true,
-    "number": 45268,
-    "review_comments_count": 2,
+    "number": 45324,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "Fix `Qwen2IntegrationTest`",
-    "updated_at": "2026-04-08T07:56:35Z"
+    "title": "Gemma4 resizing per layer inputs",
+    "updated_at": "2026-04-15T11:15:23Z"
   },
   {
-    "additions": 108,
-    "author": "KetanP1618",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Adds a missing docstring to the `FFN.forward` method in DistilBERT. The `FFN.forward` method in `modeling_distilbert.py` had no documentation at all. This PR adds proper Args and Returns sections following the exis\u2026",
-    "changed_files": 1,
+    "additions": 225,
+    "author": "remi-or",
+    "author_association": "MEMBER",
+    "body_excerpt": "# Summary This PR fixes the issue raised in https://github.com/huggingface/transformers/pull/45274 . CUDA graph reuse in continuous batching used (num_q_tokens, max_kv_read) as the graph cache key. However, FlashAttention varlen kernels al\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45267",
-    "created_at": "2026-04-06T17:59:13Z",
-    "deletions": 39,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45323",
+    "created_at": "2026-04-08T16:30:18Z",
+    "deletions": 126,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45267/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45267",
+    "files_url": "https://github.com/huggingface/transformers/pull/45323/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45323",
     "labels": [],
     "merged": false,
-    "number": 45267,
-    "review_comments_count": 0,
+    "number": 45323,
+    "review_comments_count": 2,
     "state": "open",
-    "title": "Add docstring to FFN.forward in DistilBERT",
-    "updated_at": "2026-04-08T13:43:57Z"
+    "title": "[CB] Fix capture of max_seqlen",
+    "updated_at": "2026-04-16T10:19:54Z"
   },
   {
-    "additions": 125,
-    "author": "KetanP1618",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Adds missing docstrings to two undocumented `forward` methods in the ALBERT model: - `AlbertMLMHead.forward` - Added Args and Returns sections - `AlbertSOPHead.forward` - Added Args and Returns sections Both method\u2026",
-    "changed_files": 1,
+    "additions": 20,
+    "author": "andrewor14",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "**Summary:** TorchAO recently deprecated AffineQuantizedTensor and related classes (pytorch/ao#2752). These will be removed in the next release. We should remove references of these classes in transformers before then. **Test Plan:** ``` p\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45266",
-    "created_at": "2026-04-06T17:41:14Z",
-    "deletions": 41,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45321",
+    "created_at": "2026-04-08T15:42:16Z",
+    "deletions": 29,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45266/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45266",
+    "files_url": "https://github.com/huggingface/transformers/pull/45321/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45321",
     "labels": [],
     "merged": false,
-    "number": 45266,
+    "number": 45321,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add docstrings to AlbertMLMHead and AlbertSOPHead forward methods",
-    "updated_at": "2026-04-08T13:43:34Z"
+    "state": "open",
+    "title": "Remove references to torchao's AffineQuantizedTensor",
+    "updated_at": "2026-04-09T12:21:03Z"
   },
   {
-    "additions": 1,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? **Fix tf32 issue: set `torch.backends.cudnn.conv.fp32_precision` explicitly.** (#45248) breaks when running on torch 2.8 or older ...",
+    "additions": 5,
+    "author": "Regata3010",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes a crash in assisted generation when using model pairs with different vocabulary sizes but the same tokenizer family (e.g., Qwen2.5-7B + Qwen2.5-0.5B). `map_input_embeddings` is only initialized when `len(self\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45263",
-    "created_at": "2026-04-06T10:06:40Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45320",
+    "created_at": "2026-04-08T15:30:16Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45263/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45263",
+    "files_url": "https://github.com/huggingface/transformers/pull/45320/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45320",
     "labels": [],
     "merged": true,
-    "number": 45263,
+    "number": 45320,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add `hasattr(torch.backends.cudnn, \"conv\")` to `conftest.py`",
-    "updated_at": "2026-04-06T19:46:18Z"
+    "title": "Fix AttributeError in AssistantToTargetTranslator.unmap_input_ids with cross-vocab models",
+    "updated_at": "2026-04-10T17:46:37Z"
   },
   {
-    "additions": 1,
-    "author": "lowzhao",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fix docstring spelling mistake TokenizersBackend.convert_to_native_format. ```python @classmethod def convert_to_native_format(cls, trust_remote_code=False, **kwargs): \"\"\"s <---- additional s ``` Likely caused by mi\u2026",
-    "changed_files": 1,
+    "additions": 78,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? - Removes `HUGGINGFACE_CO_STAGING` when downloading artifacts - adds a retry mechanism for external URLs (with partial file cleanup)",
+    "changed_files": 6,
     "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45262",
-    "created_at": "2026-04-06T08:41:40Z",
-    "deletions": 1,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45319",
+    "created_at": "2026-04-08T14:51:48Z",
+    "deletions": 32,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45262/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45262",
+    "files_url": "https://github.com/huggingface/transformers/pull/45319/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45319",
     "labels": [],
     "merged": true,
-    "number": 45262,
-    "review_comments_count": 0,
+    "number": 45319,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "doc: fix TokenizersBackend.convert_to_native_format docstring",
-    "updated_at": "2026-04-06T16:32:44Z"
+    "title": "fix: dont download artifacts from the test hub",
+    "updated_at": "2026-04-15T16:52:10Z"
   },
   {
-    "additions": 22,
-    "author": "ydshieh",
+    "additions": 5,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What this PR does It's working, see https://github.com/huggingface/transformers/actions/runs/24025915210 (the failing job in this PR is because the workflow needs to be on `main` to be effective).",
+    "body_excerpt": "# What does this PR do? AutoTokenizer.register() adds classes to the global `REGISTERED_TOKENIZER_CLASSES` dict and some tests did not clean up behind them, leading to leaky state between tests",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45261",
-    "created_at": "2026-04-06T07:52:29Z",
-    "deletions": 5,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45318",
+    "created_at": "2026-04-08T13:46:47Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45261/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45261",
+    "files_url": "https://github.com/huggingface/transformers/pull/45318/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45318",
     "labels": [],
     "merged": true,
-    "number": 45261,
+    "number": 45318,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "empty",
-    "updated_at": "2026-04-06T09:24:59Z"
+    "title": "fix: leak in tokenizer registry for `test_processors`",
+    "updated_at": "2026-04-09T10:12:46Z"
   },
   {
-    "additions": 18,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "## Summary PR #43514 refactored `_preprocess` to pass `resample=resample` to `resize`, but the `resize` method in `SmolVLMVideoProcessor` still had `interpolation` as its parameter name. The `resample` kwarg was silently swallowed by `**kw\u2026",
+    "additions": 24,
+    "author": "mohdfaour03",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #45081 ## Problem Loading a Mistral tokenizer with `fix_mistral_regex=True` crashes because `_patch_mistral_regex` receives a raw `tokenizers.Tokenizer` but tries to access `.backend_tokenizer.pre_tokenizer` on it \u2014 that attribute on\u2026",
     "changed_files": 2,
+    "cluster_id": "cluster-45081-3",
+    "cluster_ids": [
+      "cluster-45081-3"
+    ],
+    "cluster_role": "canonical",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45317",
+    "created_at": "2026-04-08T13:38:46Z",
+    "deletions": 3,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45317/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45317",
+    "labels": [],
+    "merged": false,
+    "number": 45317,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "Fix AttributeError in _patch_mistral_regex when fix_mistral_regex=True  ",
+    "updated_at": "2026-04-09T13:52:30Z"
+  },
+  {
+    "additions": 9,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per title and seems like there are no objections. Also added some colors in verbose logging cc @tarekziade @tomaarsen @yonigozlan if you have better ideas to style this (just tagging since you reacted \u2795 ) This is\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45258",
-    "created_at": "2026-04-06T05:09:09Z",
-    "deletions": 7,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45316",
+    "created_at": "2026-04-08T13:01:15Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45258/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45258",
+    "files_url": "https://github.com/huggingface/transformers/pull/45316/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45316",
     "labels": [],
     "merged": true,
-    "number": 45258,
-    "review_comments_count": 7,
+    "number": 45316,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Fix `SmolVLM` video processor `resize` using wrong interpolation after backend refactor",
-    "updated_at": "2026-04-06T20:40:22Z"
+    "title": "Logger has `[transformers]` prefix in non-verbose mode",
+    "updated_at": "2026-04-14T14:08:04Z"
   },
   {
-    "additions": 269,
-    "author": "lucianommartins",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# [Gemma4] Fix chat template and stop tokens for OpenAI tool calling compatibility ## What does this PR do? Rewrites the `_patch_template_for_openai_tool_role()` function in `convert_gemma4_weights.py` to fully support OpenAI Chat Completi\u2026",
-    "changed_files": 1,
+    "additions": 46,
+    "author": "Rocketknight1",
+    "author_association": "MEMBER",
+    "body_excerpt": "Reusing a variable name meant that we returned a softmaxed value instead of the original logits in some MoE routers. This generally did not affect inference, but could affect the auxiliary loss on MoE logits in training when the coefficien\u2026",
+    "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45257",
-    "created_at": "2026-04-05T22:07:53Z",
-    "deletions": 1,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45257/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45257",
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45315",
+    "created_at": "2026-04-08T12:54:52Z",
+    "deletions": 30,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45315/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45315",
     "labels": [],
     "merged": false,
-    "number": 45257,
+    "number": 45315,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Gemma4] Fix chat template and stop tokens for OpenAI tool calling compatibility",
-    "updated_at": "2026-04-10T13:28:04Z"
+    "title": "Fix softmaxing router logits",
+    "updated_at": "2026-04-10T13:25:20Z"
   },
   {
-    "additions": 39,
-    "author": "zozo123",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary When saving a Qwen3.5 VL model via `save_pretrained`, the `revert_weight_conversion` for `qwen3_5_text` replaces a leading `model.` segment. This wrongly matches keys that already start with `model.language_model.` on composite\u2026",
-    "changed_files": 2,
+    "additions": 18,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? fixes https://github.com/huggingface/transformers/issues/45216 and https://github.com/huggingface/transformers/issues/45310 and https://github.com/huggingface/transformers/issues/45313 TBH load-save-load works for t\u2026",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45256",
-    "created_at": "2026-04-05T19:00:26Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45314",
+    "created_at": "2026-04-08T11:54:53Z",
+    "deletions": 27,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45256/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45256",
+    "files_url": "https://github.com/huggingface/transformers/pull/45314/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45314",
     "labels": [],
     "merged": false,
-    "number": 45256,
+    "number": 45314,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: skip qwen3_5_text checkpoint remap for nested VL language_model",
-    "updated_at": "2026-04-06T18:50:06Z"
+    "title": "Conversion for LLM class loading with VLM ckpt ",
+    "updated_at": "2026-04-10T09:18:26Z"
   },
   {
-    "additions": 0,
-    "author": "ydshieh",
+    "additions": 61,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - `tests/models/qwen2/test_modeling_qwen2.py::Qwen2IntegrationTest::test_model_450m_logits`: - failed due to 6217adc6c8f0be7b5374e6a46129ad2214e4c6ed - `tests/models/smolvlm/test_modeling_smolvlm.py::SmolVLMForCondi\u2026",
-    "changed_files": 0,
+    "body_excerpt": "# What does this PR do? As per the title. It was confirmed that the weight matrices of shared layers are NEVER used, and that kv states should ALWAYS be shared, even during training or inference without Cache. I will fully remove them on a\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45254",
-    "created_at": "2026-04-05T18:20:02Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45312",
+    "created_at": "2026-04-08T11:33:33Z",
+    "deletions": 24,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45254/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45254",
+    "files_url": "https://github.com/huggingface/transformers/pull/45312/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45312",
     "labels": [],
-    "merged": false,
-    "number": 45254,
+    "merged": true,
+    "number": 45312,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix more integration tests for important models",
-    "updated_at": "2026-04-06T05:23:46Z"
+    "state": "closed",
+    "title": "[gemma4] Dissociate kv states sharing from the Cache",
+    "updated_at": "2026-04-09T08:08:07Z"
   },
   {
-    "additions": 33,
-    "author": "Charly21r",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes a bug where `use_cache=False` produces garbage logits in Gemma 4 models due to broken KV sharing between layers. Fixes #45242 ## Root cause of the issue Gemma 4 introduces two architectural features not presen\u2026",
-    "changed_files": 3,
+    "additions": 2,
+    "author": "KoichiYasuoka",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #45292 (seems to come from #41580) ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45253",
-    "created_at": "2026-04-05T18:12:08Z",
-    "deletions": 2,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45311",
+    "created_at": "2026-04-08T10:38:34Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45253/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45253",
+    "files_url": "https://github.com/huggingface/transformers/pull/45311/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45311",
     "labels": [],
     "merged": false,
-    "number": 45253,
+    "number": 45311,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix Gemma4 `use_cache=False` producing bad logits",
-    "updated_at": "2026-04-09T08:17:08Z"
+    "state": "open",
+    "title": "resize_token_embeddings does not effect to output_embeddings",
+    "updated_at": "2026-04-16T15:04:43Z"
   },
   {
-    "additions": 4,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? #43166 used `torch.set_float32_matmul_precision(\"high\")` which causes (likely) TF32 being used > \u201chigh\u201d, float32 matrix multiplications either use the TensorFloat32 datatype (10 mantissa bits explicitly stored) or t\u2026",
-    "changed_files": 1,
+    "additions": 301,
+    "author": "agentspan",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #45290. `ProcessorMixin.apply_chat_template` and several related code paths assumed every message in a conversation has a `content` key. Assistant messages with `tool_calls` and no textual content (a valid shape per the Op\u2026",
+    "changed_files": 9,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45252",
-    "created_at": "2026-04-05T16:51:29Z",
-    "deletions": 4,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45309",
+    "created_at": "2026-04-08T08:40:08Z",
+    "deletions": 23,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45252/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45252",
-    "labels": [],
-    "merged": true,
-    "number": 45252,
+    "files_url": "https://github.com/huggingface/transformers/pull/45309/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45309",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 45309,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix unexpected TF32 being enabled in testing",
-    "updated_at": "2026-04-07T10:14:50Z"
+    "title": "Fix KeyError in apply_chat_template when message has no content (#45290)",
+    "updated_at": "2026-04-08T11:30:37Z"
   },
   {
-    "additions": 91,
-    "author": "balgaly",
+    "additions": 10,
+    "author": "juliabush",
     "author_association": "NONE",
-    "body_excerpt": "## Problem `torch.multinomial` rejects last dimensions >= 2**24. Beam search with `do_sample=True` builds a flat distribution of size `num_beams * vocab_size`, which can exceed that limit (e.g. large beams + ~164k vocab), crashing during g\u2026",
-    "changed_files": 2,
+    "body_excerpt": "## What does this PR do? Fixes #29942 Flash Attention 2 inference equivalence tests for Whisper can fail due to higher numerical variance compared to the eager attention implementation. This PR increases the tolerance (`atol`, `rtol`) spec\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45251",
-    "created_at": "2026-04-05T15:38:58Z",
-    "deletions": 3,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45303",
+    "created_at": "2026-04-07T21:37:00Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45251/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45251",
+    "files_url": "https://github.com/huggingface/transformers/pull/45303/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45303",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 45251,
+    "number": 45303,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(generation): beam sample when num_beams * vocab_size exceeds multinomial limit",
-    "updated_at": "2026-04-08T14:50:40Z"
+    "title": "Fix FA2 inference equivalence failures for Whisper (closes #29942)",
+    "updated_at": "2026-04-08T14:42:36Z"
   },
   {
-    "additions": 16,
-    "author": "ydshieh",
+    "additions": 7,
+    "author": "jagwar",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? PR #42428 change the way to enable / disable torch's TF32 using torch new API. It turns out set > torch.backends.fp32_precision = False would still have > torch.backends.cudnn.conv.fp32_precision = \"tf32\" > torch.ba\u2026",
+    "body_excerpt": "## Security Fix Fixes a trust check bypass in `trl-ci-bot.yml` that allowed any GitHub user to trigger TRL CI on self-hosted GPU runners by commenting `/trl-ci` on any PR. ### The bug The \"Ignore untrusted commenter\" step used `exit 0`, wh\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45248",
-    "created_at": "2026-04-05T07:51:44Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45302",
+    "created_at": "2026-04-07T21:35:38Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45248/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45248",
+    "files_url": "https://github.com/huggingface/transformers/pull/45302/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45302",
     "labels": [],
     "merged": true,
-    "number": 45248,
-    "review_comments_count": 2,
+    "number": 45302,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix tf32 issue: set `torch.backends.cudnn.conv.fp32_precision` explicitly.",
-    "updated_at": "2026-04-10T21:48:56Z"
+    "title": "fix(security): prevent untrusted users from triggering TRL CI dispatch",
+    "updated_at": "2026-04-07T21:59:38Z"
   },
   {
-    "additions": 3,
-    "author": "gagandhakrey",
+    "additions": 0,
+    "author": "sahildando",
     "author_association": "NONE",
-    "body_excerpt": "\u2026 validation # What does this PR do? Problem The invert_attention_mask function in src/transformers/modeling_utils.py crashed with an UnboundLocalError when given an encoder_attention_mask shape that wasn't exactly 2D or 3D. Because it onl\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 0,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45247",
-    "created_at": "2026-04-05T03:45:23Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45301",
+    "created_at": "2026-04-07T21:12:29Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45247/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45247",
+    "files_url": "https://github.com/huggingface/transformers/pull/45301/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45301",
     "labels": [],
     "merged": false,
-    "number": 45247,
-    "review_comments_count": 1,
+    "number": 45301,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix UnboundLocalError in invert_attention_mask by adding proper shape\u2026",
-    "updated_at": "2026-04-09T14:11:43Z"
+    "title": " docs maintenance for transformers repository 979e8",
+    "updated_at": "2026-04-09T13:28:27Z"
   },
   {
-    "additions": 0,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Let's CI go great!!!!",
-    "changed_files": 0,
+    "additions": 136,
+    "author": "w4nderlust",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Nemotron-H models use standalone MLP layers in their `hybrid_override_pattern` (the `-` character), but the config parser, validators, and modeling code only know about `mamba`/`attention`/`moe`. This means every Ne\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45244",
-    "created_at": "2026-04-04T18:52:19Z",
-    "deletions": 0,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45300",
+    "created_at": "2026-04-07T20:57:45Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45244/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45244",
+    "files_url": "https://github.com/huggingface/transformers/pull/45300/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45300",
     "labels": [],
     "merged": false,
-    "number": 45244,
+    "number": 45300,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Let's CI go great",
-    "updated_at": "2026-04-04T19:59:28Z"
+    "title": "Fix Nemotron-H: add mlp layer type support",
+    "updated_at": "2026-04-09T16:16:30Z"
   },
   {
-    "additions": 2,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Use torch 2.11 for our (daily) CI since it's released for 2 weeks already. For CircleCI, we need to fix something regarding `torchvision.io.read_video`. For daily CI, torch 2.11 doesn't cause issues (for those `torc\u2026",
+    "additions": 3,
+    "author": "partacc",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Some CI security tests... I will contact you privately if there is anything worth reporting. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear i\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45243",
-    "created_at": "2026-04-04T18:09:27Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45243/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45243",
-    "labels": [],
-    "merged": true,
-    "number": 45243,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Nvidia CI with `torch 2.11`",
-    "updated_at": "2026-04-04T18:48:45Z"
-  },
-  {
-    "additions": 523,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? After the series of fixes in other previous PRs, we can now update the tiny model creation script. This update makes the script running without any failure, just 10 warnings. There are many # TODO, some of them may\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45241",
-    "created_at": "2026-04-04T12:30:35Z",
-    "deletions": 164,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45299",
+    "created_at": "2026-04-07T20:07:02Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45241/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45241",
+    "files_url": "https://github.com/huggingface/transformers/pull/45299/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45299",
     "labels": [],
-    "merged": true,
-    "number": 45241,
+    "merged": false,
+    "number": 45299,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Update tiny model creation script",
-    "updated_at": "2026-04-04T17:19:34Z"
+    "title": "[Please ignore] CI Test PR",
+    "updated_at": "2026-04-07T20:23:28Z"
   },
   {
-    "additions": 3,
-    "author": "shhKnight30",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Type checking for `PreTrainedConfig` subclasses broke in v5.4.0 and this fixes it. The culprit is `wrap_init_to_accept_kwargs` \u2014 it swaps out the dataclass-generated `__init__` with a `(**kwargs: Any)` wrapper at r\u2026",
-    "changed_files": 1,
+    "additions": 2958,
+    "author": "DatLe203",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45240",
-    "created_at": "2026-04-04T10:29:57Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45298",
+    "created_at": "2026-04-07T19:39:50Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45240/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45240",
+    "files_url": "https://github.com/huggingface/transformers/pull/45298/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45298",
     "labels": [],
-    "merged": true,
-    "number": 45240,
+    "merged": false,
+    "number": 45298,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: restore mypy type checking for PreTrainedConfig subclasses (#45071)",
-    "updated_at": "2026-04-10T11:10:27Z"
+    "title": "Add new qwen2 5 vl",
+    "updated_at": "2026-04-07T20:00:43Z"
   },
   {
-    "additions": 28,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? We have introduced `CausalLMModelTest` for some time, but haven't update `get_test_info.py` accordingly, which causes some issues, in particularly for tiny model creation, regarding the part of the attribute `all_mo\u2026",
+    "additions": 21,
+    "author": "EhteshamSid",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes mutable default argument bugs in four quantization config `__init__` methods inside `quantization_config.py`. In Python, mutable objects used as default argument values (e.g. `=[]`, `={}`) are created once at\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45238",
-    "created_at": "2026-04-04T07:25:15Z",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45297",
+    "created_at": "2026-04-07T18:40:05Z",
     "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45238/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45238",
-    "labels": [],
-    "merged": true,
-    "number": 45238,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Update `get_test_info.py` (related to tiny model creation)",
-    "updated_at": "2026-04-04T07:40:21Z"
-  },
-  {
-    "additions": 2,
-    "author": "KoichiYasuoka",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes the bug in `resize_token_embeddings` (may occur python3.10 or after). Quick reproduce: ``` from transformers import AutoModelForMaskedLM mdl = AutoModelForMaskedLM.from_pretrained(\"bert-base-uncased\") f = mdl.\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45236",
-    "created_at": "2026-04-04T07:05:53Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45236/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45236",
+    "files_url": "https://github.com/huggingface/transformers/pull/45297/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45297",
     "labels": [],
     "merged": false,
-    "number": 45236,
+    "number": 45297,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "resize_token_embeddings does not resize lm_head",
-    "updated_at": "2026-04-07T00:55:39Z"
+    "title": "Fix mutable default arguments in quantization config classes",
+    "updated_at": "2026-04-09T14:00:25Z"
   },
   {
-    "additions": 102,
-    "author": "kallewoof",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds a tiny \"agnostic.gpu\" utility that is meant to allow easy replacing of unnecessarily hard-coded vendor-specific code. The code does not use `torch.accelerator` as it is still considered experimental, bu\u2026",
-    "changed_files": 8,
+    "additions": 191,
+    "author": "UsamaKenway",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45235",
-    "created_at": "2026-04-04T06:08:22Z",
-    "deletions": 13,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45235/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45235",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45296",
+    "created_at": "2026-04-07T18:39:33Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45296/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45296",
     "labels": [],
     "merged": false,
-    "number": 45235,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "feat/rfc/poc: Agnostic GPU",
-    "updated_at": "2026-04-09T14:34:20Z"
+    "number": 45296,
+    "review_comments_count": 2,
+    "state": "open",
+    "title": "Add GGUF support to Gemma4 (31B & 26B-A4B) text ",
+    "updated_at": "2026-04-12T10:53:54Z"
   },
   {
-    "additions": 306,
-    "author": "pdufour",
+    "additions": 116,
+    "author": "jesperschlegel",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This fixes an issues with the model that made it incompatible with exporting to onnx. Specifically the following has been changed: 1. if condition ``` if input_len < context_len: ``` Will give this error when you tr\u2026",
+    "body_excerpt": "Adds the Gemma4ForSequenceClassification class which supports finetuning Gemma 4 models on sequence classification tasks.",
     "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45233",
-    "created_at": "2026-04-04T00:15:04Z",
-    "deletions": 98,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45294",
+    "created_at": "2026-04-07T17:51:38Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45233/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45233",
+    "files_url": "https://github.com/huggingface/transformers/pull/45294/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45294",
     "labels": [],
     "merged": false,
-    "number": 45233,
-    "review_comments_count": 4,
+    "number": 45294,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "feat: make timesfm2_5 onnx export compatible",
-    "updated_at": "2026-04-10T23:37:52Z"
+    "title": "feat: add Gemma4ForSequenceClassification",
+    "updated_at": "2026-04-13T13:05:41Z"
   },
   {
-    "additions": 439,
-    "author": "stevhliu",
+    "additions": 4,
+    "author": "yonigozlan",
     "author_association": "MEMBER",
-    "body_excerpt": "adds docs for static model rules so model contributors are aware of expectations > I wonder if it wouldn't make sense to auto generate that portion of the doc automatically added from @tarekziade feedback: - reformats `format_rule_details(\u2026",
-    "changed_files": 8,
+    "body_excerpt": "## Fix global state leak in `AutoTokenizer.register` causing test failures ### Problem `test_from_pretrained_dynamic_processor` was failing when run as part of the full test class with: ``` AttributeError: NewTokenizer has no attribute spe\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45232",
-    "created_at": "2026-04-03T22:41:22Z",
-    "deletions": 146,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45293",
+    "created_at": "2026-04-07T16:29:25Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45232/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45232",
+    "files_url": "https://github.com/huggingface/transformers/pull/45293/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45293",
     "labels": [],
-    "merged": true,
-    "number": 45232,
+    "merged": false,
+    "number": 45293,
     "review_comments_count": 1,
-    "state": "closed",
-    "title": "[docs] static model rules",
-    "updated_at": "2026-04-08T15:06:17Z"
+    "state": "open",
+    "title": "Fix \"AttributeError: NewTokenizer has no attribute special_attribute_present\" (Remove `REGISTERED_FAST_ALIASES`)",
+    "updated_at": "2026-04-09T12:04:21Z"
   },
   {
-    "additions": 20,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? All of this are trivial ... (maybe except \"evolla\")",
-    "changed_files": 10,
+    "additions": 2958,
+    "author": "DatLe203",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 11,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45228",
-    "created_at": "2026-04-03T18:17:29Z",
-    "deletions": 12,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45291",
+    "created_at": "2026-04-07T14:30:38Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45228/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45228",
+    "files_url": "https://github.com/huggingface/transformers/pull/45291/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45291",
     "labels": [],
-    "merged": true,
-    "number": 45228,
+    "merged": false,
+    "number": 45291,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "More fix for tiny model creation",
-    "updated_at": "2026-04-03T19:25:51Z"
+    "title": "First pull request",
+    "updated_at": "2026-04-07T14:53:52Z"
   },
   {
-    "additions": 3,
-    "author": "akhilc08",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary - Remove the unused `PILImageResampling` runtime import from `video_processing_utils.py` which causes an `ImportError` when Pillow is not installed - Also remove the now-unused `is_vision_available` import that only guarded the\u2026",
-    "changed_files": 2,
+    "additions": 24,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45265 and doesn't warn if the value is an int.",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45227",
-    "created_at": "2026-04-03T17:44:25Z",
-    "deletions": 7,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45289",
+    "created_at": "2026-04-07T13:00:52Z",
+    "deletions": 73,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45227/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45227",
+    "files_url": "https://github.com/huggingface/transformers/pull/45289/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45289",
     "labels": [],
-    "merged": false,
-    "number": 45227,
+    "merged": true,
+    "number": 45289,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: remove nonexistent PILImageResampling import from video_processing_utils",
-    "updated_at": "2026-04-05T17:04:49Z"
+    "title": "Less unnecessary RoPE warnings",
+    "updated_at": "2026-04-13T13:14:41Z"
   },
   {
-    "additions": 3,
-    "author": "akhilc08",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary - Fixes an `IndexError: string index out of range` crash in `_split_tokens_on_unicode()` when the decoded token stream ends with a dangling Unicode replacement character (U+FFFD) - Adds a bounds check so that when `unicode_offse\u2026",
+    "additions": 35,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": null,
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45226",
-    "created_at": "2026-04-03T17:43:21Z",
-    "deletions": 3,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45226/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45226",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45288",
+    "created_at": "2026-04-07T10:24:53Z",
+    "deletions": 11,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45288/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45288",
     "labels": [],
     "merged": false,
-    "number": 45226,
+    "number": 45288,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: handle trailing replacement character in Whisper word timestamp decoding",
-    "updated_at": "2026-04-05T17:04:49Z"
+    "title": "fix(cohere_asr): auto-fix failing tests",
+    "updated_at": "2026-04-10T10:41:27Z"
   },
   {
-    "additions": 6,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The `dev` extra now indirectly pulls hf-doc-builder so the install step failed. We also need to update to current main for the latest features",
+    "additions": 102,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": null,
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45225",
-    "created_at": "2026-04-03T16:46:54Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45225/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45225",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45287",
+    "created_at": "2026-04-07T10:23:45Z",
+    "deletions": 0,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45287/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45287",
     "labels": [],
-    "merged": true,
-    "number": 45225,
+    "merged": false,
+    "number": 45287,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: hf-doc-builder insallation was failing",
-    "updated_at": "2026-04-03T17:02:09Z"
+    "title": "fix(videomt): auto-fix failing tests",
+    "updated_at": "2026-04-10T10:41:30Z"
   },
   {
-    "additions": 2,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix for tiny model",
-    "changed_files": 2,
+    "additions": 29,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": null,
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45224",
-    "created_at": "2026-04-03T16:39:23Z",
-    "deletions": 4,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45224/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45224",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45286",
+    "created_at": "2026-04-07T10:22:51Z",
+    "deletions": 1,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45286/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45286",
     "labels": [],
-    "merged": true,
-    "number": 45224,
+    "merged": false,
+    "number": 45286,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "remove unnecessary entries in some auto model mappings",
-    "updated_at": "2026-04-03T17:26:24Z"
+    "title": "fix(nomic_bert): auto-fix failing tests",
+    "updated_at": "2026-04-10T10:41:28Z"
   },
   {
-    "additions": 17,
-    "author": "Talhax55z",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #31356 ## What does this PR do? The `postprocess` method in `ObjectDetectionPipeline` was hardcoding `raw_annotations[0]`, which caused batch inference to only return results for the first image, ignoring all others. This PR replaces\u2026",
-    "changed_files": 1,
+    "additions": 153,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45223",
-    "created_at": "2026-04-03T16:36:25Z",
-    "deletions": 15,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45285",
+    "created_at": "2026-04-07T09:18:11Z",
+    "deletions": 329,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45223/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45223",
+    "files_url": "https://github.com/huggingface/transformers/pull/45285/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45285",
     "labels": [],
-    "merged": false,
-    "number": 45223,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 45285,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "Fix: ObjectDetectionPipeline batch inference only returns first image results",
-    "updated_at": "2026-04-13T15:06:32Z"
+    "title": "Fix export for gemma4 and add Integration tests",
+    "updated_at": "2026-04-08T13:07:40Z"
   },
   {
-    "additions": 12,
-    "author": "jashshah999",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary When using Gemma 3 or Gemma 4 for text-only supervised fine-tuning (no images), the forward pass raises a `ValueError` because `token_type_ids` / `mm_token_type_ids` is not provided. This happens because `AutoTokenizer` does not\u2026",
-    "changed_files": 4,
+    "additions": 4,
+    "author": "Abdennacer-Badaoui",
+    "author_association": "MEMBER",
+    "body_excerpt": "Following this PR : https://github.com/huggingface/transformers/pull/45268. This PR fix Qwen2 expectations for AMD.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45222",
-    "created_at": "2026-04-03T16:27:31Z",
-    "deletions": 4,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45284",
+    "created_at": "2026-04-07T09:12:01Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45222/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45222",
-    "labels": [],
-    "merged": false,
-    "number": 45222,
+    "files_url": "https://github.com/huggingface/transformers/pull/45284/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45284",
+    "labels": [],
+    "merged": true,
+    "number": 45284,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(gemma3, gemma4): default token_type_ids to zeros for text-only training",
-    "updated_at": "2026-04-15T15:31:05Z"
+    "title": "[AMD CI] Fix Qwen2 expectations",
+    "updated_at": "2026-04-07T12:01:58Z"
   },
   {
-    "additions": 6,
-    "author": "eustlb",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per title",
-    "changed_files": 1,
+    "additions": 107,
+    "author": "jIab-b",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45221",
-    "created_at": "2026-04-03T15:49:24Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45283",
+    "created_at": "2026-04-07T09:11:14Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45221/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45221",
+    "files_url": "https://github.com/huggingface/transformers/pull/45283/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45283",
     "labels": [],
     "merged": false,
-    "number": 45221,
+    "number": 45283,
     "review_comments_count": 0,
     "state": "open",
-    "title": "user friendly error when loading audio from video",
-    "updated_at": "2026-04-09T14:16:24Z"
+    "title": "Add Qwen3.5 GGUF loading support",
+    "updated_at": "2026-04-09T14:49:22Z"
   },
   {
-    "additions": 1711,
-    "author": "SunMarc",
+    "additions": 30,
+    "author": "Abdennacer-Badaoui",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds transformers serve compatibility to multimodal models like qwen omni or gemma 4. We add support for audio with chat completion and response though `input_audio` -> the client need to `base64-encode` the\u2026",
-    "changed_files": 7,
+    "body_excerpt": "`_register_model_output_pytree_node` was calling set.__contains__ during TorchDynamo tracing, which is unsupported in PyTorch 2.8.0 (ROCm). Added an early return when `torch.compiler.is_compiling()` is True, since pytree nodes are already\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45220",
-    "created_at": "2026-04-03T14:16:33Z",
-    "deletions": 163,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45282",
+    "created_at": "2026-04-07T08:50:54Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45220/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45220",
+    "files_url": "https://github.com/huggingface/transformers/pull/45282/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45282",
     "labels": [],
     "merged": true,
-    "number": 45220,
-    "review_comments_count": 18,
+    "number": 45282,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Multimodal serve support ",
-    "updated_at": "2026-04-15T14:13:02Z"
+    "title": "[AMD CI] Fix torch.compile/export failures on AMD CI due to untraceable set.__contains__ ",
+    "updated_at": "2026-04-13T15:59:55Z"
   },
   {
-    "additions": 3,
-    "author": "sywangyi",
+    "additions": 25,
+    "author": "zhang-prog",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? google/gemma-4-26B-A4B-it tp 2, memory is 46G per rank wo the change, drop to about 25G w per rank with the change - text models: @ArthurZucker @Cyrilvallez",
-    "changed_files": 1,
+    "body_excerpt": "### Description This PR fixes a boundary issue in the `_extract_polygon_points_by_masks` method of PP-DocLayoutV3. When running inference with a low confidence threshold, or due to coordinate clipping during scaling, the extracted `cropped\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45219",
-    "created_at": "2026-04-03T13:59:02Z",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45281",
+    "created_at": "2026-04-07T08:49:24Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45219/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45219",
+    "files_url": "https://github.com/huggingface/transformers/pull/45281/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45281",
     "labels": [],
     "merged": true,
-    "number": 45219,
+    "number": 45281,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add MoE to Gemma4 TP plan",
-    "updated_at": "2026-04-08T14:09:40Z"
+    "title": "Fix resize failure caused by zero-sized masks in PP-DocLayoutV3",
+    "updated_at": "2026-04-09T13:06:44Z"
   },
   {
-    "additions": 4057,
-    "author": "LysandreJik",
-    "author_association": "MEMBER",
-    "body_excerpt": "This PR offers a new, Agentic surface for transformers. It tries to apply what is done elsewhere with CLIs to `transformers`, leveraging many current use-cases of `transformers` and exposing them as CLI endpoints. I recommend reading this\u2026",
-    "changed_files": 15,
+    "additions": 2502,
+    "author": "marvinzh",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? add Qianfan-OCR model definition - `QianfanOCRForConditionalGeneration` - image-text to text model definition - `QianfanOCRModel` - backbone of image-text to text model without lm heads - `QianfanOCRProcessor` - tex\u2026",
+    "changed_files": 18,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45218",
-    "created_at": "2026-04-03T13:31:08Z",
-    "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45218/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45218",
-    "labels": [],
+    "comments_count": 14,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45280",
+    "created_at": "2026-04-07T06:49:34Z",
+    "deletions": 165,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45280/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45280",
+    "labels": [
+      "New model"
+    ],
     "merged": false,
-    "number": 45218,
-    "review_comments_count": 0,
+    "number": 45280,
+    "review_comments_count": 73,
     "state": "open",
-    "title": "Proposal: Agent-first CLI",
-    "updated_at": "2026-04-13T13:03:00Z"
+    "title": "add Qianfan-OCR model definition",
+    "updated_at": "2026-04-16T16:29:11Z"
   },
   {
-    "additions": 5,
-    "author": "ENg-122",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Remove unnecessary masked_fill_(mask, 0) call in torch_chunk_gated_delta_rule. The decay_mask computed earlier already encodes the causal/lower-triangular structure (upper-triangle values are zero), so masking the a\u2026",
+    "additions": 73,
+    "author": "sywangyi",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45215",
-    "created_at": "2026-04-03T09:08:28Z",
-    "deletions": 5,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45279",
+    "created_at": "2026-04-07T06:40:35Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45215/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45215",
+    "files_url": "https://github.com/huggingface/transformers/pull/45279/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45279",
+    "labels": [],
+    "merged": false,
+    "number": 45279,
+    "review_comments_count": 5,
+    "state": "open",
+    "title": "add expert parallelism for gemma-4-26B-A4B-it",
+    "updated_at": "2026-04-16T07:20:17Z"
+  },
+  {
+    "additions": 4,
+    "author": "kamalrajkannan78",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? - Gemma3ForConditionalGeneration.forward & Gemma3ForSequenceClassification.forward calls self.model() without return_dict=True, so @can_return_tuple silently converts the output to a plain tuple, causing outputs.pas\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45277",
+    "created_at": "2026-04-07T05:24:43Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45277/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45277",
     "labels": [],
     "merged": true,
-    "number": 45215,
+    "number": 45277,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Qwen3_5]Remove unnecessary masked_fill_ in torch_chunk_gated_delta_rule attention computation: \"attn = (q_i @ k_i.transpose(-1, -2) * decay_mask[:, :, i]).masked_fill_(mask, 0)\"",
-    "updated_at": "2026-04-09T14:12:51Z"
+    "title": "Fix AttributeError in Gemma3ForConditionalGeneration and Gemma3ForSequenceClassification when config.return_dict=False",
+    "updated_at": "2026-04-10T10:37:16Z"
   },
   {
-    "additions": 45,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR fixes failed test case: `tests/models/cohere_asr/test_modeling_cohere_asr.py::CohereAsrModelTest::test_model_parallel_beam_search`, and add some adjustment to make the test cases pass for Intel XPU device. @ydshieh pls help review,\u2026",
+    "additions": 9,
+    "author": "avarga1",
+    "author_association": "NONE",
+    "body_excerpt": "## Problem `AutoConfig.from_pretrained(\"baidu/ERNIE-4.5-VL-28B-A3B-Paddle\", trust_remote_code=True)` raises errors that prevent the model from loading at all. Three separate bugs compound each other: ### Bug 1 \u2014 `model_type` mismatch (KeyE\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45214",
-    "created_at": "2026-04-03T08:32:34Z",
-    "deletions": 14,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45275",
+    "created_at": "2026-04-07T02:22:26Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45214/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45214",
+    "files_url": "https://github.com/huggingface/transformers/pull/45275/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45275",
     "labels": [],
-    "merged": true,
-    "number": 45214,
-    "review_comments_count": 2,
+    "merged": false,
+    "number": 45275,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "cohere_asr: fix bug for model_parallel_beam_search test case",
-    "updated_at": "2026-04-13T02:40:29Z"
+    "title": "fix(ernie4_5_vl_moe): resolve three config loading failures for ERNIE-4.5-VL MoE models",
+    "updated_at": "2026-04-09T16:57:22Z"
   },
   {
-    "additions": 6052,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "comparison https://github.com/huggingface/transformers/blob/937d61b9fa00001da1a0680ecf8061b5990fbcd7/sarvam_moe_comparison.md # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once mer\u2026",
-    "changed_files": 26,
+    "additions": 215,
+    "author": "Qubitium",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? 1. Fix: CUDA graph reuse for FA2 continuous batching was wrongly keyed causing quality collapse for specific configuration CUDA graph reuse used the wrong key: replay reuse depended on padded tensor sizes, but FA va\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45213",
-    "created_at": "2026-04-03T08:25:49Z",
-    "deletions": 209,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45213/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45213",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45274",
+    "created_at": "2026-04-07T01:43:12Z",
+    "deletions": 36,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45274/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45274",
     "labels": [],
     "merged": false,
-    "number": 45213,
+    "number": 45274,
     "review_comments_count": 0,
     "state": "open",
-    "title": "DO NOT MERGE - model creation skill",
-    "updated_at": "2026-04-03T12:26:09Z"
+    "title": "Fix CB Accuracy Regression under FA2",
+    "updated_at": "2026-04-09T03:47:47Z"
   },
   {
-    "additions": 29,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@ydshieh pls help review, thx!",
-    "changed_files": 2,
+    "additions": 2,
+    "author": "excepshenal",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Liger kernel unnecessarily materializes logits in VRAM during eval with `prediction_loss_only=True`, causing OOM. We explicitly tell Liger to `skip_logits`. <!-- Congratulations! You've made it this far! You're not\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45212",
-    "created_at": "2026-04-03T07:44:35Z",
-    "deletions": 6,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45273",
+    "created_at": "2026-04-06T21:11:21Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45212/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45212",
+    "files_url": "https://github.com/huggingface/transformers/pull/45273/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45273",
     "labels": [],
-    "merged": true,
-    "number": 45212,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "musicflamingo: add test support for Intel XPU device",
-    "updated_at": "2026-04-10T10:41:20Z"
+    "merged": false,
+    "number": 45273,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "fix: liger unnecessarily materializes logits in VRAM during eval, causing OOM",
+    "updated_at": "2026-04-09T15:32:29Z"
   },
   {
-    "additions": 4,
-    "author": "matdou",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #45208 # What does this PR do? This PR corrects an incorrect return type in `Qwen3MoeSparseMoeBlock.forward`. The method was annotated as returning `tuple[torch.Tensor, torch.Tensor]`, while the implementation returns a `torch.Tensor\u2026",
-    "changed_files": 4,
+    "additions": 6,
+    "author": "yonigozlan",
+    "author_association": "MEMBER",
+    "body_excerpt": "Cc @ydshieh",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45211",
-    "created_at": "2026-04-03T07:44:32Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45272",
+    "created_at": "2026-04-06T20:39:29Z",
+    "deletions": 20,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45211/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45211",
+    "files_url": "https://github.com/huggingface/transformers/pull/45272/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45272",
     "labels": [],
     "merged": false,
-    "number": 45211,
+    "number": 45272,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[Qwen3MoE] Fix wrong return type annotation in Qwen3MoeSparseMoeBlock.forward",
-    "updated_at": "2026-04-08T19:20:49Z"
+    "state": "open",
+    "title": "Fix redundant logic in video processing SmolVLM",
+    "updated_at": "2026-04-06T20:49:13Z"
   },
   {
-    "additions": 11,
-    "author": "ArthurZucker",
+    "additions": 94,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Commits that got in the release branch to allow pushing",
-    "changed_files": 3,
+    "body_excerpt": "adds a separate vlm contribution doc for more visibility instead of being hidden in the Contribute to Transformers doc, and integration tests are covered in #45152",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45210",
-    "created_at": "2026-04-03T05:42:21Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45271",
+    "created_at": "2026-04-06T18:56:42Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45210/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45210",
+    "files_url": "https://github.com/huggingface/transformers/pull/45271/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45271",
     "labels": [],
     "merged": true,
-    "number": 45210,
-    "review_comments_count": 0,
+    "number": 45271,
+    "review_comments_count": 8,
     "state": "closed",
-    "title": "Fix pypi release",
-    "updated_at": "2026-04-03T06:40:39Z"
+    "title": "[docs] vlm addition",
+    "updated_at": "2026-04-15T17:02:43Z"
+  },
+  {
+    "additions": 162,
+    "author": "madhav1k",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Introduce logging of individual loss components when models return a dict of losses. - Add TrainingArguments.logging_loss_components flag to enable/disable this behavior. - Track per-component running sums with _tr_loss_components and aggr\u2026",
+    "changed_files": 3,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45270",
+    "created_at": "2026-04-06T18:38:51Z",
+    "deletions": 7,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45270/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45270",
+    "labels": [],
+    "merged": false,
+    "number": 45270,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "[Trainer] Support multi-loss component logging",
+    "updated_at": "2026-04-13T08:30:40Z"
   },
   {
     "additions": 3,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@ydshieh pls help review, thx!",
+    "author": "ryota-komatsu",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fix minor typos in `src/transformers/utils/output_capturing.py`: - `install_output_capuring_hook` \u2192 `install_output_capturing_hook` (lines 98, 147) - `Tis` \u2192 `This` (line 152) ## Before submitting - [x] This PR fixe\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45209",
-    "created_at": "2026-04-03T05:40:36Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45269",
+    "created_at": "2026-04-06T18:01:42Z",
     "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45209/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45209",
+    "files_url": "https://github.com/huggingface/transformers/pull/45269/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45269",
     "labels": [],
-    "merged": true,
-    "number": 45209,
-    "review_comments_count": 2,
-    "state": "closed",
-    "title": "nomic_bert: make the test suitable for general device.",
-    "updated_at": "2026-04-13T02:40:26Z"
+    "merged": false,
+    "number": 45269,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix typos in src/transformers/utils/output_capturing.py",
+    "updated_at": "2026-04-06T20:49:06Z"
   },
   {
-    "additions": 65,
-    "author": "w4nderlust",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes #45206 ## What does this PR do? Adds documentation for the Gemma4 Per-Layer Embeddings (PLE) system, which is currently pretty hard to reverse-engineer from the code alone. I ran into this while implementing Gemma4 inference from scr\u2026",
-    "changed_files": 4,
+    "additions": 3,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix `Qwen2IntegrationTest`: - `test_speculative_generation`: - `0c89522f`: #43794 changed seed, so the actual output changed, but the expected output is not updated (cc @tarekziade more attention next time \ud83d\ude04 ) - `af\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45207",
-    "created_at": "2026-04-03T05:15:47Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45268",
+    "created_at": "2026-04-06T18:01:38Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45207/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45207",
+    "files_url": "https://github.com/huggingface/transformers/pull/45268/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45268",
     "labels": [],
     "merged": true,
-    "number": 45207,
-    "review_comments_count": 6,
+    "number": 45268,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "[Gemma4] Add docstrings for Per-Layer Embeddings (PLE) pipeline",
-    "updated_at": "2026-04-14T20:03:58Z"
+    "title": "Fix `Qwen2IntegrationTest`",
+    "updated_at": "2026-04-08T07:56:35Z"
   },
   {
-    "additions": 112,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@ydshieh pls help review, thx!",
-    "changed_files": 3,
+    "additions": 108,
+    "author": "KetanP1618",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Adds a missing docstring to the `FFN.forward` method in DistilBERT. The `FFN.forward` method in `modeling_distilbert.py` had no documentation at all. This PR adds proper Args and Returns sections following the exis\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45204",
-    "created_at": "2026-04-03T02:32:39Z",
-    "deletions": 10,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45267",
+    "created_at": "2026-04-06T17:59:13Z",
+    "deletions": 39,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45204/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45204",
+    "files_url": "https://github.com/huggingface/transformers/pull/45267/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45267",
     "labels": [],
-    "merged": true,
-    "number": 45204,
-    "review_comments_count": 2,
-    "state": "closed",
-    "title": "fix bug for videomt model device mismatch",
-    "updated_at": "2026-04-13T02:40:27Z"
+    "merged": false,
+    "number": 45267,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Add docstring to FFN.forward in DistilBERT",
+    "updated_at": "2026-04-08T13:43:57Z"
   },
   {
-    "additions": 2,
-    "author": "Qubitium",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Disable FlashAttention support for Gemm4 which FA cannot suport due to global.head-dim=512. I am very confused at the current code/test for Gemma4. I ran real inference using transformer `main` and `fa` throws head-\u2026",
-    "changed_files": 2,
+    "additions": 125,
+    "author": "KetanP1618",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Adds missing docstrings to two undocumented `forward` methods in the ALBERT model: - `AlbertMLMHead.forward` - Added Args and Returns sections - `AlbertSOPHead.forward` - Added Args and Returns sections Both method\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45202",
-    "created_at": "2026-04-02T23:37:22Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45266",
+    "created_at": "2026-04-06T17:41:14Z",
+    "deletions": 41,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45202/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45202",
+    "files_url": "https://github.com/huggingface/transformers/pull/45266/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45266",
     "labels": [],
     "merged": false,
-    "number": 45202,
-    "review_comments_count": 4,
-    "state": "open",
-    "title": "Fix gemma4 has flash-attention incompatbile head-dim=512",
-    "updated_at": "2026-04-12T03:39:09Z"
+    "number": 45266,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Add docstrings to AlbertMLMHead and AlbertSOPHead forward methods",
+    "updated_at": "2026-04-08T13:43:34Z"
   },
   {
-    "additions": 18,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following **Wav2Vec2PhonemeCTC** use cases were identified and fixed in this PR: \u2192 [05c0e1d (\"rm slow tokenizers\")](https://github.com/huggingface/transformers/pull/40936) added [self.backend = kwargs.pop(\"bac\u2026",
-    "changed_files": 2,
+    "additions": 1,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? **Fix tf32 issue: set `torch.backends.cudnn.conv.fp32_precision` explicitly.** (#45248) breaks when running on torch 2.8 or older ...",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45199",
-    "created_at": "2026-04-02T20:03:22Z",
-    "deletions": 6,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45263",
+    "created_at": "2026-04-06T10:06:40Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45199/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45199",
+    "files_url": "https://github.com/huggingface/transformers/pull/45263/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45263",
     "labels": [],
     "merged": true,
-    "number": 45199,
-    "review_comments_count": 2,
+    "number": 45263,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(models): Resolve regressions in Wav2Vec2PhonemeCTCTokenizer (wav2vec2-lv-60-espeak-cv-ft)",
-    "updated_at": "2026-04-14T14:00:01Z"
+    "title": "Add `hasattr(torch.backends.cudnn, \"conv\")` to `conftest.py`",
+    "updated_at": "2026-04-06T19:46:18Z"
   },
   {
-    "additions": 104,
-    "author": "douglas-reid",
+    "additions": 1,
+    "author": "lowzhao",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes issues with the Gemma 4 model docs. Mainly, this is updating examples to point at the actual models, with FC and Audio examples added. - [ X ] I confirm that this is not a pure code agent PR. ## Before submitting - [ X ] This PR fixe\u2026",
+    "body_excerpt": "# What does this PR do? Fix docstring spelling mistake TokenizersBackend.convert_to_native_format. ```python @classmethod def convert_to_native_format(cls, trust_remote_code=False, **kwargs): \"\"\"s <---- additional s ``` Likely caused by mi\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45197",
-    "created_at": "2026-04-02T19:05:35Z",
-    "deletions": 42,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45262",
+    "created_at": "2026-04-06T08:41:40Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45197/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45197",
+    "files_url": "https://github.com/huggingface/transformers/pull/45262/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45262",
     "labels": [],
     "merged": true,
-    "number": 45197,
+    "number": 45262,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(docs): correct gemma4 docs and examples",
-    "updated_at": "2026-04-02T22:23:16Z"
+    "title": "doc: fix TokenizersBackend.convert_to_native_format docstring",
+    "updated_at": "2026-04-06T16:32:44Z"
   },
   {
-    "additions": 1,
-    "author": "stevhliu",
+    "additions": 22,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "fixes `<hfoptions>` tag",
+    "body_excerpt": "# What this PR does It's working, see https://github.com/huggingface/transformers/actions/runs/24025915210 (the failing job in this PR is because the workflow needs to be on `main` to be effective).",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45196",
-    "created_at": "2026-04-02T18:02:44Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45261",
+    "created_at": "2026-04-06T07:52:29Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45196/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45196",
+    "files_url": "https://github.com/huggingface/transformers/pull/45261/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45261",
     "labels": [],
     "merged": true,
-    "number": 45196,
+    "number": 45261,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] formatting",
-    "updated_at": "2026-04-03T09:37:52Z"
+    "title": "empty",
+    "updated_at": "2026-04-06T09:24:59Z"
   },
   {
-    "additions": 90,
-    "author": "yonigozlan",
+    "additions": 18,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What this PR does Adds a new `load_image_as_tensor` utility leveraging torchvision's `decode_image` to `image_utils.py` and overrides `fetch_images` in `TorchvisionBackend` to use it. Previously, all image loading went through PIL regard\u2026",
-    "changed_files": 5,
+    "body_excerpt": "## Summary PR #43514 refactored `_preprocess` to pass `resample=resample` to `resize`, but the `resize` method in `SmolVLMVideoProcessor` still had `interpolation` as its parameter name. The `resample` kwarg was silently swallowed by `**kw\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45195",
-    "created_at": "2026-04-02T17:58:51Z",
-    "deletions": 20,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45258",
+    "created_at": "2026-04-06T05:09:09Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45195/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45195",
+    "files_url": "https://github.com/huggingface/transformers/pull/45258/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45258",
     "labels": [],
     "merged": true,
-    "number": 45195,
-    "review_comments_count": 5,
+    "number": 45258,
+    "review_comments_count": 7,
     "state": "closed",
-    "title": " Use torchvision `decode_image` to load images in the torchvision backend",
-    "updated_at": "2026-04-09T17:08:04Z"
+    "title": "Fix `SmolVLM` video processor `resize` using wrong interpolation after backend refactor",
+    "updated_at": "2026-04-06T20:40:22Z"
   },
   {
-    "additions": 222,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes: - Replace `PretrainedConfig` with `PreTrainedConfig` - Don't import from other models inside `configuration_model.py`, instead resolve via modular - Text/vision sub-configs are documented (only kosmos was mis\u2026",
-    "changed_files": 23,
+    "additions": 269,
+    "author": "lucianommartins",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# [Gemma4] Fix chat template and stop tokens for OpenAI tool calling compatibility ## What does this PR do? Rewrites the `_patch_template_for_openai_tool_role()` function in `convert_gemma4_weights.py` to fully support OpenAI Chat Completi\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45194",
-    "created_at": "2026-04-02T16:39:06Z",
-    "deletions": 40,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45194/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45194",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45257",
+    "created_at": "2026-04-05T22:07:53Z",
+    "deletions": 1,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45257/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45257",
     "labels": [],
     "merged": false,
-    "number": 45194,
-    "review_comments_count": 1,
+    "number": 45257,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Configuration insoncistencies",
-    "updated_at": "2026-04-10T11:39:18Z"
+    "title": "[Gemma4] Fix chat template and stop tokens for OpenAI tool calling compatibility",
+    "updated_at": "2026-04-10T13:28:04Z"
   },
   {
-    "additions": 15,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45070 Though I am convinced that configs should not be pyndatic BaseClass, just because we already wrap all subclasses as dataclass. You never know what happe\u2026",
+    "additions": 39,
+    "author": "zozo123",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary When saving a Qwen3.5 VL model via `save_pretrained`, the `revert_weight_conversion` for `qwen3_5_text` replaces a leading `model.` segment. This wrongly matches keys that already start with `model.language_model.` on composite\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45193",
-    "created_at": "2026-04-02T15:36:56Z",
-    "deletions": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45256",
+    "created_at": "2026-04-05T19:00:26Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45193/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45193",
+    "files_url": "https://github.com/huggingface/transformers/pull/45256/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45256",
     "labels": [],
     "merged": false,
-    "number": 45193,
+    "number": 45256,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Config can apply pyndatic validation without torch-dependence",
-    "updated_at": "2026-04-02T17:00:43Z"
+    "state": "closed",
+    "title": "fix: skip qwen3_5_text checkpoint remap for nested VL language_model",
+    "updated_at": "2026-04-06T18:50:06Z"
   },
   {
-    "additions": 9896,
-    "author": "RyanMullins",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "--------- # What does this PR do? model previously unable to use tools ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenec\u2026",
-    "changed_files": 41,
+    "additions": 0,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? - `tests/models/qwen2/test_modeling_qwen2.py::Qwen2IntegrationTest::test_model_450m_logits`: - failed due to 6217adc6c8f0be7b5374e6a46129ad2214e4c6ed - `tests/models/smolvlm/test_modeling_smolvlm.py::SmolVLMForCondi\u2026",
+    "changed_files": 0,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45192",
-    "created_at": "2026-04-02T14:35:18Z",
-    "deletions": 79,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45254",
+    "created_at": "2026-04-05T18:20:02Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45192/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45192",
-    "labels": [
-      "New model"
-    ],
-    "merged": true,
-    "number": 45192,
-    "review_comments_count": 2,
-    "state": "closed",
-    "title": "casually dropping the most capable open weights on the planet",
-    "updated_at": "2026-04-03T12:28:06Z"
+    "files_url": "https://github.com/huggingface/transformers/pull/45254/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45254",
+    "labels": [],
+    "merged": false,
+    "number": 45254,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix more integration tests for important models",
+    "updated_at": "2026-04-06T05:23:46Z"
   },
   {
-    "additions": 13,
-    "author": "saslifat-gif",
-    "author_association": "NONE",
-    "body_excerpt": "The Qwen2 tokenizer test file had no custom test methods \u2014 only integration constants inherited from TokenizerTesterMixin. This PR adds a test documenting two untested edge cases in decode(): **Before (no test, behavior undocumented):** ``\u2026",
-    "changed_files": 1,
+    "additions": 33,
+    "author": "Charly21r",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes a bug where `use_cache=False` produces garbage logits in Gemma 4 models due to broken KV sharing between layers. Fixes #45242 ## Root cause of the issue Gemma 4 introduces two architectural features not presen\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45191",
-    "created_at": "2026-04-02T14:08:15Z",
-    "deletions": 0,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45253",
+    "created_at": "2026-04-05T18:12:08Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45191/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45191",
+    "files_url": "https://github.com/huggingface/transformers/pull/45253/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45253",
     "labels": [],
     "merged": false,
-    "number": 45191,
+    "number": 45253,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add edge case tests for out-of-range token id decoding in Qwen2 tokenizer",
-    "updated_at": "2026-04-08T12:18:56Z"
+    "title": "Fix Gemma4 `use_cache=False` producing bad logits",
+    "updated_at": "2026-04-09T08:17:08Z"
   },
   {
-    "additions": 92,
-    "author": "SunMarc",
+    "additions": 4,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This patch extends ty check to src/transformers/cli. Supersedes https://github.com/huggingface/transformers/pull/44566. I've added some of the changes in this PR",
-    "changed_files": 10,
+    "body_excerpt": "# What does this PR do? #43166 used `torch.set_float32_matmul_precision(\"high\")` which causes (likely) TF32 being used > \u201chigh\u201d, float32 matrix multiplications either use the TensorFloat32 datatype (10 mantissa bits explicitly stored) or t\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45190",
-    "created_at": "2026-04-02T13:54:23Z",
-    "deletions": 32,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45252",
+    "created_at": "2026-04-05T16:51:29Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45190/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45190",
+    "files_url": "https://github.com/huggingface/transformers/pull/45252/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45252",
     "labels": [],
     "merged": true,
-    "number": 45190,
-    "review_comments_count": 18,
+    "number": 45252,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix ty for transformers cli",
-    "updated_at": "2026-04-14T16:53:22Z"
+    "title": "Fix unexpected TF32 being enabled in testing",
+    "updated_at": "2026-04-07T10:14:50Z"
   },
   {
-    "additions": 409,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "## Summary - Add two optional, backward-compatible inputs (`test_path_prefix`, `test_file_suffix`) to `model_jobs.yml` and `self-scheduled.yml` \u2014 defaults preserve all existing behavior - Extend the `set-matrix` step in `self-scheduled.yml\u2026",
-    "changed_files": 5,
+    "additions": 91,
+    "author": "balgaly",
+    "author_association": "NONE",
+    "body_excerpt": "## Problem `torch.multinomial` rejects last dimensions >= 2**24. Beam search with `do_sample=True` builds a flat distribution of size `num_beams * vocab_size`, which can exceed that limit (e.g. large beams + ~164k vocab), crashing during g\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45189",
-    "created_at": "2026-04-02T13:43:29Z",
-    "deletions": 185,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45251",
+    "created_at": "2026-04-05T15:38:58Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45189/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45189",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45251/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45251",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 45189,
+    "number": 45251,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add doc test CI workflow reusing existing model job infrastructure",
-    "updated_at": "2026-04-03T07:30:17Z"
+    "state": "closed",
+    "title": "fix(generation): beam sample when num_beams * vocab_size exceeds multinomial limit",
+    "updated_at": "2026-04-08T14:50:40Z"
   },
   {
-    "additions": 5,
-    "author": "SunMarc",
+    "additions": 16,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do ? This PR fixes the `test_register_result_handler`. Not sure how it passed in the past when i added it but since CB returns `generated_tokens` from the same list to avoid copy, len(results[i].generated_tokens) for i\u2026",
+    "body_excerpt": "# What does this PR do? PR #42428 change the way to enable / disable torch's TF32 using torch new API. It turns out set > torch.backends.fp32_precision = False would still have > torch.backends.cudnn.conv.fp32_precision = \"tf32\" > torch.ba\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45188",
-    "created_at": "2026-04-02T13:15:57Z",
-    "deletions": 8,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45248",
+    "created_at": "2026-04-05T07:51:44Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45188/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45188",
+    "files_url": "https://github.com/huggingface/transformers/pull/45248/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45248",
     "labels": [],
     "merged": true,
-    "number": 45188,
-    "review_comments_count": 0,
+    "number": 45248,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "fix `test_register_result_handler`",
-    "updated_at": "2026-04-03T09:37:31Z"
+    "title": "Fix tf32 issue: set `torch.backends.cudnn.conv.fp32_precision` explicitly.",
+    "updated_at": "2026-04-10T21:48:56Z"
   },
   {
     "additions": 3,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Close file handler.",
+    "author": "gagandhakrey",
+    "author_association": "NONE",
+    "body_excerpt": "\u2026 validation # What does this PR do? Problem The invert_attention_mask function in src/transformers/modeling_utils.py crashed with an UnboundLocalError when given an encoder_attention_mask shape that wasn't exactly 2D or 3D. Because it onl\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45187",
-    "created_at": "2026-04-02T13:10:31Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45247",
+    "created_at": "2026-04-05T03:45:23Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45187/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45187",
+    "files_url": "https://github.com/huggingface/transformers/pull/45247/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45247",
     "labels": [],
-    "merged": true,
-    "number": 45187,
+    "merged": false,
+    "number": 45247,
     "review_comments_count": 1,
     "state": "closed",
-    "title": "Close file handler",
-    "updated_at": "2026-04-10T11:31:40Z"
+    "title": "Fix UnboundLocalError in invert_attention_mask by adding proper shape\u2026",
+    "updated_at": "2026-04-09T14:11:43Z"
   },
   {
-    "additions": 7248,
-    "author": "zucchini-nlp",
+    "additions": 0,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Same as https://github.com/huggingface/transformers/pull/40962 but cleans up the code to match transformers API. Couldn't test due to errors, the integration test is failing atm. Still need to clean the testing file\u2026",
-    "changed_files": 21,
+    "body_excerpt": "# What does this PR do? Let's CI go great!!!!",
+    "changed_files": 0,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45186",
-    "created_at": "2026-04-02T12:29:46Z",
-    "deletions": 2,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45244",
+    "created_at": "2026-04-04T18:52:19Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45186/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45186",
+    "files_url": "https://github.com/huggingface/transformers/pull/45244/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45244",
     "labels": [],
     "merged": false,
-    "number": 45186,
+    "number": 45244,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add new model: Isaac ",
-    "updated_at": "2026-04-02T12:55:56Z"
+    "title": "Let's CI go great",
+    "updated_at": "2026-04-04T19:59:28Z"
   },
   {
-    "additions": 57,
-    "author": "zucchini-nlp",
+    "additions": 2,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? If we have videos, the token type ids will be `2` but the current fn checks only image token types. This PR generalizes it rely only on `vision_group_ids` instead of token types",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Use torch 2.11 for our (daily) CI since it's released for 2 weeks already. For CircleCI, we need to fix something regarding `torchvision.io.read_video`. For daily CI, torch 2.11 doesn't cause issues (for those `torc\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45185",
-    "created_at": "2026-04-02T11:35:53Z",
-    "deletions": 104,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45243",
+    "created_at": "2026-04-04T18:09:27Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45185/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45185",
+    "files_url": "https://github.com/huggingface/transformers/pull/45243/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45243",
     "labels": [],
     "merged": true,
-    "number": 45185,
-    "review_comments_count": 2,
+    "number": 45243,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Generalize gemma vision mask to videos",
-    "updated_at": "2026-04-02T13:15:46Z"
+    "title": "Nvidia CI with `torch 2.11`",
+    "updated_at": "2026-04-04T18:48:45Z"
   },
   {
-    "additions": 445,
-    "author": "remi-or",
+    "additions": 523,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# Summary This PR adds CPU offloading to continuous batching. It's in raft until perf and test status are reported. When the GPU KV cache is full and a request must be evicted, we check if there is enough VRAM to copy the request's KV cach\u2026",
-    "changed_files": 7,
+    "body_excerpt": "# What does this PR do? After the series of fixes in other previous PRs, we can now update the tiny model creation script. This update makes the script running without any failure, just 10 warnings. There are many # TODO, some of them may\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45184",
-    "created_at": "2026-04-02T10:12:00Z",
-    "deletions": 56,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45184/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45184",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45241",
+    "created_at": "2026-04-04T12:30:35Z",
+    "deletions": 164,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45241/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45241",
     "labels": [],
-    "merged": false,
-    "number": 45184,
+    "merged": true,
+    "number": 45241,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[CB] [Major] Add CPU request offloading",
-    "updated_at": "2026-04-15T04:26:39Z"
+    "state": "closed",
+    "title": "Update tiny model creation script",
+    "updated_at": "2026-04-04T17:19:34Z"
   },
   {
-    "additions": 232,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The `transformers` CLI currently does a lot of work before command dispatch. In particular, the top-level entrypoint eagerly imports `transformers` and CLI subcommands with heavy dependencies, so even simple invocat\u2026",
-    "changed_files": 6,
+    "additions": 3,
+    "author": "shhKnight30",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Type checking for `PreTrainedConfig` subclasses broke in v5.4.0 and this fixes it. The culprit is `wrap_init_to_accept_kwargs` \u2014 it swaps out the dataclass-generated `__init__` with a `(**kwargs: Any)` wrapper at r\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45181",
-    "created_at": "2026-04-02T08:03:40Z",
-    "deletions": 23,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45181/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45181",
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45240",
+    "created_at": "2026-04-04T10:29:57Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45240/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45240",
     "labels": [],
-    "merged": false,
-    "number": 45181,
+    "merged": true,
+    "number": 45240,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Make the cli a top-level package",
-    "updated_at": "2026-04-09T12:01:41Z"
+    "state": "closed",
+    "title": "fix: restore mypy type checking for PreTrainedConfig subclasses (#45071)",
+    "updated_at": "2026-04-10T11:10:27Z"
   },
   {
-    "additions": 48,
-    "author": "paulinebm",
+    "additions": 28,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "## \ud83d\udd12 Pin GitHub Actions to commit SHAs This PR pins all GitHub Actions to their exact commit SHA instead of mutable tags or branch names. **Why?** Pinning to a SHA prevents supply chain attacks where a tag (e.g. `v4`) could be moved to poi\u2026",
-    "changed_files": 18,
+    "body_excerpt": "# What does this PR do? We have introduced `CausalLMModelTest` for some time, but haven't update `get_test_info.py` accordingly, which causes some issues, in particularly for tiny model creation, regarding the part of the attribute `all_mo\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45180",
-    "created_at": "2026-04-02T08:00:02Z",
-    "deletions": 48,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45238",
+    "created_at": "2026-04-04T07:25:15Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45180/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45180",
+    "files_url": "https://github.com/huggingface/transformers/pull/45238/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45238",
     "labels": [],
     "merged": true,
-    "number": 45180,
+    "number": 45238,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "\ud83d\udd12 Pin GitHub Actions to commit SHAs",
-    "updated_at": "2026-04-02T09:12:55Z"
+    "title": "Update `get_test_info.py` (related to tiny model creation)",
+    "updated_at": "2026-04-04T07:40:21Z"
   },
   {
-    "additions": 327,
-    "author": "remi-or",
-    "author_association": "MEMBER",
-    "body_excerpt": "## Summary This PR ads minor changes to `cache.update`, updates the memory handler with all new features and refactors a few parts of the code to make it more readable. Cache indexing: - Replace fancy indexing (cache[idx, :, :]) with expli\u2026",
-    "changed_files": 6,
+    "additions": 2,
+    "author": "KoichiYasuoka",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes the bug in `resize_token_embeddings` (may occur python3.10 or after). Quick reproduce: ``` from transformers import AutoModelForMaskedLM mdl = AutoModelForMaskedLM.from_pretrained(\"bert-base-uncased\") f = mdl.\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45179",
-    "created_at": "2026-04-02T06:15:08Z",
-    "deletions": 288,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45236",
+    "created_at": "2026-04-04T07:05:53Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45179/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45179",
+    "files_url": "https://github.com/huggingface/transformers/pull/45236/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45236",
     "labels": [],
-    "merged": true,
-    "number": 45179,
-    "review_comments_count": 3,
+    "merged": false,
+    "number": 45236,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[CB] Tweaks to update and minor fixes",
-    "updated_at": "2026-04-03T09:36:51Z"
+    "title": "resize_token_embeddings does not resize lm_head",
+    "updated_at": "2026-04-07T00:55:39Z"
   },
   {
-    "additions": 4966,
-    "author": "masoudpz",
+    "additions": 102,
+    "author": "kallewoof",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds a tiny \"agnostic.gpu\" utility that is meant to allow easy replacing of unnecessarily hard-coded vendor-specific code. The code does not use `torch.accelerator` as it is still considered experimental, bu\u2026",
+    "changed_files": 8,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45235",
+    "created_at": "2026-04-04T06:08:22Z",
+    "deletions": 13,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45235/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45235",
+    "labels": [],
+    "merged": false,
+    "number": 45235,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "feat/rfc/poc: Agnostic GPU",
+    "updated_at": "2026-04-09T14:34:20Z"
+  },
+  {
+    "additions": 306,
+    "author": "pdufour",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 20,
+    "body_excerpt": "# What does this PR do? This fixes an issues with the model that made it incompatible with exporting to onnx. Specifically the following has been changed: 1. if condition ``` if input_len < context_len: ``` Will give this error when you tr\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45176",
-    "created_at": "2026-04-02T00:47:45Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45233",
+    "created_at": "2026-04-04T00:15:04Z",
+    "deletions": 98,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45176/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45176",
+    "files_url": "https://github.com/huggingface/transformers/pull/45233/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45233",
     "labels": [],
     "merged": false,
-    "number": 45176,
-    "review_comments_count": 0,
+    "number": 45233,
+    "review_comments_count": 4,
     "state": "open",
-    "title": "added efficietvitsam model to HF",
-    "updated_at": "2026-04-02T16:01:34Z"
+    "title": "feat: make timesfm2_5 onnx export compatible",
+    "updated_at": "2026-04-10T23:37:52Z"
   },
   {
-    "additions": 49,
+    "additions": 439,
     "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "updates the docs with changes from #44796: - added section for `--compile` in the serve optimization docs - added section for `--model-timeout` in the Loading models section (useful when a model is silently kicked off and a user doesn't kn\u2026",
-    "changed_files": 2,
+    "body_excerpt": "adds docs for static model rules so model contributors are aware of expectations > I wonder if it wouldn't make sense to auto generate that portion of the doc automatically added from @tarekziade feedback: - reformats `format_rule_details(\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45174",
-    "created_at": "2026-04-01T23:29:39Z",
-    "deletions": 21,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45232",
+    "created_at": "2026-04-03T22:41:22Z",
+    "deletions": 146,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45174/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45174",
+    "files_url": "https://github.com/huggingface/transformers/pull/45232/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45232",
     "labels": [],
     "merged": true,
-    "number": 45174,
-    "review_comments_count": 2,
+    "number": 45232,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "[docs] transformers serve",
-    "updated_at": "2026-04-02T16:39:12Z"
+    "title": "[docs] static model rules",
+    "updated_at": "2026-04-08T15:06:17Z"
   },
   {
-    "additions": 16,
-    "author": "JJJYmmm",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixed the `qwen3_5` / `qwen3_5_moe` reverse-loading tests by correcting the text model type used in the setup, and aligned the reverse-mapping behavior with gemma3n since they are all native multimodal. This also re\u2026",
-    "changed_files": 2,
+    "additions": 20,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? All of this are trivial ... (maybe except \"evolla\")",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45173",
-    "created_at": "2026-04-01T20:20:37Z",
-    "deletions": 90,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45228",
+    "created_at": "2026-04-03T18:17:29Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45173/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45173",
+    "files_url": "https://github.com/huggingface/transformers/pull/45228/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45228",
     "labels": [],
     "merged": true,
-    "number": 45173,
+    "number": 45228,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[misc] fix qwen35 tests: correct the text model type and skip reverse_mapping",
-    "updated_at": "2026-04-02T13:05:39Z"
+    "title": "More fix for tiny model creation",
+    "updated_at": "2026-04-03T19:25:51Z"
   },
   {
-    "additions": 156,
-    "author": "ezylopx5",
+    "additions": 3,
+    "author": "akhilc08",
     "author_association": "NONE",
-    "body_excerpt": "## Problem Transformers currently provides sampling filters such as top-k, top-p, min-p, and top-h, but does not include top-n-sigma sampling from \"Top-n\u03c3: Not All Logits Are You Need\". This makes it harder to use a temperature-invariant t\u2026",
-    "changed_files": 8,
+    "body_excerpt": "## Summary - Remove the unused `PILImageResampling` runtime import from `video_processing_utils.py` which causes an `ImportError` when Pillow is not installed - Also remove the now-unused `is_vision_available` import that only guarded the\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45172",
-    "created_at": "2026-04-01T19:25:12Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45227",
+    "created_at": "2026-04-03T17:44:25Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45172/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45172",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/45227/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45227",
+    "labels": [],
     "merged": false,
-    "number": 45172,
+    "number": 45227,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add TopNSigmaLogitsWarper and top_n_sigma generation config support",
-    "updated_at": "2026-04-04T06:24:26Z"
+    "title": "fix: remove nonexistent PILImageResampling import from video_processing_utils",
+    "updated_at": "2026-04-05T17:04:49Z"
   },
   {
-    "additions": 113,
-    "author": "Kash6",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "When input_boxes contains a mix of None and real box entries (e.g. input_boxes=[None, [[x1,y1,x2,y2]]]), the processor pads None entries with [-10,-10,0,0] but does not generate corresponding input_boxes_labels. The model's geometry encode\u2026",
-    "changed_files": 2,
+    "additions": 3,
+    "author": "akhilc08",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Fixes an `IndexError: string index out of range` crash in `_split_tokens_on_unicode()` when the decoded token stream ends with a dangling Unicode replacement character (U+FFFD) - Adds a bounds check so that when `unicode_offse\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45171",
-    "created_at": "2026-04-01T18:26:48Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45226",
+    "created_at": "2026-04-03T17:43:21Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45171/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45171",
+    "files_url": "https://github.com/huggingface/transformers/pull/45226/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45226",
     "labels": [],
-    "merged": true,
-    "number": 45171,
+    "merged": false,
+    "number": 45226,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Sam3Processor missing input_boxes_labels for padded None entries",
-    "updated_at": "2026-04-15T15:22:19Z"
+    "title": "fix: handle trailing replacement character in Whisper word timestamp decoding",
+    "updated_at": "2026-04-05T17:04:49Z"
   },
   {
-    "additions": 19,
-    "author": "zucchini-nlp",
+    "additions": 6,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Before I forget, opening a PR. Will cause conflicts in https://github.com/huggingface/transformers/pull/44431, so I will better merge this after refactoring",
-    "changed_files": 10,
+    "body_excerpt": "# What does this PR do? The `dev` extra now indirectly pulls hf-doc-builder so the install step failed. We also need to update to current main for the latest features",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45170",
-    "created_at": "2026-04-01T17:50:33Z",
-    "deletions": 17,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45225",
+    "created_at": "2026-04-03T16:46:54Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45170/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45170",
+    "files_url": "https://github.com/huggingface/transformers/pull/45225/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45225",
     "labels": [],
-    "merged": false,
-    "number": 45170,
+    "merged": true,
+    "number": 45225,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "`layrnorm` -> `layernorm`",
-    "updated_at": "2026-04-01T18:03:31Z"
+    "state": "closed",
+    "title": "fix: hf-doc-builder insallation was failing",
+    "updated_at": "2026-04-03T17:02:09Z"
   },
   {
-    "additions": 10,
-    "author": "hmellor",
+    "additions": 2,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "In https://github.com/huggingface/transformers/pull/45094 I introduced some errors to the remote code resolution when trying to detect if local code belonged to Transformers or not. These tests were: ```bash pytest tests/models/cohere_asr/\u2026",
+    "body_excerpt": "# What does this PR do? Fix for tiny model",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45169",
-    "created_at": "2026-04-01T15:27:43Z",
-    "deletions": 6,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45224",
+    "created_at": "2026-04-03T16:39:23Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45169/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45169",
+    "files_url": "https://github.com/huggingface/transformers/pull/45224/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45224",
     "labels": [],
     "merged": true,
-    "number": 45169,
+    "number": 45224,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix explicit local code resolution for tokenizers and image processors",
-    "updated_at": "2026-04-01T21:48:02Z"
+    "title": "remove unnecessary entries in some auto model mappings",
+    "updated_at": "2026-04-03T17:26:24Z"
   },
   {
-    "additions": 4,
-    "author": "w601sxs",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Based on our experimentation min and max lr for LLMs need to be set properly as defaults. Please refer to paper. For the broader community 1e-7 to 1e-4 are decent defaults # What does this PR do? <!-- Congratulations! You've made it this f\u2026",
+    "additions": 17,
+    "author": "Talhax55z",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #31356 ## What does this PR do? The `postprocess` method in `ObjectDetectionPipeline` was hardcoding `raw_annotations[0]`, which caused batch inference to only return results for the first image, ignoring all others. This PR replaces\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45168",
-    "created_at": "2026-04-01T15:02:15Z",
-    "deletions": 4,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45223",
+    "created_at": "2026-04-03T16:36:25Z",
+    "deletions": 15,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45168/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45168",
+    "files_url": "https://github.com/huggingface/transformers/pull/45223/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45223",
     "labels": [],
     "merged": false,
-    "number": 45168,
+    "number": 45223,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Update min_lr and max_lr default values to better defaults",
-    "updated_at": "2026-04-08T16:58:59Z"
+    "state": "closed",
+    "title": "Fix: ObjectDetectionPipeline batch inference only returns first image results",
+    "updated_at": "2026-04-13T15:06:32Z"
   },
   {
-    "additions": 16,
-    "author": "xu-song",
+    "additions": 12,
+    "author": "jashshah999",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds support for [Anthropic's JSON function style](https://platform.claude.com/docs/en/agents-and-tools/tool-use/define-tools): `{\"name\": \"...\", \"description\": \"...\", \"input_schema\": {...}}` ## Usage Example\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary When using Gemma 3 or Gemma 4 for text-only supervised fine-tuning (no images), the forward pass raises a `ValueError` because `token_type_ids` / `mm_token_type_ids` is not provided. This happens because `AutoTokenizer` does not\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45167",
-    "created_at": "2026-04-01T14:50:33Z",
-    "deletions": 6,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45222",
+    "created_at": "2026-04-03T16:27:31Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45167/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45167",
+    "files_url": "https://github.com/huggingface/transformers/pull/45222/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45222",
     "labels": [],
     "merged": false,
-    "number": 45167,
+    "number": 45222,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add anthropic style of function schema",
-    "updated_at": "2026-04-03T05:49:56Z"
+    "state": "closed",
+    "title": "fix(gemma3, gemma4): default token_type_ids to zeros for text-only training",
+    "updated_at": "2026-04-15T15:31:05Z"
   },
   {
-    "additions": 67,
-    "author": "Rocketknight1",
+    "additions": 6,
+    "author": "eustlb",
     "author_association": "MEMBER",
-    "body_excerpt": "We didn't think we needed them, but I think we do after all!",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? As per title",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45166",
-    "created_at": "2026-04-01T14:46:33Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45221",
+    "created_at": "2026-04-03T15:49:24Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45166/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45166",
+    "files_url": "https://github.com/huggingface/transformers/pull/45221/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45221",
     "labels": [],
-    "merged": true,
-    "number": 45166,
+    "merged": false,
+    "number": 45221,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Re-add regex substitutions to the response parsing spec",
-    "updated_at": "2026-04-01T15:46:34Z"
+    "state": "open",
+    "title": "user friendly error when loading audio from video",
+    "updated_at": "2026-04-09T14:16:24Z"
   },
   {
-    "additions": 11,
-    "author": "zucchini-nlp",
+    "additions": 1711,
+    "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Emu3 was not updated in recent refactor and blip files were swapped. This PR fixes it Do we need anything to support BC importing from old files, or does it happen in `LazyImports` @yonigozlan ?",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? This PR adds transformers serve compatibility to multimodal models like qwen omni or gemma 4. We add support for audio with chat completion and response though `input_audio` -> the client need to `base64-encode` the\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45165",
-    "created_at": "2026-04-01T14:18:38Z",
-    "deletions": 6,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45220",
+    "created_at": "2026-04-03T14:16:33Z",
+    "deletions": 163,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45165/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45165",
+    "files_url": "https://github.com/huggingface/transformers/pull/45220/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45220",
     "labels": [],
     "merged": true,
-    "number": 45165,
-    "review_comments_count": 0,
+    "number": 45220,
+    "review_comments_count": 18,
     "state": "closed",
-    "title": "Fix missing image processors backends",
-    "updated_at": "2026-04-07T13:46:24Z"
+    "title": "Multimodal serve support ",
+    "updated_at": "2026-04-15T14:13:02Z"
   },
   {
-    "additions": 1,
-    "author": "albertvillanova",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix TypeError: 'NoneType' object is not iterable in `GenerationMixin.generate` - Fix for None layer_types <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is goin\u2026",
+    "additions": 3,
+    "author": "sywangyi",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? google/gemma-4-26B-A4B-it tp 2, memory is 46G per rank wo the change, drop to about 25G w per rank with the change - text models: @ArthurZucker @Cyrilvallez",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45164",
-    "created_at": "2026-04-01T13:53:41Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45219",
+    "created_at": "2026-04-03T13:59:02Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45164/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45164",
+    "files_url": "https://github.com/huggingface/transformers/pull/45219/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45219",
     "labels": [],
     "merged": true,
-    "number": 45164,
+    "number": 45219,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix TypeError: 'NoneType' object is not iterable in GenerationMixin.generate",
-    "updated_at": "2026-04-03T04:57:18Z"
+    "title": "Add MoE to Gemma4 TP plan",
+    "updated_at": "2026-04-08T14:09:40Z"
   },
   {
-    "additions": 131,
-    "author": "tarekziade",
+    "additions": 4057,
+    "author": "LysandreJik",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Makes sure the full error is displayed on errors",
-    "changed_files": 2,
+    "body_excerpt": "This PR offers a new, Agentic surface for transformers. It tries to apply what is done elsewhere with CLIs to `transformers`, leveraging many current use-cases of `transformers` and exposing them as CLI endpoints. I recommend reading this\u2026",
+    "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45163",
-    "created_at": "2026-04-01T13:41:18Z",
-    "deletions": 20,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45163/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45163",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45218",
+    "created_at": "2026-04-03T13:31:08Z",
+    "deletions": 0,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45218/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45218",
     "labels": [],
-    "merged": true,
-    "number": 45163,
+    "merged": false,
+    "number": 45218,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "tweak checkers output on errors",
-    "updated_at": "2026-04-09T11:50:16Z"
+    "state": "open",
+    "title": "Proposal: Agent-first CLI",
+    "updated_at": "2026-04-13T13:03:00Z"
   },
   {
-    "additions": 513,
-    "author": "onwp",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Description Re-creates the Turkish documentation that was accidentally removed during the TF/Flax cleanup (commit fce74651). This PR adds the foundational Turkish docs with the complete \"Get Started\" section. ### Files added - `docs/sou\u2026",
-    "changed_files": 6,
+    "additions": 5,
+    "author": "ENg-122",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Remove unnecessary masked_fill_(mask, 0) call in torch_chunk_gated_delta_rule. The decay_mask computed earlier already encodes the causal/lower-triangular structure (upper-triangle values are zero), so masking the a\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45159",
-    "created_at": "2026-04-01T05:15:32Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45215",
+    "created_at": "2026-04-03T09:08:28Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45159/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45159",
+    "files_url": "https://github.com/huggingface/transformers/pull/45215/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45215",
     "labels": [],
-    "merged": false,
-    "number": 45159,
+    "merged": true,
+    "number": 45215,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add Turkish documentation: Get Started section",
-    "updated_at": "2026-04-01T05:16:49Z"
+    "title": "[Qwen3_5]Remove unnecessary masked_fill_ in torch_chunk_gated_delta_rule attention computation: \"attn = (q_i @ k_i.transpose(-1, -2) * decay_mask[:, :, i]).masked_fill_(mask, 0)\"",
+    "updated_at": "2026-04-09T14:12:51Z"
   },
   {
-    "additions": 525,
-    "author": "onwp",
+    "additions": 45,
+    "author": "kaixuanliu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary This PR re-creates the Turkish (`tr`) documentation for the Transformers library, starting with the **Get Started** section. The original Turkish translation was accidentally removed in commit fce74651 (#40999). This contributio\u2026",
-    "changed_files": 6,
+    "body_excerpt": "This PR fixes failed test case: `tests/models/cohere_asr/test_modeling_cohere_asr.py::CohereAsrModelTest::test_model_parallel_beam_search`, and add some adjustment to make the test cases pass for Intel XPU device. @ydshieh pls help review,\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45158",
-    "created_at": "2026-04-01T05:05:40Z",
-    "deletions": 1,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45214",
+    "created_at": "2026-04-03T08:32:34Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45158/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45158",
+    "files_url": "https://github.com/huggingface/transformers/pull/45214/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45214",
     "labels": [],
     "merged": true,
-    "number": 45158,
-    "review_comments_count": 0,
+    "number": 45214,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Add Turkish (tr) translation for Get Started section",
-    "updated_at": "2026-04-02T17:50:46Z"
+    "title": "cohere_asr: fix bug for model_parallel_beam_search test case",
+    "updated_at": "2026-04-13T02:40:29Z"
   },
   {
-    "additions": 190,
-    "author": "Qubitium",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Add PrismML 1bit (gguf based, group-size 128) model inference suppport. ## Code Agent Policy - [ ] I confirm that this is not a pure code agent PR. ## Before submitting - [ ] This PR fixes a typo or improves the doc\u2026",
-    "changed_files": 4,
+    "additions": 6052,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "comparison https://github.com/huggingface/transformers/blob/937d61b9fa00001da1a0680ecf8061b5990fbcd7/sarvam_moe_comparison.md # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once mer\u2026",
+    "changed_files": 26,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45157",
-    "created_at": "2026-04-01T03:11:51Z",
-    "deletions": 23,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45213",
+    "created_at": "2026-04-03T08:25:49Z",
+    "deletions": 209,
     "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45157/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45157",
+    "files_url": "https://github.com/huggingface/transformers/pull/45213/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45213",
     "labels": [],
     "merged": false,
-    "number": 45157,
+    "number": 45213,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[WIP] PrismML Bonsai model support",
-    "updated_at": "2026-04-02T20:53:12Z"
+    "title": "DO NOT MERGE - model creation skill",
+    "updated_at": "2026-04-03T12:26:09Z"
   },
   {
-    "additions": 50,
-    "author": "Cursx",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? After the `merge_and_unload()` operation in PEFT, embed_tokens and lm_head become independent tensors with different values, but config.tie_word_embeddings remains True. The load-side already detects this using torc\u2026",
+    "additions": 29,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@ydshieh pls help review, thx!",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45156",
-    "created_at": "2026-04-01T02:36:38Z",
-    "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45156/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45156",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45212",
+    "created_at": "2026-04-03T07:44:35Z",
+    "deletions": 6,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45212/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45212",
     "labels": [],
-    "merged": false,
-    "number": 45156,
+    "merged": true,
+    "number": 45212,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix save_pretrained writing incorrect tie_word_embeddings=True config after PEFT merge",
-    "updated_at": "2026-04-01T09:34:05Z"
+    "title": "musicflamingo: add test support for Intel XPU device",
+    "updated_at": "2026-04-10T10:41:20Z"
   },
   {
-    "additions": 46,
-    "author": "michaelbenayoun",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Add support for loading adapters with `PreTrainedModel.load_adapter` when using TP. See: https://github.com/huggingface/peft/pull/3096",
-    "changed_files": 1,
+    "additions": 4,
+    "author": "matdou",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #45208 # What does this PR do? This PR corrects an incorrect return type in `Qwen3MoeSparseMoeBlock.forward`. The method was annotated as returning `tuple[torch.Tensor, torch.Tensor]`, while the implementation returns a `torch.Tensor\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45155",
-    "created_at": "2026-03-31T22:23:06Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45211",
+    "created_at": "2026-04-03T07:44:32Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45155/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45155",
+    "files_url": "https://github.com/huggingface/transformers/pull/45211/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45211",
     "labels": [],
-    "merged": true,
-    "number": 45155,
-    "review_comments_count": 2,
+    "merged": false,
+    "number": 45211,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Load adapter with TP",
-    "updated_at": "2026-04-09T22:35:45Z"
+    "title": "[Qwen3MoE] Fix wrong return type annotation in Qwen3MoeSparseMoeBlock.forward",
+    "updated_at": "2026-04-08T19:20:49Z"
   },
   {
-    "additions": 676,
-    "author": "FaizanImran-blip",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixed bug #45072 / #45071 where PretrainedConfig type checking and Pydantic model field validation were broken in v5.4.0. Added proper type checking for 'num_labels'. Added unit tests in test.py to verify correct and incorrect types. Verif\u2026",
-    "changed_files": 5,
+    "additions": 11,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Commits that got in the release branch to allow pushing",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45154",
-    "created_at": "2026-03-31T20:28:32Z",
-    "deletions": 153,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45154/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45154",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 45154,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45210",
+    "created_at": "2026-04-03T05:42:21Z",
+    "deletions": 4,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45210/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45210",
+    "labels": [],
+    "merged": true,
+    "number": 45210,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Pretrained-config bug(45072/huggingfacebug)",
-    "updated_at": "2026-04-09T11:43:55Z"
+    "title": "Fix pypi release",
+    "updated_at": "2026-04-03T06:40:39Z"
   },
   {
-    "additions": 502,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "As per title, with torch releasing the varlen API, we can somewhat use native FA (with limited feature support) Restrictions - Unsupported features - Dropout - Learnable sinks (attention sinks) - Determinism - Softcap - CB KV cache native\u2026",
-    "changed_files": 19,
+    "additions": 3,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@ydshieh pls help review, thx!",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45153",
-    "created_at": "2026-03-31T19:43:19Z",
-    "deletions": 386,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45153/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45153",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45209",
+    "created_at": "2026-04-03T05:40:36Z",
+    "deletions": 3,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45209/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45209",
     "labels": [],
-    "merged": false,
-    "number": 45153,
-    "review_comments_count": 25,
-    "state": "open",
-    "title": "[`FA`] Native torch integration",
-    "updated_at": "2026-04-01T20:01:16Z"
+    "merged": true,
+    "number": 45209,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "nomic_bert: make the test suitable for general device.",
+    "updated_at": "2026-04-13T02:40:26Z"
   },
   {
-    "additions": 230,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "refactors the testing docs to be more contributor-facing organized around writing model tests instead of being a collection of pytest usage examples and CI maintenance. also updates the pr checks doc so contributors are better prepared to\u2026",
-    "changed_files": 3,
+    "additions": 65,
+    "author": "w4nderlust",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fixes #45206 ## What does this PR do? Adds documentation for the Gemma4 Per-Layer Embeddings (PLE) system, which is currently pretty hard to reverse-engineer from the code alone. I ran into this while implementing Gemma4 inference from scr\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45152",
-    "created_at": "2026-03-31T18:12:09Z",
-    "deletions": 1330,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45207",
+    "created_at": "2026-04-03T05:15:47Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45152/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45152",
+    "files_url": "https://github.com/huggingface/transformers/pull/45207/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45207",
     "labels": [],
-    "merged": false,
-    "number": 45152,
+    "merged": true,
+    "number": 45207,
     "review_comments_count": 6,
-    "state": "open",
-    "title": "[docs] model testing",
-    "updated_at": "2026-04-15T16:06:59Z"
+    "state": "closed",
+    "title": "[Gemma4] Add docstrings for Per-Layer Embeddings (PLE) pipeline",
+    "updated_at": "2026-04-14T20:03:58Z"
   },
   {
-    "additions": 1,
-    "author": "maanas1234",
+    "additions": 112,
+    "author": "kaixuanliu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? The previous code snippet had an error in how `TrainingArguments` was being used, which could lead to confusion or incorrect implementation. Corrected the example to properly demonstrate how to initialize and use `T\u2026",
-    "changed_files": 1,
+    "body_excerpt": "@ydshieh pls help review, thx!",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45150",
-    "created_at": "2026-03-31T17:31:37Z",
-    "deletions": 1,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45204",
+    "created_at": "2026-04-03T02:32:39Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45150/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45150",
+    "files_url": "https://github.com/huggingface/transformers/pull/45204/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45204",
     "labels": [],
     "merged": true,
-    "number": 45150,
-    "review_comments_count": 1,
+    "number": 45204,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix incorrect TrainingArguments example in training.md",
-    "updated_at": "2026-03-31T18:31:25Z"
+    "title": "fix bug for videomt model device mismatch",
+    "updated_at": "2026-04-13T02:40:27Z"
   },
   {
-    "additions": 4633,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "First pass: ~1M+ tokens in, ~115K+ out, Opus mainly, $42, 1h30 # PR #44320 vs Our Implementation ## What we got right - Same overall structure: modular file + generated standalone files + conversion script + tests + docs - Same model direc\u2026",
-    "changed_files": 20,
+    "additions": 2,
+    "author": "Qubitium",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Disable FlashAttention support for Gemm4 which FA cannot suport due to global.head-dim=512. I am very confused at the current code/test for Gemma4. I ran real inference using transformer `main` and `fa` throws head-\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45149",
-    "created_at": "2026-03-31T15:56:11Z",
-    "deletions": 1,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45149/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45149",
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45202",
+    "created_at": "2026-04-02T23:37:22Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45202/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45202",
     "labels": [],
     "merged": false,
-    "number": 45149,
-    "review_comments_count": 0,
+    "number": 45202,
+    "review_comments_count": 4,
     "state": "open",
-    "title": "DO NOT MERGE adding SAML3-LiteText with a skill, first pass",
-    "updated_at": "2026-04-01T06:28:20Z"
+    "title": "Fix gemma4 has flash-attention incompatbile head-dim=512",
+    "updated_at": "2026-04-12T03:39:09Z"
   },
   {
-    "additions": 13,
-    "author": "HallerPatrick",
+    "additions": 18,
+    "author": "harshaljanjani",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #45146 ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by our ability to review and r\u2026",
+    "body_excerpt": "### What does this PR do? The following **Wav2Vec2PhonemeCTC** use cases were identified and fixed in this PR: \u2192 [05c0e1d (\"rm slow tokenizers\")](https://github.com/huggingface/transformers/pull/40936) added [self.backend = kwargs.pop(\"bac\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45148",
-    "created_at": "2026-03-31T15:20:02Z",
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45199",
+    "created_at": "2026-04-02T20:03:22Z",
     "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45148/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45148",
+    "files_url": "https://github.com/huggingface/transformers/pull/45199/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45199",
     "labels": [],
-    "merged": false,
-    "number": 45148,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 45199,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Allow for all layers in Qwen3.5 architecture to be Gated Deltanet.",
-    "updated_at": "2026-04-02T11:18:12Z"
+    "title": "fix(models): Resolve regressions in Wav2Vec2PhonemeCTCTokenizer (wav2vec2-lv-60-espeak-cv-ft)",
+    "updated_at": "2026-04-14T14:00:01Z"
   },
   {
-    "additions": 309,
-    "author": "mobicham",
+    "additions": 104,
+    "author": "douglas-reid",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR fixes hqq support that has been broken for a couple of months now after a refactoring: * Online quantization works fine now. * Serialization to load/save HQQ models is fixed too. ## Code Agent Policy - [x] I\u2026",
-    "changed_files": 3,
+    "body_excerpt": "Fixes issues with the Gemma 4 model docs. Mainly, this is updating examples to point at the actual models, with FC and Audio examples added. - [ X ] I confirm that this is not a pure code agent PR. ## Before submitting - [ X ] This PR fixe\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45147",
-    "created_at": "2026-03-31T14:59:47Z",
-    "deletions": 12,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45197",
+    "created_at": "2026-04-02T19:05:35Z",
+    "deletions": 42,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45147/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45147",
+    "files_url": "https://github.com/huggingface/transformers/pull/45197/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45197",
     "labels": [],
-    "merged": false,
-    "number": 45147,
-    "review_comments_count": 8,
-    "state": "open",
-    "title": "Fix broken HQQ support",
-    "updated_at": "2026-04-09T16:14:06Z"
+    "merged": true,
+    "number": 45197,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix(docs): correct gemma4 docs and examples",
+    "updated_at": "2026-04-02T22:23:16Z"
   },
   {
-    "additions": 2150,
-    "author": "casinca",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Hello, this PR aims to add the MiMo-V2-Flash model to the Transformers library Fixes https://github.com/huggingface/transformers/issues/42954 MiMo-V2 is \"The last of the SOTAs\" that isn't natively supported by the T\u2026",
-    "changed_files": 12,
+    "additions": 1,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "fixes `<hfoptions>` tag",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45144",
-    "created_at": "2026-03-31T13:53:28Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45196",
+    "created_at": "2026-04-02T18:02:44Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45144/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45144",
-    "labels": [
-      "New model"
-    ],
-    "merged": false,
-    "number": 45144,
-    "review_comments_count": 30,
-    "state": "open",
-    "title": "Add Xiaomi MiMo-V2",
-    "updated_at": "2026-04-15T18:35:49Z"
+    "files_url": "https://github.com/huggingface/transformers/pull/45196/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45196",
+    "labels": [],
+    "merged": true,
+    "number": 45196,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[docs] formatting",
+    "updated_at": "2026-04-03T09:37:52Z"
   },
   {
-    "additions": 102,
-    "author": "Rocketknight1",
+    "additions": 90,
+    "author": "yonigozlan",
     "author_association": "MEMBER",
-    "body_excerpt": "This PR adds `parse_response` to Processor classes by wrapping the `Tokenizer` method! cc @zucchini-nlp",
-    "changed_files": 6,
+    "body_excerpt": "# What this PR does Adds a new `load_image_as_tensor` utility leveraging torchvision's `decode_image` to `image_utils.py` and overrides `fetch_images` in `TorchvisionBackend` to use it. Previously, all image loading went through PIL regard\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45143",
-    "created_at": "2026-03-31T13:11:49Z",
-    "deletions": 9,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45195",
+    "created_at": "2026-04-02T17:58:51Z",
+    "deletions": 20,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45143/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45143",
+    "files_url": "https://github.com/huggingface/transformers/pull/45195/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45195",
     "labels": [],
     "merged": true,
-    "number": 45143,
-    "review_comments_count": 4,
+    "number": 45195,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "Add parse_response to Processor, make it a bit more official",
-    "updated_at": "2026-03-31T17:07:16Z"
+    "title": " Use torchvision `decode_image` to load images in the torchvision backend",
+    "updated_at": "2026-04-09T17:08:04Z"
   },
   {
-    "additions": 4,
-    "author": "casinca",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 2,
+    "additions": 222,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes: - Replace `PretrainedConfig` with `PreTrainedConfig` - Don't import from other models inside `configuration_model.py`, instead resolve via modular - Text/vision sub-configs are documented (only kosmos was mis\u2026",
+    "changed_files": 23,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45142",
-    "created_at": "2026-03-31T12:47:46Z",
-    "deletions": 4,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45194",
+    "created_at": "2026-04-02T16:39:06Z",
+    "deletions": 40,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45142/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45142",
+    "files_url": "https://github.com/huggingface/transformers/pull/45194/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45194",
     "labels": [],
     "merged": false,
-    "number": 45142,
-    "review_comments_count": 0,
+    "number": 45194,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "refactor(gpt-oss): rename `eager_attention_forward` to `eager_attention_forward_with_sink`",
-    "updated_at": "2026-04-02T16:44:14Z"
+    "title": "Configuration insoncistencies",
+    "updated_at": "2026-04-10T11:39:18Z"
   },
   {
-    "additions": 19,
-    "author": "ydshieh",
+    "additions": 15,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Really stupid ... Currently, change a docstring will count as \"modified files\", then the impacted files and test files to run are computed. Say, we add a comment to bert, many files impacted (via dependency) then ma\u2026",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45070 Though I am convinced that configs should not be pyndatic BaseClass, just because we already wrap all subclasses as dataclass. You never know what happe\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45140",
-    "created_at": "2026-03-31T09:41:28Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45193",
+    "created_at": "2026-04-02T15:36:56Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45140/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45140",
+    "files_url": "https://github.com/huggingface/transformers/pull/45193/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45193",
     "labels": [],
-    "merged": true,
-    "number": 45140,
+    "merged": false,
+    "number": 45193,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix stupid test fetcher",
-    "updated_at": "2026-03-31T11:06:51Z"
+    "state": "open",
+    "title": "Config can apply pyndatic validation without torch-dependence",
+    "updated_at": "2026-04-02T17:00:43Z"
   },
   {
-    "additions": 757,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? More fixes",
-    "changed_files": 99,
+    "additions": 9896,
+    "author": "RyanMullins",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "--------- # What does this PR do? model previously unable to use tools ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenec\u2026",
+    "changed_files": 41,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45139",
-    "created_at": "2026-03-31T09:26:28Z",
-    "deletions": 976,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45192",
+    "created_at": "2026-04-02T14:35:18Z",
+    "deletions": 79,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45139/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45139",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45192/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45192",
+    "labels": [
+      "New model"
+    ],
     "merged": true,
-    "number": 45139,
-    "review_comments_count": 26,
+    "number": 45192,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix vllm cis",
-    "updated_at": "2026-04-08T11:19:41Z"
+    "title": "casually dropping the most capable open weights on the planet",
+    "updated_at": "2026-04-03T12:28:06Z"
   },
   {
-    "additions": 2,
-    "author": "Abdennacer-Badaoui",
-    "author_association": "MEMBER",
-    "body_excerpt": "This is a small T5 expectations update. It is the same for both AMD and NVIDIA A10 GPUs.",
+    "additions": 13,
+    "author": "saslifat-gif",
+    "author_association": "NONE",
+    "body_excerpt": "The Qwen2 tokenizer test file had no custom test methods \u2014 only integration constants inherited from TokenizerTesterMixin. This PR adds a test documenting two untested edge cases in decode(): **Before (no test, behavior undocumented):** ``\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45138",
-    "created_at": "2026-03-31T08:41:40Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45191",
+    "created_at": "2026-04-02T14:08:15Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45138/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45138",
+    "files_url": "https://github.com/huggingface/transformers/pull/45191/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45191",
     "labels": [],
-    "merged": true,
-    "number": 45138,
+    "merged": false,
+    "number": 45191,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "CI] Small T5 expectations updated",
-    "updated_at": "2026-04-02T08:21:25Z"
+    "title": "Add edge case tests for out-of-range token id decoding in Qwen2 tokenizer",
+    "updated_at": "2026-04-08T12:18:56Z"
   },
   {
-    "additions": 83,
-    "author": "Cursx",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? This PR fixes a bug in `PreTrainedModel.save_pretrained()` where `config.tie_word_embeddings` can be inconsistent with the actual weight state, leading to silent model corruption for downstream consumers. ### Proble\u2026",
-    "changed_files": 2,
+    "additions": 92,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This patch extends ty check to src/transformers/cli. Supersedes https://github.com/huggingface/transformers/pull/44566. I've added some of the changes in this PR",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45136",
-    "created_at": "2026-03-31T06:45:53Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45190",
+    "created_at": "2026-04-02T13:54:23Z",
+    "deletions": 32,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45136/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45136",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 45136,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/45190/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45190",
+    "labels": [],
+    "merged": true,
+    "number": 45190,
+    "review_comments_count": 18,
     "state": "closed",
-    "title": "Fix #45127: Auto-fix diverged tie_word_embeddings config on save to prevent silent weight corruption",
-    "updated_at": "2026-03-31T12:21:32Z"
+    "title": "Fix ty for transformers cli",
+    "updated_at": "2026-04-14T16:53:22Z"
   },
   {
-    "additions": 44,
-    "author": "Cursx",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Fixes an issue where PEFT adapters applied independently to tied embeddings (`embed_tokens` and [lm_head](cci:1://file:///d:/transformers/transformers/src/transformers/modeling_utils.py:2858:4-2985:26)) cause silent\u2026",
-    "changed_files": 1,
+    "additions": 409,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "## Summary - Add two optional, backward-compatible inputs (`test_path_prefix`, `test_file_suffix`) to `model_jobs.yml` and `self-scheduled.yml` \u2014 defaults preserve all existing behavior - Extend the `set-matrix` step in `self-scheduled.yml\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45135",
-    "created_at": "2026-03-31T02:28:26Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45189",
+    "created_at": "2026-04-02T13:43:29Z",
+    "deletions": 185,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45135/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45135",
+    "files_url": "https://github.com/huggingface/transformers/pull/45189/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45189",
     "labels": [],
     "merged": false,
-    "number": 45135,
+    "number": 45189,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix model saving corruption for dynamically untied embeddings",
-    "updated_at": "2026-03-31T04:36:56Z"
+    "state": "open",
+    "title": "Add doc test CI workflow reusing existing model job infrastructure",
+    "updated_at": "2026-04-03T07:30:17Z"
   },
   {
-    "additions": 91,
-    "author": "milesial",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Add support for CUDA parakeet preprocessor, running STFT and mel spectrogram extraction on the GPU. This refactor also speeds up the CPU implementation. Tested on `nvidia/parakeet-ctc-0.6b`, B200, 300s audio: Before\u2026",
+    "additions": 5,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do ? This PR fixes the `test_register_result_handler`. Not sure how it passed in the past when i added it but since CB returns `generated_tokens` from the same list to avoid copy, len(results[i].generated_tokens) for i\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45134",
-    "created_at": "2026-03-31T01:59:28Z",
-    "deletions": 56,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45188",
+    "created_at": "2026-04-02T13:15:57Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45134/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45134",
+    "files_url": "https://github.com/huggingface/transformers/pull/45188/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45188",
     "labels": [],
-    "merged": false,
-    "number": 45134,
+    "merged": true,
+    "number": 45188,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Optimize Parakeet feature extraction on CUDA",
-    "updated_at": "2026-04-07T14:56:51Z"
+    "state": "closed",
+    "title": "fix `test_register_result_handler`",
+    "updated_at": "2026-04-03T09:37:31Z"
   },
   {
-    "additions": 1944,
-    "author": "itazap",
+    "additions": 3,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "## Summary - Auto-generated modular integration for `sarvam` - `modular_sarvam.py` written via HF Inference API guided by `modular_model_detector.py` - `modeling_sarvam.py` regenerated from modular via `modular_model_converter.py` ## Test\u2026",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Close file handler.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45133",
-    "created_at": "2026-03-31T00:00:15Z",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45187",
+    "created_at": "2026-04-02T13:10:31Z",
     "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45133/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45133",
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45187/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45187",
     "labels": [],
-    "merged": false,
-    "number": 45133,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Add sarvam model",
-    "updated_at": "2026-03-31T00:09:59Z"
+    "merged": true,
+    "number": 45187,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "Close file handler",
+    "updated_at": "2026-04-10T11:31:40Z"
   },
   {
-    "additions": 215,
-    "author": "akintunero",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary This PR fixes GitHub issue #45120: \"Double softmax in MoE router load-balancing loss\". MoE routers in Mixtral, Qwen2MoE, and Qwen3VLMoE were applying softmax inside forward(), then the load_balancing_loss_func applied softmax AG\u2026",
-    "changed_files": 9,
+    "additions": 7248,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Same as https://github.com/huggingface/transformers/pull/40962 but cleans up the code to match transformers API. Couldn't test due to errors, the integration test is failing atm. Still need to clean the testing file\u2026",
+    "changed_files": 21,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45132",
-    "created_at": "2026-03-30T22:45:01Z",
-    "deletions": 34,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45186",
+    "created_at": "2026-04-02T12:29:46Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45132/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45132",
+    "files_url": "https://github.com/huggingface/transformers/pull/45186/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45186",
     "labels": [],
     "merged": false,
-    "number": 45132,
+    "number": 45186,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix: Remove double softmax in MoE router load-balancing loss (Mixtral, Qwen2MoE, Qwen3VLMoE)",
-    "updated_at": "2026-03-30T23:36:57Z"
+    "state": "open",
+    "title": "Add new model: Isaac ",
+    "updated_at": "2026-04-02T12:55:56Z"
   },
   {
-    "additions": 30,
-    "author": "yacinemebarki",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes issue #45120: Several MoE routers returned softmaxed probabilities as `router_logits`, which caused `load_balancing_loss_func` to compute softmax(softmax(logits)), flattening routing distributions and weakenin\u2026",
-    "changed_files": 13,
+    "additions": 57,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? If we have videos, the token type ids will be `2` but the current fn checks only image token types. This PR generalizes it rely only on `vision_group_ids` instead of token types",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45131",
-    "created_at": "2026-03-30T21:18:47Z",
-    "deletions": 30,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45185",
+    "created_at": "2026-04-02T11:35:53Z",
+    "deletions": 104,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45131/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45131",
+    "files_url": "https://github.com/huggingface/transformers/pull/45185/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45185",
     "labels": [],
     "merged": true,
-    "number": 45131,
-    "review_comments_count": 0,
+    "number": 45185,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix MoE routers returning probabilities instead of logits",
-    "updated_at": "2026-04-13T09:53:56Z"
+    "title": "Generalize gemma vision mask to videos",
+    "updated_at": "2026-04-02T13:15:46Z"
   },
   {
-    "additions": 122,
-    "author": "stevhliu",
+    "additions": 445,
+    "author": "remi-or",
     "author_association": "MEMBER",
-    "body_excerpt": "updates the `@auto_docstring` docs as part of the process of improving the model addition docs - updates title to be more precise as it can mean documenting a model in a `model.md` file as well - flat hierarchy before \u2192 each component-type\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# Summary This PR adds CPU offloading to continuous batching. It's in raft until perf and test status are reported. When the GPU KV cache is full and a request must be evicted, we check if there is enough VRAM to copy the request's KV cach\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45130",
-    "created_at": "2026-03-30T20:14:33Z",
-    "deletions": 154,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45130/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45130",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45184",
+    "created_at": "2026-04-02T10:12:00Z",
+    "deletions": 56,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45184/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45184",
     "labels": [],
-    "merged": true,
-    "number": 45130,
-    "review_comments_count": 2,
-    "state": "closed",
-    "title": "[docs] @auto_docstring decorator",
-    "updated_at": "2026-04-15T15:47:33Z"
+    "merged": false,
+    "number": 45184,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "[CB] [Major] Add CPU request offloading",
+    "updated_at": "2026-04-15T04:26:39Z"
   },
   {
-    "additions": 5,
-    "author": "IgnazioDS",
-    "author_association": "NONE",
-    "body_excerpt": "## Problem Fixes #45070. `PreTrainedConfig.dtype` was annotated as `Union[str, \"torch.dtype\"] | None`. Since `torch` is only imported under `TYPE_CHECKING`, pydantic's schema builder encounters the `\"torch.dtype\"` forward reference at runt\u2026",
-    "changed_files": 1,
+    "additions": 232,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? The `transformers` CLI currently does a lot of work before command dispatch. In particular, the top-level entrypoint eagerly imports `transformers` and CLI subcommands with heavy dependencies, so even simple invocat\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45129",
-    "created_at": "2026-03-30T19:13:52Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45129/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45129",
-    "labels": [
-      "Code agent slop"
-    ],
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45181",
+    "created_at": "2026-04-02T08:03:40Z",
+    "deletions": 23,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45181/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45181",
+    "labels": [],
     "merged": false,
-    "number": 45129,
+    "number": 45181,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(config): annotate PreTrainedConfig.dtype as Any to fix pydantic schema generation (#45070)",
-    "updated_at": "2026-03-31T12:33:39Z"
+    "state": "open",
+    "title": "Make the cli a top-level package",
+    "updated_at": "2026-04-09T12:01:41Z"
   },
   {
-    "additions": 443,
-    "author": "FaizanImran-blip",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "This PR fixes a crash in `_process_kwargs_parameters` that occurred when a module uses `from __future__ import annotations`. - Prevents AttributeError crash. - Adds a test `test_future_annotations.py` to verify the fix works. No other func\u2026",
-    "changed_files": 3,
+    "additions": 48,
+    "author": "paulinebm",
+    "author_association": "MEMBER",
+    "body_excerpt": "## \ud83d\udd12 Pin GitHub Actions to commit SHAs This PR pins all GitHub Actions to their exact commit SHA instead of mutable tags or branch names. **Why?** Pinning to a SHA prevents supply chain attacks where a tag (e.g. `v4`) could be moved to poi\u2026",
+    "changed_files": 18,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45128",
-    "created_at": "2026-03-30T19:10:48Z",
-    "deletions": 95,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45180",
+    "created_at": "2026-04-02T08:00:02Z",
+    "deletions": 48,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45128/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45128",
+    "files_url": "https://github.com/huggingface/transformers/pull/45180/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45180",
     "labels": [],
-    "merged": false,
-    "number": 45128,
+    "merged": true,
+    "number": 45180,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix: handle future annotations in _process_kwargs_parameters",
-    "updated_at": "2026-03-31T10:44:40Z"
+    "state": "closed",
+    "title": "\ud83d\udd12 Pin GitHub Actions to commit SHAs",
+    "updated_at": "2026-04-02T09:12:55Z"
   },
   {
-    "additions": 186,
-    "author": "tarekziade",
+    "additions": 327,
+    "author": "remi-or",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This patch adds exponential back-off for `load_audio_librosa` / `load_audio_as` / `load_audio_torchcodec` when they try to download a file. That generic utility is also used within `hub_retry` notice that there were\u2026",
-    "changed_files": 4,
+    "body_excerpt": "## Summary This PR ads minor changes to `cache.update`, updates the memory handler with all new features and refactors a few parts of the code to make it more readable. Cache indexing: - Replace fancy indexing (cache[idx, :, :]) with expli\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45126",
-    "created_at": "2026-03-30T16:34:40Z",
-    "deletions": 38,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45179",
+    "created_at": "2026-04-02T06:15:08Z",
+    "deletions": 288,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45126/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45126",
+    "files_url": "https://github.com/huggingface/transformers/pull/45179/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45179",
     "labels": [],
     "merged": true,
-    "number": 45126,
-    "review_comments_count": 2,
+    "number": 45179,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "http retries on audio file downloads",
-    "updated_at": "2026-04-09T15:12:40Z"
+    "title": "[CB] Tweaks to update and minor fixes",
+    "updated_at": "2026-04-03T09:36:51Z"
   },
   {
-    "additions": 3,
-    "author": "danielquintas8",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds `_tp_plan = {\"lm_head\": \"colwise_gather_output\"}` to `Qwen3_5MoeForConditionalGeneration` (the VL wrapper class). The text-only `Qwen3_5MoeForCausalLM` already had `_tp_plan`, but the VL variant was missing it.\u2026",
-    "changed_files": 2,
+    "additions": 4966,
+    "author": "masoudpz",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 20,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45124",
-    "created_at": "2026-03-30T16:23:11Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45176",
+    "created_at": "2026-04-02T00:47:45Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45124/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45124",
+    "files_url": "https://github.com/huggingface/transformers/pull/45176/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45176",
     "labels": [],
-    "merged": true,
-    "number": 45124,
+    "merged": false,
+    "number": 45176,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[Qwen3.5 MoE] Add _tp_plan to ForConditionalGeneration",
-    "updated_at": "2026-04-02T14:10:01Z"
+    "state": "open",
+    "title": "added efficietvitsam model to HF",
+    "updated_at": "2026-04-02T16:01:34Z"
   },
   {
-    "additions": 1,
-    "author": "Rocketknight1",
+    "additions": 49,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "The `test_ocr_queries` assertion value was wrong, even at the initial commit! I'm not sure how tests passed at the time but they're failing now in the CI. This PR fixes the target value!",
-    "changed_files": 1,
+    "body_excerpt": "updates the docs with changes from #44796: - added section for `--compile` in the serve optimization docs - added section for `--model-timeout` in the Loading models section (useful when a model is silently kicked off and a user doesn't kn\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45123",
-    "created_at": "2026-03-30T15:48:42Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45174",
+    "created_at": "2026-04-01T23:29:39Z",
+    "deletions": 21,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45123/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45123",
+    "files_url": "https://github.com/huggingface/transformers/pull/45174/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45174",
     "labels": [],
     "merged": true,
-    "number": 45123,
-    "review_comments_count": 0,
+    "number": 45174,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix PP test_ocr_queries",
-    "updated_at": "2026-03-30T16:28:38Z"
+    "title": "[docs] transformers serve",
+    "updated_at": "2026-04-02T16:39:12Z"
   },
   {
-    "additions": 106,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "Native model integration should never have anything related to remote code execution: That just means that we would need to add that model natively, not via remote - it can and will not be maintained by us. Additionally added a new linter\u2026",
-    "changed_files": 5,
+    "additions": 16,
+    "author": "JJJYmmm",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixed the `qwen3_5` / `qwen3_5_moe` reverse-loading tests by correcting the text model type used in the setup, and aligned the reverse-mapping behavior with gemma3n since they are all native multimodal. This also re\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45122",
-    "created_at": "2026-03-30T15:40:30Z",
-    "deletions": 34,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45122/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45122",
-    "labels": [
-      "for patch"
-    ],
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45173",
+    "created_at": "2026-04-01T20:20:37Z",
+    "deletions": 90,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45173/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45173",
+    "labels": [],
     "merged": true,
-    "number": 45122,
+    "number": 45173,
     "review_comments_count": 0,
     "state": "closed",
-    "title": ":rotating_light: [`LightGlue`] Remove remote code execution",
-    "updated_at": "2026-03-31T12:11:19Z"
+    "title": "[misc] fix qwen35 tests: correct the text model type and skip reverse_mapping",
+    "updated_at": "2026-04-02T13:05:39Z"
   },
   {
-    "additions": 56,
-    "author": "orbisai0security",
+    "additions": 156,
+    "author": "ezylopx5",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Fix critical severity security issue in `src/transformers/cli/serve.py`. ## Vulnerability | Field | Value | |-------|-------| | **ID** | V-007 | | **Severity** | CRITICAL | | **Scanner** | multi_agent_ai | | **Rule** | `V-007` |\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Problem Transformers currently provides sampling filters such as top-k, top-p, min-p, and top-h, but does not include top-n-sigma sampling from \"Top-n\u03c3: Not All Logits Are You Need\". This makes it harder to use a temperature-invariant t\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45121",
-    "created_at": "2026-03-30T15:02:06Z",
-    "deletions": 3,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45172",
+    "created_at": "2026-04-01T19:25:12Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45121/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45121",
+    "files_url": "https://github.com/huggingface/transformers/pull/45172/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45172",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 45121,
+    "number": 45172,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: remove unsafe exec() in serve.py",
-    "updated_at": "2026-03-31T12:09:59Z"
+    "title": "Add TopNSigmaLogitsWarper and top_n_sigma generation config support",
+    "updated_at": "2026-04-04T06:24:26Z"
   },
   {
-    "additions": 39,
-    "author": "akintunero",
+    "additions": 113,
+    "author": "Kash6",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary This PR fixes GitHub issue #45071: \"v5.4.0 breaks PretrainedConfig type checking\". The regression prevents type checkers (mypy, pyright) from validating `PretrainedConfig` subclass instantiation with valid parameters. ## Root Ca\u2026",
+    "body_excerpt": "When input_boxes contains a mix of None and real box entries (e.g. input_boxes=[None, [[x1,y1,x2,y2]]]), the processor pads None entries with [-10,-10,0,0] but does not generate corresponding input_boxes_labels. The model's geometry encode\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45119",
-    "created_at": "2026-03-30T13:48:47Z",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45171",
+    "created_at": "2026-04-01T18:26:48Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45119/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45119",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 45119,
+    "files_url": "https://github.com/huggingface/transformers/pull/45171/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45171",
+    "labels": [],
+    "merged": true,
+    "number": 45171,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix: Preserve PreTrainedConfig __init__ signatures for type checkers (fixes #45071)",
-    "updated_at": "2026-03-31T12:17:09Z"
-  },
-  {
-    "additions": 195,
-    "author": "sirzechs66",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds full GGUF loading support for GPT\u2011OSS models (20B/120B). It allows Transformers (and consequently vLLM) to directly load GPT\u2011OSS GGUF files without falling back to a wrong architecture. The changes incl\u2026",
-    "changed_files": 4,
-    "cluster_id": "cluster-43366-3",
-    "cluster_ids": [
-      "cluster-43366-3"
-    ],
-    "cluster_role": "canonical",
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45118",
-    "created_at": "2026-03-30T13:10:36Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45118/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45118",
-    "labels": [],
-    "merged": false,
-    "number": 45118,
-    "review_comments_count": 7,
-    "state": "open",
-    "title": "Add full GGUF loading support for GPT\u2011OSS (fixes #43366, supersedes #43757)",
-    "updated_at": "2026-04-15T13:54:59Z"
+    "title": "Fix Sam3Processor missing input_boxes_labels for padded None entries",
+    "updated_at": "2026-04-15T15:22:19Z"
   },
   {
-    "additions": 3,
-    "author": "Rocketknight1",
+    "additions": 19,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "Fixes #45084. We need to resolve the chat template in the Voxtral code to avoid `None` being passed to `_get_template_variables()`! cc @zucchini-nlp, follow-up to #44881",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Before I forget, opening a PR. Will cause conflicts in https://github.com/huggingface/transformers/pull/44431, so I will better merge this after refactoring",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45117",
-    "created_at": "2026-03-30T12:34:19Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45170",
+    "created_at": "2026-04-01T17:50:33Z",
+    "deletions": 17,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45117/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45117",
+    "files_url": "https://github.com/huggingface/transformers/pull/45170/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45170",
     "labels": [],
-    "merged": true,
-    "number": 45117,
+    "merged": false,
+    "number": 45170,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Copy the template resolution logic from the base apply_chat_template to Voxtral",
-    "updated_at": "2026-04-10T14:36:38Z"
+    "state": "open",
+    "title": "`layrnorm` -> `layernorm`",
+    "updated_at": "2026-04-01T18:03:31Z"
   },
   {
-    "additions": 222,
-    "author": "sirzechs66",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds full GGUF loading support for GPT\u2011OSS models (20B/120B). It allows Transformers (and consequently vLLM) to directly load GPT\u2011OSS GGUF files without falling back to a wrong architecture. The changes incl\u2026",
-    "changed_files": 5,
+    "additions": 10,
+    "author": "hmellor",
+    "author_association": "MEMBER",
+    "body_excerpt": "In https://github.com/huggingface/transformers/pull/45094 I introduced some errors to the remote code resolution when trying to detect if local code belonged to Transformers or not. These tests were: ```bash pytest tests/models/cohere_asr/\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45116",
-    "created_at": "2026-03-30T12:18:41Z",
-    "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45116/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45116",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45169",
+    "created_at": "2026-04-01T15:27:43Z",
+    "deletions": 6,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45169/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45169",
     "labels": [],
-    "merged": false,
-    "number": 45116,
+    "merged": true,
+    "number": 45169,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add full GGUF loading support for GPT\u2011OSS (fixes #43366)",
-    "updated_at": "2026-03-30T16:50:24Z"
+    "title": "Fix explicit local code resolution for tokenizers and image processors",
+    "updated_at": "2026-04-01T21:48:02Z"
   },
   {
-    "additions": 327,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 6,
+    "additions": 4,
+    "author": "w601sxs",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Based on our experimentation min and max lr for LLMs need to be set properly as defaults. Please refer to paper. For the broader community 1e-7 to 1e-4 are decent defaults # What does this PR do? <!-- Congratulations! You've made it this f\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45115",
-    "created_at": "2026-03-30T12:09:31Z",
-    "deletions": 375,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45115/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45115",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45168",
+    "created_at": "2026-04-01T15:02:15Z",
+    "deletions": 4,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45168/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45168",
     "labels": [],
     "merged": false,
-    "number": 45115,
+    "number": 45168,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Refactor/nemotron h inherit granitemoehybrid",
-    "updated_at": "2026-03-30T12:23:34Z"
+    "title": "Update min_lr and max_lr default values to better defaults",
+    "updated_at": "2026-04-08T16:58:59Z"
   },
   {
     "additions": 16,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This patch fixes all doctests for the run_doctest job",
-    "changed_files": 4,
+    "author": "xu-song",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds support for [Anthropic's JSON function style](https://platform.claude.com/docs/en/agents-and-tools/tool-use/define-tools): `{\"name\": \"...\", \"description\": \"...\", \"input_schema\": {...}}` ## Usage Example\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45114",
-    "created_at": "2026-03-30T11:08:02Z",
-    "deletions": 10,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45167",
+    "created_at": "2026-04-01T14:50:33Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45114/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45114",
+    "files_url": "https://github.com/huggingface/transformers/pull/45167/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45167",
     "labels": [],
     "merged": false,
-    "number": 45114,
+    "number": 45167,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: lets fix all doctests",
-    "updated_at": "2026-03-30T11:30:04Z"
+    "title": "Add anthropic style of function schema",
+    "updated_at": "2026-04-03T05:49:56Z"
   },
   {
-    "additions": 181,
-    "author": "cyyever",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds GPU Direct Storage (GDS) support for safetensors model loading via `torch.cuda.gds.GdsFile`. GDS is disabled by default, `HF_ENABLE_GDS=1` env is used to enable it. ## Benchmark A100 PCIe 40GB, Samsung\u2026",
-    "changed_files": 3,
+    "additions": 67,
+    "author": "Rocketknight1",
+    "author_association": "MEMBER",
+    "body_excerpt": "We didn't think we needed them, but I think we do after all!",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45113",
-    "created_at": "2026-03-30T10:55:35Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45166",
+    "created_at": "2026-04-01T14:46:33Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45113/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45113",
+    "files_url": "https://github.com/huggingface/transformers/pull/45166/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45166",
     "labels": [],
-    "merged": false,
-    "number": 45113,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "Add GDS support for safetensors loading ",
-    "updated_at": "2026-03-31T15:17:16Z"
+    "merged": true,
+    "number": 45166,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Re-add regex substitutions to the response parsing spec",
+    "updated_at": "2026-04-01T15:46:34Z"
   },
   {
-    "additions": 233,
-    "author": "remi-or",
+    "additions": 11,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "This PR adds a warmup phase before generation starts, turned on by default. It allows for better diagnostics and a more representative user experience than without warmup, where the cost of wamup is payed during the first request rather th\u2026",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? Emu3 was not updated in recent refactor and blip files were swapped. This PR fixes it Do we need anything to support BC importing from old files, or does it happen in `LazyImports` @yonigozlan ?",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45112",
-    "created_at": "2026-03-30T10:43:32Z",
-    "deletions": 46,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45165",
+    "created_at": "2026-04-01T14:18:38Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45112/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45112",
+    "files_url": "https://github.com/huggingface/transformers/pull/45165/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45165",
     "labels": [],
     "merged": true,
-    "number": 45112,
-    "review_comments_count": 5,
+    "number": 45165,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[CB] Add warmup feature",
-    "updated_at": "2026-03-31T07:42:18Z"
+    "title": "Fix missing image processors backends",
+    "updated_at": "2026-04-07T13:46:24Z"
   },
   {
-    "additions": 42,
-    "author": "ionut-anghelina",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary - Several MoE routers applied softmax inside `forward()` but returned the result as `router_logits`. The `load_balancing_loss_func` then applied softmax **again**, computing the aux loss on `softmax(softmax(logits))` which flatt\u2026",
-    "changed_files": 13,
+    "additions": 1,
+    "author": "albertvillanova",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix TypeError: 'NoneType' object is not iterable in `GenerationMixin.generate` - Fix for None layer_types <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is goin\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45111",
-    "created_at": "2026-03-30T08:23:07Z",
-    "deletions": 42,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45164",
+    "created_at": "2026-04-01T13:53:41Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45111/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45111",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 45111,
+    "files_url": "https://github.com/huggingface/transformers/pull/45164/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45164",
+    "labels": [],
+    "merged": true,
+    "number": 45164,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix double softmax in MoE router load-balancing loss",
-    "updated_at": "2026-03-30T14:09:10Z"
+    "title": "Fix TypeError: 'NoneType' object is not iterable in GenerationMixin.generate",
+    "updated_at": "2026-04-03T04:57:18Z"
   },
   {
-    "additions": 7035,
-    "author": "NielsRogge",
+    "additions": 131,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? [disclaimer: PR was entirely written by Codex where I just nudge it in the right directions, similar to #44285] ### Feature request I'd like to add support for Meta's [SAM 3.1](https://huggingface.co/facebook/sam3.1\u2026",
-    "changed_files": 20,
+    "body_excerpt": "# What does this PR do? Makes sure the full error is displayed on errors",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45110",
-    "created_at": "2026-03-30T08:19:42Z",
-    "deletions": 29,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45163",
+    "created_at": "2026-04-01T13:41:18Z",
+    "deletions": 20,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45110/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45110",
+    "files_url": "https://github.com/huggingface/transformers/pull/45163/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45163",
     "labels": [],
-    "merged": false,
-    "number": 45110,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "Add SAM 3.1",
-    "updated_at": "2026-03-30T12:33:41Z"
+    "merged": true,
+    "number": 45163,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "tweak checkers output on errors",
+    "updated_at": "2026-04-09T11:50:16Z"
   },
   {
-    "additions": 42,
-    "author": "aws-zhanxun",
+    "additions": 513,
+    "author": "onwp",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? `T5Attention.forward` hard-codes `n_heads` and `inner_dim` in `view()` calls. When using PyTorch Tensor Parallelism, `ColwiseParallel` shards the q/k/v projection output dim from `inner_dim` to `inner_dim / tp_size`\u2026",
+    "body_excerpt": "## Description Re-creates the Turkish documentation that was accidentally removed during the TF/Flax cleanup (commit fce74651). This PR adds the foundational Turkish docs with the complete \"Get Started\" section. ### Files added - `docs/sou\u2026",
     "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45109",
-    "created_at": "2026-03-30T07:06:19Z",
-    "deletions": 48,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45159",
+    "created_at": "2026-04-01T05:15:32Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45109/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45109",
+    "files_url": "https://github.com/huggingface/transformers/pull/45159/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45159",
     "labels": [],
-    "merged": true,
-    "number": 45109,
-    "review_comments_count": 3,
+    "merged": false,
+    "number": 45159,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix T5Attention shape mismatch under Tensor Parallelism",
-    "updated_at": "2026-04-01T16:21:32Z"
+    "title": "Add Turkish documentation: Get Started section",
+    "updated_at": "2026-04-01T05:16:49Z"
   },
   {
-    "additions": 1,
-    "author": "jiqing-feng",
+    "additions": 525,
+    "author": "onwp",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Description Some wav2vec2 models (e.g. audio classification variants) have `vocab_size: null` in their `config.json`. The current type annotation `vocab_size: int = 32` causes `huggingface_hub`'s strict dataclass validation to reject `N\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary This PR re-creates the Turkish (`tr`) documentation for the Transformers library, starting with the **Get Started** section. The original Turkish translation was accidentally removed in commit fce74651 (#40999). This contributio\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45108",
-    "created_at": "2026-03-30T03:06:01Z",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45158",
+    "created_at": "2026-04-01T05:05:40Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45108/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45108",
+    "files_url": "https://github.com/huggingface/transformers/pull/45158/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45158",
     "labels": [],
     "merged": true,
-    "number": 45108,
+    "number": 45158,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix `Wav2Vec2Config.vocab_size` type to allow `None`",
-    "updated_at": "2026-04-09T14:56:15Z"
+    "title": "Add Turkish (tr) translation for Get Started section",
+    "updated_at": "2026-04-02T17:50:46Z"
   },
   {
-    "additions": 1,
-    "author": "jiqing-feng",
+    "additions": 190,
+    "author": "Qubitium",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Description The `text-to-speech` pipeline crashes when loading models whose `generation_config` contains fields set to `None` (e.g. `use_cache=None`). This is because `TextToAudioPipeline.__init__` blindly copies all generation config v\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Add PrismML 1bit (gguf based, group-size 128) model inference suppport. ## Code Agent Policy - [ ] I confirm that this is not a pure code agent PR. ## Before submitting - [ ] This PR fixes a typo or improves the doc\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45107",
-    "created_at": "2026-03-30T02:32:13Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45107/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45107",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45157",
+    "created_at": "2026-04-01T03:11:51Z",
+    "deletions": 23,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45157/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45157",
     "labels": [],
-    "merged": true,
-    "number": 45107,
+    "merged": false,
+    "number": 45157,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix `text-to-speech` pipeline crash when generation config contains `None` values",
-    "updated_at": "2026-04-08T17:13:05Z"
+    "state": "open",
+    "title": "[WIP] PrismML Bonsai model support",
+    "updated_at": "2026-04-02T20:53:12Z"
   },
   {
-    "additions": 72,
-    "author": "rpathade",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes `_process_kwargs_parameters` crashing with `AttributeError` when `@auto_docstring` is applied in a module that uses `from __future__ import annotations`. Fixes #45103 ## Root cause `from __future__ import anno\u2026",
+    "additions": 50,
+    "author": "Cursx",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? After the `merge_and_unload()` operation in PEFT, embed_tokens and lm_head become independent tensors with different values, but config.tie_word_embeddings remains True. The load-side already detects this using torc\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45105",
-    "created_at": "2026-03-29T23:37:13Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45105/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45105",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45156",
+    "created_at": "2026-04-01T02:36:38Z",
+    "deletions": 0,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45156/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45156",
     "labels": [],
     "merged": false,
-    "number": 45105,
+    "number": 45156,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix @auto_docstring crash with from __future__ import annotations in _process_kwargs_parameters",
-    "updated_at": "2026-03-30T00:20:41Z"
+    "state": "closed",
+    "title": "Fix save_pretrained writing incorrect tie_word_embeddings=True config after PEFT merge",
+    "updated_at": "2026-04-01T09:34:05Z"
   },
   {
-    "additions": 13,
-    "author": "hkc5",
-    "author_association": "NONE",
-    "body_excerpt": "## Description Fixes #45103 The `@auto_docstring` decorator crashes at import time when applied to a class in a module that uses `from __future__ import annotations`. This is because `from __future__ import annotations` makes all annotatio\u2026",
+    "additions": 46,
+    "author": "michaelbenayoun",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Add support for loading adapters with `PreTrainedModel.load_adapter` when using TP. See: https://github.com/huggingface/peft/pull/3096",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45104",
-    "created_at": "2026-03-29T23:26:49Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45155",
+    "created_at": "2026-03-31T22:23:06Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45104/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45104",
+    "files_url": "https://github.com/huggingface/transformers/pull/45155/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45155",
+    "labels": [],
+    "merged": true,
+    "number": 45155,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "Load adapter with TP",
+    "updated_at": "2026-04-09T22:35:45Z"
+  },
+  {
+    "additions": 676,
+    "author": "FaizanImran-blip",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixed bug #45072 / #45071 where PretrainedConfig type checking and Pydantic model field validation were broken in v5.4.0. Added proper type checking for 'num_labels'. Added unit tests in test.py to verify correct and incorrect types. Verif\u2026",
+    "changed_files": 5,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45154",
+    "created_at": "2026-03-31T20:28:32Z",
+    "deletions": 153,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45154/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45154",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 45104,
+    "number": 45154,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix auto_docstring crash with from __future__ import annotations",
-    "updated_at": "2026-03-30T12:13:20Z"
+    "title": "Pretrained-config bug(45072/huggingfacebug)",
+    "updated_at": "2026-04-09T11:43:55Z"
+  },
+  {
+    "additions": 502,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "As per title, with torch releasing the varlen API, we can somewhat use native FA (with limited feature support) Restrictions - Unsupported features - Dropout - Learnable sinks (attention sinks) - Determinism - Softcap - CB KV cache native\u2026",
+    "changed_files": 19,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45153",
+    "created_at": "2026-03-31T19:43:19Z",
+    "deletions": 386,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45153/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45153",
+    "labels": [],
+    "merged": false,
+    "number": 45153,
+    "review_comments_count": 25,
+    "state": "open",
+    "title": "[`FA`] Native torch integration",
+    "updated_at": "2026-04-01T20:01:16Z"
   },
   {
-    "additions": 1664,
-    "author": "HemanthSai7",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Co-authored-by: Vishesht27 This PR adds support for codes for the upcoming Nandi series models. We also appreciate the valuable feedback and thorough review provided by @vasqu and @ArthurZucker \ud83e\udd17\ud83d\ude4f",
-    "changed_files": 11,
+    "additions": 230,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "refactors the testing docs to be more contributor-facing organized around writing model tests instead of being a collection of pytest usage examples and CI maintenance. also updates the pr checks doc so contributors are better prepared to\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45101",
-    "created_at": "2026-03-29T20:35:56Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45152",
+    "created_at": "2026-03-31T18:12:09Z",
+    "deletions": 1330,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45101/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45101",
+    "files_url": "https://github.com/huggingface/transformers/pull/45152/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45152",
     "labels": [],
     "merged": false,
-    "number": 45101,
-    "review_comments_count": 4,
+    "number": 45152,
+    "review_comments_count": 6,
     "state": "open",
-    "title": "Adding support for Nandi Models",
-    "updated_at": "2026-04-14T17:11:15Z"
+    "title": "[docs] model testing",
+    "updated_at": "2026-04-15T16:06:59Z"
   },
   {
     "additions": 1,
-    "author": "code-runner77",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "Improve wording in accelerator selection documentation # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you\u2026",
+    "author": "maanas1234",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? The previous code snippet had an error in how `TrainingArguments` was being used, which could lead to confusion or incorrect implementation. Corrected the example to properly demonstrate how to initialize and use `T\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45100",
-    "created_at": "2026-03-29T17:28:59Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45150",
+    "created_at": "2026-03-31T17:31:37Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45100/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45100",
+    "files_url": "https://github.com/huggingface/transformers/pull/45150/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45150",
+    "labels": [],
+    "merged": true,
+    "number": 45150,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "Fix incorrect TrainingArguments example in training.md",
+    "updated_at": "2026-03-31T18:31:25Z"
+  },
+  {
+    "additions": 4633,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "First pass: ~1M+ tokens in, ~115K+ out, Opus mainly, $42, 1h30 # PR #44320 vs Our Implementation ## What we got right - Same overall structure: modular file + generated standalone files + conversion script + tests + docs - Same model direc\u2026",
+    "changed_files": 20,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45149",
+    "created_at": "2026-03-31T15:56:11Z",
+    "deletions": 1,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45149/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45149",
     "labels": [],
     "merged": false,
-    "number": 45100,
+    "number": 45149,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Update accelerator_selection.md",
-    "updated_at": "2026-03-30T13:40:51Z"
+    "state": "open",
+    "title": "DO NOT MERGE adding SAML3-LiteText with a skill, first pass",
+    "updated_at": "2026-04-01T06:28:20Z"
   },
   {
-    "additions": 2,
-    "author": "zendy199x",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Fixes # (issue) ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by our ability to review an\u2026",
-    "changed_files": 1,
+    "additions": 13,
+    "author": "HallerPatrick",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #45146 ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by our ability to review and r\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45098",
-    "created_at": "2026-03-29T15:50:04Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45148",
+    "created_at": "2026-03-31T15:20:02Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45098/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45098",
+    "files_url": "https://github.com/huggingface/transformers/pull/45148/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45148",
     "labels": [],
     "merged": false,
-    "number": 45098,
+    "number": 45148,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: incomplete string literal causes syntax error in config docstring checker",
-    "updated_at": "2026-03-30T13:27:17Z"
+    "title": "Allow for all layers in Qwen3.5 architecture to be Gated Deltanet.",
+    "updated_at": "2026-04-02T11:18:12Z"
   },
   {
-    "additions": 771,
-    "author": "baonudesifeizhai",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR extends the InternVL conversion script to support the old `OpenGVLab/InternVL2-1B` and `OpenGVLab/InternVL2-2B` checkpoints. These checkpoints currently rely on remote code and are problematic for downstream\u2026",
-    "changed_files": 14,
+    "additions": 309,
+    "author": "mobicham",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR fixes hqq support that has been broken for a couple of months now after a refactoring: * Online quantization works fine now. * Serialization to load/save HQQ models is fixed too. ## Code Agent Policy - [x] I\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45097",
-    "created_at": "2026-03-29T05:29:42Z",
-    "deletions": 30,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45147",
+    "created_at": "2026-03-31T14:59:47Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45097/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45097",
+    "files_url": "https://github.com/huggingface/transformers/pull/45147/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45147",
     "labels": [],
     "merged": false,
-    "number": 45097,
-    "review_comments_count": 0,
+    "number": 45147,
+    "review_comments_count": 8,
     "state": "open",
-    "title": "Add old InternVL2-1B/2B support to the InternVL conversion script #45092",
-    "updated_at": "2026-04-02T07:59:13Z"
+    "title": "Fix broken HQQ support",
+    "updated_at": "2026-04-09T16:14:06Z"
   },
   {
-    "additions": 5,
-    "author": "hkc5",
-    "author_association": "NONE",
-    "body_excerpt": "## Problem Old remote-code checkpoints (like InternVL2) perform real-tensor operations during model construction (e.g., calling `.item()` on tensors). This causes `RuntimeError: Tensor.item() cannot be called on meta tensors` when models a\u2026",
-    "changed_files": 1,
+    "additions": 2018,
+    "author": "casinca",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Hello, this PR aims to add the MiMo-V2-Flash model to the Transformers library Fixes https://github.com/huggingface/transformers/issues/42954 MiMo-V2 is \"The last of the SOTAs\" that isn't natively supported by the T\u2026",
+    "changed_files": 12,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45096",
-    "created_at": "2026-03-29T05:13:12Z",
-    "deletions": 1,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45144",
+    "created_at": "2026-03-31T13:53:28Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45096/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45096",
+    "files_url": "https://github.com/huggingface/transformers/pull/45144/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45144",
     "labels": [
-      "Code agent slop"
+      "New model"
     ],
     "merged": false,
-    "number": 45096,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix: Skip meta device initialization for remote code models",
-    "updated_at": "2026-03-30T12:13:08Z"
+    "number": 45144,
+    "review_comments_count": 32,
+    "state": "open",
+    "title": "Add Xiaomi MiMo-V2",
+    "updated_at": "2026-04-16T14:50:26Z"
   },
   {
-    "additions": 100,
-    "author": "HanFa",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "When a config class has been explicitly registered via AutoConfig.register(), it should take precedence over auto_map remote code. Previously, `trust_remote_code=True` with auto_map.AutoConfig in config.json would always load remote code,\u2026",
-    "changed_files": 14,
+    "additions": 102,
+    "author": "Rocketknight1",
+    "author_association": "MEMBER",
+    "body_excerpt": "This PR adds `parse_response` to Processor classes by wrapping the `Tokenizer` method! cc @zucchini-nlp",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45094",
-    "created_at": "2026-03-29T04:21:29Z",
-    "deletions": 17,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45143",
+    "created_at": "2026-03-31T13:11:49Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45094/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45094",
+    "files_url": "https://github.com/huggingface/transformers/pull/45143/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45143",
     "labels": [],
     "merged": true,
-    "number": 45094,
-    "review_comments_count": 0,
+    "number": 45143,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "fix: prefer registered config over remote code in AutoConfig.from_pretrained",
-    "updated_at": "2026-03-31T14:56:49Z"
+    "title": "Add parse_response to Processor, make it a bit more official",
+    "updated_at": "2026-03-31T17:07:16Z"
   },
   {
-    "additions": 6,
-    "author": "hkc5",
-    "author_association": "NONE",
-    "body_excerpt": "This PR fixes the unexpected behaviour of helper function `_get_feat_extract_output_lengths` in qwen3_omni_moe as reported in #45083. ## Problem The current implementation incorrectly calculates the output length of the convolutional layer\u2026",
-    "changed_files": 3,
+    "additions": 4,
+    "author": "casinca",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45091",
-    "created_at": "2026-03-29T00:37:06Z",
-    "deletions": 9,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45142",
+    "created_at": "2026-03-31T12:47:46Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45091/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45091",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/45142/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45142",
+    "labels": [],
     "merged": false,
-    "number": 45091,
+    "number": 45142,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix _get_feat_extract_output_lengths in qwen3_omni_moe",
-    "updated_at": "2026-03-30T12:12:57Z"
+    "title": "refactor(gpt-oss): rename `eager_attention_forward` to `eager_attention_forward_with_sink`",
+    "updated_at": "2026-04-02T16:44:14Z"
   },
   {
-    "additions": 17,
-    "author": "hkc5",
-    "author_association": "NONE",
-    "body_excerpt": "## Description Fixes #45084 The `VoxtralProcessor.apply_chat_template` method was calling `_get_template_variables(chat_template)` without first checking if `chat_template` was None. This caused a `TypeError: Can't compile non template nod\u2026",
-    "changed_files": 1,
+    "additions": 19,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Really stupid ... Currently, change a docstring will count as \"modified files\", then the impacted files and test files to run are computed. Say, we add a comment to bert, many files impacted (via dependency) then ma\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45090",
-    "created_at": "2026-03-29T00:35:00Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45140",
+    "created_at": "2026-03-31T09:41:28Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45090/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45090",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 45090,
+    "files_url": "https://github.com/huggingface/transformers/pull/45140/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45140",
+    "labels": [],
+    "merged": true,
+    "number": 45140,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix TypeError when chat_template is None in VoxtralProcessor",
-    "updated_at": "2026-03-30T12:10:25Z"
+    "title": "Fix stupid test fetcher",
+    "updated_at": "2026-03-31T11:06:51Z"
   },
   {
-    "additions": 4,
-    "author": "Krishnachaitanyakc",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #45003 `_can_set_attn_implementation` and `_can_set_experts_implementation` in `PreTrainedModel` use `sys.modules[cls.__module__]`, which raises `KeyError` when a module has been removed from `sys.modules` at runtime. This\u2026",
-    "changed_files": 1,
+    "additions": 757,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? More fixes",
+    "changed_files": 99,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45089",
-    "created_at": "2026-03-28T16:44:06Z",
-    "deletions": 4,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45139",
+    "created_at": "2026-03-31T09:26:28Z",
+    "deletions": 976,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45089/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45089",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 45089,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/45139/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45139",
+    "labels": [],
+    "merged": true,
+    "number": 45139,
+    "review_comments_count": 26,
     "state": "closed",
-    "title": "fix: use sys.modules.get() to avoid KeyError in modeling_utils",
-    "updated_at": "2026-03-30T14:19:30Z"
+    "title": "Fix vllm cis",
+    "updated_at": "2026-04-08T11:19:41Z"
   },
   {
-    "additions": 6,
-    "author": "knQzx",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "corrects the conv output length calculation in _get_feat_extract_output_lengths which was computing wrong values for the audio encoder. fixes #45083",
-    "changed_files": 3,
+    "additions": 2,
+    "author": "Abdennacer-Badaoui",
+    "author_association": "MEMBER",
+    "body_excerpt": "This is a small T5 expectations update. It is the same for both AMD and NVIDIA A10 GPUs.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45088",
-    "created_at": "2026-03-28T16:40:07Z",
-    "deletions": 9,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45138",
+    "created_at": "2026-03-31T08:41:40Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45088/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45088",
+    "files_url": "https://github.com/huggingface/transformers/pull/45138/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45138",
     "labels": [],
-    "merged": false,
-    "number": 45088,
+    "merged": true,
+    "number": 45138,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix audio encoder output length formula in qwen3_omni_moe",
-    "updated_at": "2026-03-30T12:53:42Z"
+    "title": "CI] Small T5 expectations updated",
+    "updated_at": "2026-04-02T08:21:25Z"
   },
   {
-    "additions": 5,
-    "author": "hkc5",
+    "additions": 83,
+    "author": "Cursx",
     "author_association": "NONE",
-    "body_excerpt": "This PR fixes issue #45071 where mypy type checking was broken for PretrainedConfig subclasses. ## Problem In transformers v5.4.0, the PretrainedConfig class was converted to a dataclass with a wrapper around __init__ to accept arbitrary k\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? This PR fixes a bug in `PreTrainedModel.save_pretrained()` where `config.tie_word_embeddings` can be inconsistent with the actual weight state, leading to silent model corruption for downstream consumers. ### Proble\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45087",
-    "created_at": "2026-03-28T16:38:11Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45136",
+    "created_at": "2026-03-31T06:45:53Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45087/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45087",
+    "files_url": "https://github.com/huggingface/transformers/pull/45136/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45136",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 45087,
+    "number": 45136,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix PretrainedConfig type checking with mypy",
-    "updated_at": "2026-03-30T12:12:48Z"
-  },
-  {
-    "additions": 3,
-    "author": "knQzx",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "the function accesses backend_tokenizer.pre_tokenizer but the tokenizer passed is already the raw rust object, so it should be pre_tokenizer directly. fixes #45081",
-    "changed_files": 1,
-    "cluster_id": "cluster-45081-3",
-    "cluster_ids": [
-      "cluster-45081-3"
-    ],
-    "cluster_role": "member",
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45086",
-    "created_at": "2026-03-28T16:37:49Z",
-    "deletions": 3,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45086/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45086",
-    "labels": [],
-    "merged": false,
-    "number": 45086,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "fix AttributeError in _patch_mistral_regex",
-    "updated_at": "2026-03-28T16:37:49Z"
+    "title": "Fix #45127: Auto-fix diverged tie_word_embeddings config on save to prevent silent weight corruption",
+    "updated_at": "2026-03-31T12:21:32Z"
   },
   {
-    "additions": 7,
-    "author": "hkc5",
+    "additions": 44,
+    "author": "Cursx",
     "author_association": "NONE",
-    "body_excerpt": "This PR fixes #45072. ## Changes ### SwitchTransformers - Fixed a bug in `SwitchTransformersTop1Router.forward()` where `router_logits` was being reassigned to the max probability values instead of keeping the raw logits from the classifie\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Fixes an issue where PEFT adapters applied independently to tied embeddings (`embed_tokens` and [lm_head](cci:1://file:///d:/transformers/transformers/src/transformers/modeling_utils.py:2858:4-2985:26)) cause silent\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45085",
-    "created_at": "2026-03-28T16:28:27Z",
-    "deletions": 7,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45135",
+    "created_at": "2026-03-31T02:28:26Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45085/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45085",
+    "files_url": "https://github.com/huggingface/transformers/pull/45135/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45135",
     "labels": [],
     "merged": false,
-    "number": 45085,
+    "number": 45135,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix dtype mismatches in SwitchTransformers and TimmWrapperModel for bfloat16",
-    "updated_at": "2026-03-30T11:25:14Z"
+    "title": "Fix model saving corruption for dynamically untied embeddings",
+    "updated_at": "2026-03-31T04:36:56Z"
   },
   {
-    "additions": 143,
-    "author": "NielsRogge",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR updates the conversion script of VidEoMT to convert all remaining checkpoints. Find them here: https://huggingface.co/papers/2602.17807",
-    "changed_files": 2,
+    "additions": 91,
+    "author": "milesial",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Add support for CUDA parakeet preprocessor, running STFT and mel spectrogram extraction on the GPU. This refactor also speeds up the CPU implementation. Tested on `nvidia/parakeet-ctc-0.6b`, B200, 300s audio: Before\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45082",
-    "created_at": "2026-03-28T14:07:12Z",
-    "deletions": 14,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45134",
+    "created_at": "2026-03-31T01:59:28Z",
+    "deletions": 56,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45082/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45082",
+    "files_url": "https://github.com/huggingface/transformers/pull/45134/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45134",
     "labels": [],
     "merged": false,
-    "number": 45082,
+    "number": 45134,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[VidEoMT] Update conversion script",
-    "updated_at": "2026-03-28T14:16:54Z"
+    "title": "Optimize Parakeet feature extraction on CUDA",
+    "updated_at": "2026-04-07T14:56:51Z"
   },
   {
-    "additions": 13,
-    "author": "joaquinhuigomez",
-    "author_association": "NONE",
-    "body_excerpt": "## Root cause The v5.4.0 release converted `PreTrainedConfig` from a regular class to a `@dataclass`. This changes how Pydantic handles it when used as a field type in a `BaseModel`: instead of treating it as an opaque arbitrary type, Pyda\u2026",
-    "changed_files": 1,
+    "additions": 1944,
+    "author": "itazap",
+    "author_association": "MEMBER",
+    "body_excerpt": "## Summary - Auto-generated modular integration for `sarvam` - `modular_sarvam.py` written via HF Inference API guided by `modular_model_detector.py` - `modeling_sarvam.py` regenerated from modular via `modular_model_converter.py` ## Test\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45080",
-    "created_at": "2026-03-28T12:13:57Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45133",
+    "created_at": "2026-03-31T00:00:15Z",
     "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45080/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45080",
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45133/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45133",
     "labels": [],
     "merged": false,
-    "number": 45080,
+    "number": 45133,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix PreTrainedConfig as Pydantic field type after dataclass conversion",
-    "updated_at": "2026-03-29T19:36:24Z"
+    "state": "open",
+    "title": "Add sarvam model",
+    "updated_at": "2026-03-31T00:09:59Z"
   },
   {
-    "additions": 50,
-    "author": "javierdejesusda",
+    "additions": 215,
+    "author": "akintunero",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #35141 When `tie_word_embeddings=False`, calling `resize_token_embeddings()` then `post_init()` overwrites the LM head weights with random values. This happens because `_get_resized_lm_head()` returns a new `\u2026",
-    "changed_files": 2,
+    "body_excerpt": "## Summary This PR fixes GitHub issue #45120: \"Double softmax in MoE router load-balancing loss\". MoE routers in Mixtral, Qwen2MoE, and Qwen3VLMoE were applying softmax inside forward(), then the load_balancing_loss_func applied softmax AG\u2026",
+    "changed_files": 9,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45079",
-    "created_at": "2026-03-28T00:06:03Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45132",
+    "created_at": "2026-03-30T22:45:01Z",
+    "deletions": 34,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45079/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45079",
+    "files_url": "https://github.com/huggingface/transformers/pull/45132/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45132",
     "labels": [],
-    "merged": true,
-    "number": 45079,
+    "merged": false,
+    "number": 45132,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix resized LM head weights being overwritten by post_init",
-    "updated_at": "2026-04-02T14:13:31Z"
+    "title": "Fix: Remove double softmax in MoE router load-balancing loss (Mixtral, Qwen2MoE, Qwen3VLMoE)",
+    "updated_at": "2026-03-30T23:36:57Z"
   },
   {
-    "additions": 33,
-    "author": "itazap",
-    "author_association": "MEMBER",
-    "body_excerpt": "fixes fallback https://github.com/huggingface/transformers/issues/44993",
-    "changed_files": 2,
+    "additions": 30,
+    "author": "yacinemebarki",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes issue #45120: Several MoE routers returned softmaxed probabilities as `router_logits`, which caused `load_balancing_loss_func` to compute softmax(softmax(logits)), flattening routing distributions and weakenin\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45078",
-    "created_at": "2026-03-27T23:06:36Z",
-    "deletions": 9,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45131",
+    "created_at": "2026-03-30T21:18:47Z",
+    "deletions": 30,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45078/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45078",
+    "files_url": "https://github.com/huggingface/transformers/pull/45131/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45131",
     "labels": [],
-    "merged": false,
-    "number": 45078,
-    "review_comments_count": 5,
-    "state": "open",
-    "title": "throw error when conversion required",
-    "updated_at": "2026-04-15T13:59:35Z"
+    "merged": true,
+    "number": 45131,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix MoE routers returning probabilities instead of logits",
+    "updated_at": "2026-04-13T09:53:56Z"
   },
   {
-    "additions": 312,
-    "author": "dagecko",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Re-submission of #45010. Had a problem with my fork and had to delete it, which closed the original PR. Apologies for the noise. @tarekziade @ydshieh I noticed you fixed the critical findings from the original PR, which is great. This resu\u2026",
-    "changed_files": 22,
+    "additions": 122,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "updates the `@auto_docstring` docs as part of the process of improving the model addition docs - updates title to be more precise as it can mean documenting a model in a `model.md` file as well - flat hierarchy before \u2192 each component-type\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45077",
-    "created_at": "2026-03-27T22:20:56Z",
-    "deletions": 312,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45130",
+    "created_at": "2026-03-30T20:14:33Z",
+    "deletions": 154,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45077/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45077",
+    "files_url": "https://github.com/huggingface/transformers/pull/45130/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45130",
     "labels": [],
-    "merged": false,
-    "number": 45077,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "fix: pin 50 unpinned actions to commit SHA, extract 1 secret to env var",
-    "updated_at": "2026-03-27T22:20:56Z"
+    "merged": true,
+    "number": 45130,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "[docs] @auto_docstring decorator",
+    "updated_at": "2026-04-15T15:47:33Z"
   },
   {
-    "additions": 376,
-    "author": "osman-akkawi",
+    "additions": 5,
+    "author": "IgnazioDS",
     "author_association": "NONE",
-    "body_excerpt": "As Osman Akkawi, I am proud to submit this comprehensive Pull Request which introduces two world-first, unique innovations to the transformers library alongside essential codebase maintenance. This PR transforms how users interact with and\u2026",
-    "changed_files": 8,
+    "body_excerpt": "## Problem Fixes #45070. `PreTrainedConfig.dtype` was annotated as `Union[str, \"torch.dtype\"] | None`. Since `torch` is only imported under `TYPE_CHECKING`, pydantic's schema builder encounters the `\"torch.dtype\"` forward reference at runt\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45076",
-    "created_at": "2026-03-27T20:30:37Z",
-    "deletions": 9,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45129",
+    "created_at": "2026-03-30T19:13:52Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45076/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45076",
+    "files_url": "https://github.com/huggingface/transformers/pull/45129/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45129",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 45076,
+    "number": 45129,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Osman-Level Innovations: Hardware-Aware Advisor & Selective Weight Surgery CLI",
-    "updated_at": "2026-04-06T10:49:18Z"
+    "title": "fix(config): annotate PreTrainedConfig.dtype as Any to fix pydantic schema generation (#45070)",
+    "updated_at": "2026-03-31T12:33:39Z"
   },
   {
-    "additions": 5168,
-    "author": "thisisiron",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds the **DeepSeek-OCR2** model. ### Reference - Arxiv Paper: [DeepSeek-OCR 2: Visual Causal Flow](https://arxiv.org/abs/2601.20552) - Huggingface hub: [deepseek-ai/DeepSeek-OCR-2](https://huggingface.co/deepseek-a\u2026",
-    "changed_files": 21,
+    "additions": 443,
+    "author": "FaizanImran-blip",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "This PR fixes a crash in `_process_kwargs_parameters` that occurred when a module uses `from __future__ import annotations`. - Prevents AttributeError crash. - Adds a test `test_future_annotations.py` to verify the fix works. No other func\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45075",
-    "created_at": "2026-03-27T20:14:27Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45128",
+    "created_at": "2026-03-30T19:10:48Z",
+    "deletions": 95,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45075/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45075",
+    "files_url": "https://github.com/huggingface/transformers/pull/45128/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45128",
     "labels": [],
     "merged": false,
-    "number": 45075,
-    "review_comments_count": 57,
+    "number": 45128,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Add Deepseek-OCR-2 model",
-    "updated_at": "2026-04-15T11:14:06Z"
+    "title": "Fix: handle future annotations in _process_kwargs_parameters",
+    "updated_at": "2026-03-31T10:44:40Z"
   },
   {
-    "additions": 12,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following dtype mismatch use cases were identified and fixed in this PR: \u2192 **Switch Transformers:** [7938e91fa](https://github.com/harshaljanjani/transformers/commit/7938e91faabb051f3a001cd39c173d4697c2d81c) r\u2026",
+    "additions": 186,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This patch adds exponential back-off for `load_audio_librosa` / `load_audio_as` / `load_audio_torchcodec` when they try to download a file. That generic utility is also used within `hub_retry` notice that there were\u2026",
     "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45074",
-    "created_at": "2026-03-27T20:02:28Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45126",
+    "created_at": "2026-03-30T16:34:40Z",
+    "deletions": 38,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45074/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45074",
+    "files_url": "https://github.com/huggingface/transformers/pull/45126/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45126",
     "labels": [],
     "merged": true,
-    "number": 45074,
+    "number": 45126,
     "review_comments_count": 2,
     "state": "closed",
-    "title": "fix(models): Fix dtype mismatch in SwitchTransformers and TimmWrapperModel",
-    "updated_at": "2026-04-02T13:59:46Z"
+    "title": "http retries on audio file downloads",
+    "updated_at": "2026-04-09T15:12:40Z"
   },
   {
-    "additions": 1239,
-    "author": "Aravind-11",
+    "additions": 3,
+    "author": "danielquintas8",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "What does this pr do? - Add modular_owlvit.py inheriting CLIP vision/text embeddings, MLP, encoder layer, encoder - Import box IoU helpers from loss_for_object_detection; eager_attention from BERT - Regenerate modeling_owlvit.py via modula\u2026",
+    "body_excerpt": "# What does this PR do? Adds `_tp_plan = {\"lm_head\": \"colwise_gather_output\"}` to `Qwen3_5MoeForConditionalGeneration` (the VL wrapper class). The text-only `Qwen3_5MoeForCausalLM` already had `_tp_plan`, but the VL variant was missing it.\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45073",
-    "created_at": "2026-03-27T20:00:41Z",
-    "deletions": 135,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45124",
+    "created_at": "2026-03-30T16:23:11Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45073/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45073",
+    "files_url": "https://github.com/huggingface/transformers/pull/45124/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45124",
     "labels": [],
-    "merged": false,
-    "number": 45073,
+    "merged": true,
+    "number": 45124,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Refactor OwlViT to modular Transformers",
-    "updated_at": "2026-04-15T22:29:53Z"
+    "state": "closed",
+    "title": "[Qwen3.5 MoE] Add _tp_plan to ForConditionalGeneration",
+    "updated_at": "2026-04-02T14:10:01Z"
   },
   {
     "additions": 1,
-    "author": "Fr0do",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes a `TypeError` in `_check_received_keys` (line 919 of `modeling_rope_utils.py`) where `received_keys -= ignore_keys` fails when `ignore_keys` is a `list` instead of a `set`. ## Root cause Model configs (Qwen3.\u2026",
+    "author": "Rocketknight1",
+    "author_association": "MEMBER",
+    "body_excerpt": "The `test_ocr_queries` assertion value was wrong, even at the initial commit! I'm not sure how tests passed at the time but they're failing now in the CI. This PR fixes the target value!",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45069",
-    "created_at": "2026-03-27T19:21:01Z",
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45123",
+    "created_at": "2026-03-30T15:48:42Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45069/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45069",
+    "files_url": "https://github.com/huggingface/transformers/pull/45123/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45123",
     "labels": [],
     "merged": true,
-    "number": 45069,
+    "number": 45123,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix TypeError in rope validation when ignore_keys is a list",
-    "updated_at": "2026-03-30T11:41:12Z"
+    "title": "Fix PP test_ocr_queries",
+    "updated_at": "2026-03-30T16:28:38Z"
   },
   {
-    "additions": 22,
-    "author": "aarushisingh04",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### what does this PR do? this pr enables `trainer.train(resume_from_checkpoint=...)` to accept hugging face hub repository ids. instead of only local paths, users can now pass `user/repo@revision` and the trainer will automatically downlo\u2026",
+    "additions": 106,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "Native model integration should never have anything related to remote code execution: That just means that we would need to add that model natively, not via remote - it can and will not be maintained by us. Additionally added a new linter\u2026",
+    "changed_files": 5,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45122",
+    "created_at": "2026-03-30T15:40:30Z",
+    "deletions": 34,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45122/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45122",
+    "labels": [
+      "for patch"
+    ],
+    "merged": true,
+    "number": 45122,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": ":rotating_light: [`LightGlue`] Remove remote code execution",
+    "updated_at": "2026-03-31T12:11:19Z"
+  },
+  {
+    "additions": 56,
+    "author": "orbisai0security",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fix critical severity security issue in `src/transformers/cli/serve.py`. ## Vulnerability | Field | Value | |-------|-------| | **ID** | V-007 | | **Severity** | CRITICAL | | **Scanner** | multi_agent_ai | | **Rule** | `V-007` |\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45067",
-    "created_at": "2026-03-27T18:26:03Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45121",
+    "created_at": "2026-03-30T15:02:06Z",
     "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45067/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45067",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45121/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45121",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 45067,
+    "number": 45121,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "feat: trainer resume_from_checkpoint support hub downloads (#43375)",
-    "updated_at": "2026-03-27T18:52:10Z"
+    "state": "closed",
+    "title": "fix: remove unsafe exec() in serve.py",
+    "updated_at": "2026-03-31T12:09:59Z"
+  },
+  {
+    "additions": 39,
+    "author": "akintunero",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary This PR fixes GitHub issue #45071: \"v5.4.0 breaks PretrainedConfig type checking\". The regression prevents type checkers (mypy, pyright) from validating `PretrainedConfig` subclass instantiation with valid parameters. ## Root Ca\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45119",
+    "created_at": "2026-03-30T13:48:47Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45119/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45119",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 45119,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix: Preserve PreTrainedConfig __init__ signatures for type checkers (fixes #45071)",
+    "updated_at": "2026-03-31T12:17:09Z"
   },
   {
-    "additions": 305,
-    "author": "osman-akkawi",
-    "author_association": "NONE",
-    "body_excerpt": "As **Osman Akkawi**, I am proud to submit this Pull Request which introduces a world-first, unique feature to the `transformers` library alongside essential codebase maintenance. This PR focuses on one goal: making state-of-the-art models\u2026",
-    "changed_files": 7,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45066",
-    "created_at": "2026-03-27T17:04:14Z",
-    "deletions": 4,
+    "additions": 195,
+    "author": "sirzechs66",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds full GGUF loading support for GPT\u2011OSS models (20B/120B). It allows Transformers (and consequently vLLM) to directly load GPT\u2011OSS GGUF files without falling back to a wrong architecture. The changes incl\u2026",
+    "changed_files": 4,
+    "cluster_id": "cluster-43366-3",
+    "cluster_ids": [
+      "cluster-43366-3"
+    ],
+    "cluster_role": "canonical",
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45118",
+    "created_at": "2026-03-30T13:10:36Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45066/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45066",
+    "files_url": "https://github.com/huggingface/transformers/pull/45118/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45118",
     "labels": [],
     "merged": false,
-    "number": 45066,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[PR] Unique Enhancement: Transformers Model Advisor & Legacy Cleanup",
-    "updated_at": "2026-03-27T20:31:59Z"
+    "number": 45118,
+    "review_comments_count": 7,
+    "state": "open",
+    "title": "Add full GGUF loading support for GPT\u2011OSS (fixes #43366, supersedes #43757)",
+    "updated_at": "2026-04-16T03:42:33Z"
   },
   {
-    "additions": 0,
-    "author": "Sai-Suraj-27",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Removes TensorFlow's `\"TF_CPP_MIN_LOG_LEVEL\"` env var. This is no longer needed since TF/Jax are gone. This `utils/print_env.py` script is being used in CI running tests to print, req useful env vars. ## Code Agent\u2026",
+    "additions": 3,
+    "author": "Rocketknight1",
+    "author_association": "MEMBER",
+    "body_excerpt": "Fixes #45084. We need to resolve the chat template in the Voxtral code to avoid `None` being passed to `_get_template_variables()`! cc @zucchini-nlp, follow-up to #44881",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45065",
-    "created_at": "2026-03-27T16:43:30Z",
-    "deletions": 3,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45117",
+    "created_at": "2026-03-30T12:34:19Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45065/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45065",
+    "files_url": "https://github.com/huggingface/transformers/pull/45117/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45117",
     "labels": [],
     "merged": true,
-    "number": 45065,
+    "number": 45117,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Remove unused TensorFlow env var",
-    "updated_at": "2026-03-27T17:24:30Z"
+    "title": "Copy the template resolution logic from the base apply_chat_template to Voxtral",
+    "updated_at": "2026-04-10T14:36:38Z"
   },
   {
-    "additions": 101,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? check modular import can be extremely slow (8mn in CI) we're investigating speeding it up in https://github.com/huggingface/transformers/pull/45046 But we can also shard jobs in CI to mitigate a little bit. This pat\u2026",
-    "changed_files": 3,
+    "additions": 222,
+    "author": "sirzechs66",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds full GGUF loading support for GPT\u2011OSS models (20B/120B). It allows Transformers (and consequently vLLM) to directly load GPT\u2011OSS GGUF files without falling back to a wrong architecture. The changes incl\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45064",
-    "created_at": "2026-03-27T16:26:27Z",
-    "deletions": 34,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45116",
+    "created_at": "2026-03-30T12:18:41Z",
+    "deletions": 0,
     "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45064/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45064",
+    "files_url": "https://github.com/huggingface/transformers/pull/45116/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45116",
     "labels": [],
     "merged": false,
-    "number": 45064,
+    "number": 45116,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "refactor: shard checkers",
-    "updated_at": "2026-03-27T17:05:37Z"
-  },
-  {
-    "additions": 176,
-    "author": "SunMarc",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds some features that makes serving more efficient. It shouldn't impact `generate_batch` at all: - Per-request result delivery via callbacks (replaces shared queue contention). Added `_request_callbacks` d\u2026",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45063",
-    "created_at": "2026-03-27T16:07:43Z",
-    "deletions": 30,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45063/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45063",
-    "labels": [],
-    "merged": true,
-    "number": 45063,
-    "review_comments_count": 10,
     "state": "closed",
-    "title": "CB improvements for serving ",
-    "updated_at": "2026-03-30T18:48:33Z"
+    "title": "Add full GGUF loading support for GPT\u2011OSS (fixes #43366)",
+    "updated_at": "2026-03-30T16:50:24Z"
   },
   {
-    "additions": 55,
-    "author": "ErenAta16",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "This PR adds a regression test for Unicode corruption when decoding `added_tokens` with ByteLevel tokenizers (e.g. GPT-2 family). In affected cases, characters such as `\u010d`, `\u0107`, `\u0111` can decode into control characters (`\\r`, `\\x07`, `\\x11`)\u2026",
-    "changed_files": 2,
+    "additions": 327,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45062",
-    "created_at": "2026-03-27T15:23:38Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45062/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45062",
-    "labels": [
-      "Code agent slop"
-    ],
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45115",
+    "created_at": "2026-03-30T12:09:31Z",
+    "deletions": 375,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45115/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45115",
+    "labels": [],
     "merged": false,
-    "number": 45062,
-    "review_comments_count": 1,
-    "state": "closed",
-    "title": "Add regression test for ByteLevel added-token Unicode decode corruption",
-    "updated_at": "2026-03-27T23:36:35Z"
+    "number": 45115,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Refactor/nemotron h inherit granitemoehybrid",
+    "updated_at": "2026-03-30T12:23:34Z"
   },
   {
-    "additions": 12,
-    "author": "vasqu",
+    "additions": 16,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "As per title, see https://github.com/huggingface/transformers/pull/42435#issuecomment-4143234736",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? This patch fixes all doctests for the run_doctest job",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45061",
-    "created_at": "2026-03-27T15:09:50Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45114",
+    "created_at": "2026-03-30T11:08:02Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45061/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45061",
-    "labels": [
-      "for patch"
-    ],
-    "merged": true,
-    "number": 45061,
+    "files_url": "https://github.com/huggingface/transformers/pull/45114/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45114",
+    "labels": [],
+    "merged": false,
+    "number": 45114,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[`FA`] Fix BC support for a few versions + add deprecation cycle",
-    "updated_at": "2026-03-27T15:37:13Z"
+    "state": "open",
+    "title": "fix: lets fix all doctests",
+    "updated_at": "2026-03-30T11:30:04Z"
   },
   {
-    "additions": 54,
-    "author": "ErenAta16",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "This PR fixes a regression where PIL-based image/video processors were incorrectly treated as requiring `torchvision`. As a result, `AutoProcessor` / `AutoImageProcessor` could fail in environments without `torchvision`, even though a vali\u2026",
+    "additions": 181,
+    "author": "cyyever",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds GPU Direct Storage (GDS) support for safetensors model loading via `torch.cuda.gds.GdsFile`. GDS is disabled by default, `HF_ENABLE_GDS=1` env is used to enable it. ## Benchmark A100 PCIe 40GB, Samsung\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45060",
-    "created_at": "2026-03-27T13:43:38Z",
-    "deletions": 3,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45113",
+    "created_at": "2026-03-30T10:55:35Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45060/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45060",
+    "files_url": "https://github.com/huggingface/transformers/pull/45113/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45113",
     "labels": [],
     "merged": false,
-    "number": 45060,
-    "review_comments_count": 0,
+    "number": 45113,
+    "review_comments_count": 3,
     "state": "open",
-    "title": "Fix PIL backend fallback when torchvision is unavailable",
-    "updated_at": "2026-03-30T13:50:34Z"
+    "title": "Add GDS support for safetensors loading ",
+    "updated_at": "2026-03-31T15:17:16Z"
   },
   {
-    "additions": 9,
-    "author": "hmellor",
+    "additions": 233,
+    "author": "remi-or",
     "author_association": "MEMBER",
-    "body_excerpt": "Some checkpoints, such as https://huggingface.co/omni-research/Tarsier2-Recap-7b, have the wrong `model_type` in their `config.json`. This PR allows advanced users (vLLM) to pass `model_type` into `AutoConfig.from_pretrained` via `kwargs`\u2026",
-    "changed_files": 1,
+    "body_excerpt": "This PR adds a warmup phase before generation starts, turned on by default. It allows for better diagnostics and a more representative user experience than without warmup, where the cost of wamup is payed during the first request rather th\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45058",
-    "created_at": "2026-03-27T13:24:05Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45112",
+    "created_at": "2026-03-30T10:43:32Z",
+    "deletions": 46,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45058/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45058",
+    "files_url": "https://github.com/huggingface/transformers/pull/45112/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45112",
     "labels": [],
     "merged": true,
-    "number": 45058,
-    "review_comments_count": 0,
+    "number": 45112,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "Allow advanced users to override `model_type` in `AutoConfig.from_pretrained`",
-    "updated_at": "2026-03-27T14:29:53Z"
+    "title": "[CB] Add warmup feature",
+    "updated_at": "2026-03-31T07:42:18Z"
   },
   {
-    "additions": 318,
-    "author": "NathanHB",
-    "author_association": "MEMBER",
-    "body_excerpt": "Change model_dump_json() to model_dump() to avoid double JSON encoding. When using continuous batching with stream=false, the response was being double-encoded as a string instead of returning a proper JSON object. Added a UV script to run\u2026",
-    "changed_files": 3,
+    "additions": 42,
+    "author": "ionut-anghelina",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Several MoE routers applied softmax inside `forward()` but returned the result as `router_logits`. The `load_balancing_loss_func` then applied softmax **again**, computing the aux loss on `softmax(softmax(logits))` which flatt\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45057",
-    "created_at": "2026-03-27T13:02:59Z",
-    "deletions": 37,
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45111",
+    "created_at": "2026-03-30T08:23:07Z",
+    "deletions": 42,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45057/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45057",
-    "labels": [],
-    "merged": true,
-    "number": 45057,
+    "files_url": "https://github.com/huggingface/transformers/pull/45111/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45111",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 45111,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[serving] Fix continuous batching JSON response serialization",
-    "updated_at": "2026-03-31T13:04:32Z"
+    "title": "Fix double softmax in MoE router load-balancing loss",
+    "updated_at": "2026-03-30T14:09:10Z"
   },
   {
-    "additions": 331,
-    "author": "ArthurZucker",
+    "additions": 7035,
+    "author": "NielsRogge",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This is mega long due I wanted to check benches. Its not super super huge but a win is a win",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? [disclaimer: PR was entirely written by Codex where I just nudge it in the right directions, similar to #44285] ### Feature request I'd like to add support for Meta's [SAM 3.1](https://huggingface.co/facebook/sam3.1\u2026",
+    "changed_files": 20,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45056",
-    "created_at": "2026-03-27T11:36:10Z",
-    "deletions": 28,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45110",
+    "created_at": "2026-03-30T08:19:42Z",
+    "deletions": 29,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45056/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45056",
+    "files_url": "https://github.com/huggingface/transformers/pull/45110/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45110",
     "labels": [],
     "merged": false,
-    "number": 45056,
-    "review_comments_count": 5,
+    "number": 45110,
+    "review_comments_count": 3,
     "state": "open",
-    "title": "[`auto_docstring`] needs to be only run on __doc__ ",
-    "updated_at": "2026-03-29T12:57:04Z"
+    "title": "Add SAM 3.1",
+    "updated_at": "2026-03-30T12:33:41Z"
   },
   {
-    "additions": 3,
-    "author": "vasanthrpjan1-boop",
+    "additions": 42,
+    "author": "aws-zhanxun",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? When `Trainer` saves a checkpoint for a model that is not a `PreTrainedModel` (e.g. a custom `nn.Module`), it only saves the state dict but not the model config. This means `Model.from_pretrained(ckpt_path)` requir\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? `T5Attention.forward` hard-codes `n_heads` and `inner_dim` in `view()` calls. When using PyTorch Tensor Parallelism, `ColwiseParallel` shards the q/k/v projection output dim from `inner_dim` to `inner_dim / tp_size`\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45055",
-    "created_at": "2026-03-27T11:31:10Z",
-    "deletions": 0,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45109",
+    "created_at": "2026-03-30T07:06:19Z",
+    "deletions": 48,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45055/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45055",
+    "files_url": "https://github.com/huggingface/transformers/pull/45109/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45109",
     "labels": [],
-    "merged": false,
-    "number": 45055,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Save model config in Trainer checkpoints for non-PreTrainedModel models",
-    "updated_at": "2026-03-27T11:31:10Z"
+    "merged": true,
+    "number": 45109,
+    "review_comments_count": 3,
+    "state": "closed",
+    "title": "Fix T5Attention shape mismatch under Tensor Parallelism",
+    "updated_at": "2026-04-01T16:21:32Z"
   },
   {
-    "additions": 3,
-    "author": "hf-security-analysis[bot]",
+    "additions": 1,
+    "author": "jiqing-feng",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Update `.github/workflows/update_metdata.yml` workflow configuration. cc @tarekziade @ydshieh Closes huggingface/tracking-issues#33",
+    "body_excerpt": "## Description Some wav2vec2 models (e.g. audio classification variants) have `vocab_size: null` in their `config.json`. The current type annotation `vocab_size: int = 32` causes `huggingface_hub`'s strict dataclass validation to reject `N\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45054",
-    "created_at": "2026-03-27T11:16:11Z",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45108",
+    "created_at": "2026-03-30T03:06:01Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45054/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45054",
+    "files_url": "https://github.com/huggingface/transformers/pull/45108/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45108",
     "labels": [],
     "merged": true,
-    "number": 45054,
+    "number": 45108,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "chore: update update_metdata.yml",
-    "updated_at": "2026-03-27T15:57:14Z"
+    "title": "Fix `Wav2Vec2Config.vocab_size` type to allow `None`",
+    "updated_at": "2026-04-09T14:56:15Z"
   },
   {
     "additions": 1,
-    "author": "Sai-Suraj-27",
+    "author": "jiqing-feng",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes failing [`XCLIPModelIntegrationTests`](https://github.com/huggingface/transformers/actions/runs/23230643883/job/67524760869#step:14:1384). [`self.get_attributes()`](https://github.com/huggingface/transformers/\u2026",
+    "body_excerpt": "## Description The `text-to-speech` pipeline crashes when loading models whose `generation_config` contains fields set to `None` (e.g. `use_cache=None`). This is because `TextToAudioPipeline.__init__` blindly copies all generation config v\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45053",
-    "created_at": "2026-03-27T11:11:18Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45107",
+    "created_at": "2026-03-30T02:32:13Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45053/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45053",
+    "files_url": "https://github.com/huggingface/transformers/pull/45107/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45107",
     "labels": [],
-    "merged": false,
-    "number": 45053,
-    "review_comments_count": 3,
+    "merged": true,
+    "number": 45107,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix failing `XCLIPModelIntegrationTest`",
-    "updated_at": "2026-04-13T11:05:45Z"
+    "title": "Fix `text-to-speech` pipeline crash when generation config contains `None` values",
+    "updated_at": "2026-04-08T17:13:05Z"
   },
   {
-    "additions": 1,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? `.gitignore` was not updated when `mlinter` was refactored",
-    "changed_files": 1,
+    "additions": 72,
+    "author": "rpathade",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes `_process_kwargs_parameters` crashing with `AttributeError` when `@auto_docstring` is applied in a module that uses `from __future__ import annotations`. Fixes #45103 ## Root cause `from __future__ import anno\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45052",
-    "created_at": "2026-03-27T10:16:17Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45105",
+    "created_at": "2026-03-29T23:37:13Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45052/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45052",
+    "files_url": "https://github.com/huggingface/transformers/pull/45105/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45105",
     "labels": [],
-    "merged": true,
-    "number": 45052,
+    "merged": false,
+    "number": 45105,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "chore: Fix mlinter cache location",
-    "updated_at": "2026-03-27T10:26:38Z"
+    "state": "open",
+    "title": "Fix @auto_docstring crash with from __future__ import annotations in _process_kwargs_parameters",
+    "updated_at": "2026-03-30T00:20:41Z"
   },
   {
-    "additions": 2,
-    "author": "albertvillanova",
-    "author_association": "MEMBER",
-    "body_excerpt": "Fix NaN weights on non-rank-0 FSDP processes by using `zeros_like` instead of `empty_like` in `_move_missing_keys_from_meta_to_device` Follow-up to: - #44473 See related downstream issue in `trl` : - https://github.com/huggingface/trl/issu\u2026",
+    "additions": 13,
+    "author": "hkc5",
+    "author_association": "NONE",
+    "body_excerpt": "## Description Fixes #45103 The `@auto_docstring` decorator crashes at import time when applied to a class in a module that uses `from __future__ import annotations`. This is because `from __future__ import annotations` makes all annotatio\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45050",
-    "created_at": "2026-03-27T09:19:32Z",
-    "deletions": 2,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45104",
+    "created_at": "2026-03-29T23:26:49Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45050/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45050",
-    "labels": [],
-    "merged": true,
-    "number": 45050,
-    "review_comments_count": 8,
+    "files_url": "https://github.com/huggingface/transformers/pull/45104/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45104",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 45104,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix NaN weights on non-rank-0 FSDP processes",
-    "updated_at": "2026-04-13T13:35:35Z"
+    "title": "Fix auto_docstring crash with from __future__ import annotations",
+    "updated_at": "2026-03-30T12:13:20Z"
   },
   {
-    "additions": 6,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45030 After the config validation, all validations are now run after config is initialized. So this config has been wrong from the beginning but we didn't com\u2026",
-    "changed_files": 1,
+    "additions": 1499,
+    "author": "HemanthSai7",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Co-authored-by: Vishesht27 This PR adds support for codes for the upcoming Nandi series models. We also appreciate the valuable feedback and thorough review provided by @vasqu and @ArthurZucker \ud83e\udd17\ud83d\ude4f",
+    "changed_files": 11,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45049",
-    "created_at": "2026-03-27T08:46:46Z",
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45101",
+    "created_at": "2026-03-29T20:35:56Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45049/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45049",
-    "labels": [
-      "for patch"
-    ],
-    "merged": true,
-    "number": 45049,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix when RoPE params are in kwargs",
-    "updated_at": "2026-03-27T16:28:13Z"
+    "files_url": "https://github.com/huggingface/transformers/pull/45101/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45101",
+    "labels": [],
+    "merged": false,
+    "number": 45101,
+    "review_comments_count": 4,
+    "state": "open",
+    "title": "Adding support for Nandi Models",
+    "updated_at": "2026-04-16T17:05:47Z"
   },
   {
-    "additions": 4,
-    "author": "Sai-Suraj-27",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes failing [`SmolLM3IntegrationTest:test_model_3b_long_prompt`](https://github.com/huggingface/transformers/actions/runs/23629638266/job/68826332952#step:14:216). `SmolLM3` has [`do_sample=True` by default](https\u2026",
+    "additions": 1,
+    "author": "code-runner77",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "Improve wording in accelerator selection documentation # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45048",
-    "created_at": "2026-03-27T08:45:26Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45100",
+    "created_at": "2026-03-29T17:28:59Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45048/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45048",
+    "files_url": "https://github.com/huggingface/transformers/pull/45100/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45100",
     "labels": [],
-    "merged": true,
-    "number": 45048,
+    "merged": false,
+    "number": 45100,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix failing `SmolLM3IntegrationTest`",
-    "updated_at": "2026-03-27T14:18:05Z"
-  },
-  {
-    "additions": 503,
-    "author": "Akshay404error",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "Fixes #44568 \u2014 restores add_special_tokens behavior for mDeBERTa tokenizer and non-persistent buffers in v5 ## What does this PR do? This PR fixes two v5 regressions: 1. `add_special_tokens=True` no longer added BOS/EOS tokens for the `mic\u2026",
-    "changed_files": 5,
+    "title": "Update accelerator_selection.md",
+    "updated_at": "2026-03-30T13:40:51Z"
+  },
+  {
+    "additions": 2,
+    "author": "zendy199x",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Fixes # (issue) ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by our ability to review an\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45047",
-    "created_at": "2026-03-27T08:33:01Z",
-    "deletions": 14,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45098",
+    "created_at": "2026-03-29T15:50:04Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45047/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45047",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/45098/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45098",
+    "labels": [],
     "merged": false,
-    "number": 45047,
+    "number": 45098,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: restore add_special_tokens behavior for mDeBERTa tokenizer and n\u2026",
-    "updated_at": "2026-03-27T13:15:58Z"
+    "title": "fix: incomplete string literal causes syntax error in config docstring checker",
+    "updated_at": "2026-03-30T13:27:17Z"
   },
   {
-    "additions": 548,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Investigate speedups on modular conversion",
-    "changed_files": 2,
+    "additions": 771,
+    "author": "baonudesifeizhai",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR extends the InternVL conversion script to support the old `OpenGVLab/InternVL2-1B` and `OpenGVLab/InternVL2-2B` checkpoints. These checkpoints currently rely on remote code and are problematic for downstream\u2026",
+    "changed_files": 14,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45046",
-    "created_at": "2026-03-27T08:31:58Z",
-    "deletions": 42,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45046/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45046",
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45097",
+    "created_at": "2026-03-29T05:29:42Z",
+    "deletions": 30,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45097/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45097",
     "labels": [],
     "merged": false,
-    "number": 45046,
+    "number": 45097,
     "review_comments_count": 0,
     "state": "open",
-    "title": "investigate modular conversion speedups",
-    "updated_at": "2026-03-27T16:33:45Z"
+    "title": "Add old InternVL2-1B/2B support to the InternVL conversion script #45092",
+    "updated_at": "2026-04-02T07:59:13Z"
   },
   {
-    "additions": 5095,
-    "author": "Lidang-Jiang",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Isolate dependencies, make PIL independant from Torchvision backend Fixes #45042 PR #45029 added `@requires(backends=(\"vision\", \"torch\", \"torchvision\"))` to 67 PIL backend `image_processing_pil_*.py` files. This causes PIL backend class\u2026",
-    "changed_files": 188,
+    "additions": 5,
+    "author": "hkc5",
+    "author_association": "NONE",
+    "body_excerpt": "## Problem Old remote-code checkpoints (like InternVL2) perform real-tensor operations during model construction (e.g., calling `.item()` on tensors). This causes `RuntimeError: Tensor.item() cannot be called on meta tensors` when models a\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45045",
-    "created_at": "2026-03-27T08:19:19Z",
-    "deletions": 1270,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45096",
+    "created_at": "2026-03-29T05:13:12Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45045/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45045",
+    "files_url": "https://github.com/huggingface/transformers/pull/45096/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45096",
     "labels": [
-      "for patch"
+      "Code agent slop"
     ],
-    "merged": true,
-    "number": 45045,
-    "review_comments_count": 24,
+    "merged": false,
+    "number": 45096,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[Bugfix] Remove incorrect torchvision requirement from PIL backend image processors",
-    "updated_at": "2026-03-30T07:25:49Z"
+    "title": "Fix: Skip meta device initialization for remote code models",
+    "updated_at": "2026-03-30T12:13:08Z"
   },
   {
-    "additions": 35,
-    "author": "kaixuanliu",
+    "additions": 100,
+    "author": "HanFa",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fix issue in https://github.com/huggingface/transformers/issues/44792. @zucchini-nlp @ydshieh pls help review, thx!",
-    "changed_files": 3,
+    "body_excerpt": "When a config class has been explicitly registered via AutoConfig.register(), it should take precedence over auto_map remote code. Previously, `trust_remote_code=True` with auto_map.AutoConfig in config.json would always load remote code,\u2026",
+    "changed_files": 14,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45044",
-    "created_at": "2026-03-27T07:50:21Z",
-    "deletions": 26,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45094",
+    "created_at": "2026-03-29T04:21:29Z",
+    "deletions": 17,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45044/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45044",
+    "files_url": "https://github.com/huggingface/transformers/pull/45094/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45094",
     "labels": [],
     "merged": true,
-    "number": 45044,
-    "review_comments_count": 11,
+    "number": 45094,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix bug for janus model image generation",
-    "updated_at": "2026-04-02T02:46:14Z"
+    "title": "fix: prefer registered config over remote code in AutoConfig.from_pretrained",
+    "updated_at": "2026-03-31T14:56:49Z"
   },
   {
-    "additions": 31,
-    "author": "Lidang-Jiang",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #45003. `_can_set_attn_implementation` and `_can_set_experts_implementation` in `modeling_utils.py` crash with `KeyError` when `cls.__module__` is absent from `sys.modules`. This happens in real-world scenari\u2026",
-    "changed_files": 2,
+    "additions": 6,
+    "author": "hkc5",
+    "author_association": "NONE",
+    "body_excerpt": "This PR fixes the unexpected behaviour of helper function `_get_feat_extract_output_lengths` in qwen3_omni_moe as reported in #45083. ## Problem The current implementation incorrectly calculates the output length of the convolutional layer\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45043",
-    "created_at": "2026-03-27T06:07:20Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45091",
+    "created_at": "2026-03-29T00:37:06Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45043/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45043",
+    "files_url": "https://github.com/huggingface/transformers/pull/45091/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45091",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 45043,
+    "number": 45091,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Bugfix] Use sys.modules.get() to avoid KeyError in modeling_utils",
-    "updated_at": "2026-03-27T13:02:19Z"
-  },
-  {
-    "additions": 677,
-    "author": "JJJYmmm",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR addresses the performance issues observed with nn.Conv3d across different PyTorch/cuDNN, such as https://github.com/vllm-project/vllm/pull/27418, https://mp.weixin.qq.com/s/hKRIpB561EdrMY8cbg1hEw. We replace\u2026",
-    "changed_files": 7,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 14,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45041",
-    "created_at": "2026-03-27T03:50:54Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45041/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45041",
-    "labels": [],
-    "merged": true,
-    "number": 45041,
-    "review_comments_count": 61,
-    "state": "closed",
-    "title": "[inference_fusion] convert conv3d patch embed to linear",
-    "updated_at": "2026-04-13T15:30:39Z"
+    "title": "Fix _get_feat_extract_output_lengths in qwen3_omni_moe",
+    "updated_at": "2026-03-30T12:12:57Z"
   },
   {
-    "additions": 13,
-    "author": "sywangyi",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "- vision models: @yonigozlan @molbap - CIs: @ydshieh when running tests/models/video_llama_3/test_modeling_video_llama_3.py::VideoLlama3IntegrationTest all fail cause by lm_head.weight is missing.",
-    "changed_files": 3,
+    "additions": 17,
+    "author": "hkc5",
+    "author_association": "NONE",
+    "body_excerpt": "## Description Fixes #45084 The `VoxtralProcessor.apply_chat_template` method was calling `_get_template_variables(chat_template)` without first checking if `chat_template` was None. This caused a `TypeError: Can't compile non template nod\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45040",
-    "created_at": "2026-03-27T02:56:58Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45090",
+    "created_at": "2026-03-29T00:35:00Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45040/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45040",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45090/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45090",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 45040,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "Llama3 video fix",
-    "updated_at": "2026-03-27T10:55:26Z"
+    "number": 45090,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix TypeError when chat_template is None in VoxtralProcessor",
+    "updated_at": "2026-03-30T12:10:25Z"
   },
   {
-    "additions": 26,
-    "author": "Lidang-Jiang",
+    "additions": 4,
+    "author": "Krishnachaitanyakc",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #44716 - Propagate `interpolate_pos_encoding` parameter through `PixioModel.forward()`, `PixioBackbone.forward()`, and `PixioEmbeddings.forward()` down to `PixioPatchEmbeddings.forward()` - Follows the same pattern used by\u2026",
-    "changed_files": 2,
+    "body_excerpt": "## Summary Fixes #45003 `_can_set_attn_implementation` and `_can_set_experts_implementation` in `PreTrainedModel` use `sys.modules[cls.__module__]`, which raises `KeyError` when a module has been removed from `sys.modules` at runtime. This\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45039",
-    "created_at": "2026-03-27T02:54:31Z",
-    "deletions": 10,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45089",
+    "created_at": "2026-03-28T16:44:06Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45039/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45039",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45089/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45089",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 45039,
+    "number": 45089,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Bugfix] Propagate interpolate_pos_encoding through Pixio model",
-    "updated_at": "2026-03-27T12:57:30Z"
+    "title": "fix: use sys.modules.get() to avoid KeyError in modeling_utils",
+    "updated_at": "2026-03-30T14:19:30Z"
   },
   {
-    "additions": 1,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
+    "additions": 6,
+    "author": "knQzx",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "corrects the conv output length calculation in _get_feat_extract_output_lengths which was computing wrong values for the audio encoder. fixes #45083",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45038",
-    "created_at": "2026-03-27T00:18:35Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45088",
+    "created_at": "2026-03-28T16:40:07Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45038/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45038",
+    "files_url": "https://github.com/huggingface/transformers/pull/45088/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45088",
     "labels": [],
-    "merged": true,
-    "number": 45038,
+    "merged": false,
+    "number": 45088,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "style was missing sorry @ydshieh :)",
-    "updated_at": "2026-03-27T00:28:23Z"
+    "title": "fix audio encoder output length formula in qwen3_omni_moe",
+    "updated_at": "2026-03-30T12:53:42Z"
   },
   {
-    "additions": 1,
-    "author": "asuryateja",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "\u2026_interface docs The `custom_attention` function definition in the attention_interface documentation was missing a colon at the end of the return type annotation, making it invalid Python syntax. # What does this PR do? <!-- Congratulation\u2026",
+    "additions": 5,
+    "author": "hkc5",
+    "author_association": "NONE",
+    "body_excerpt": "This PR fixes issue #45071 where mypy type checking was broken for PretrainedConfig subclasses. ## Problem In transformers v5.4.0, the PretrainedConfig class was converted to a dataclass with a wrapper around __init__ to accept arbitrary k\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45037",
-    "created_at": "2026-03-26T23:48:04Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45087",
+    "created_at": "2026-03-28T16:38:11Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45037/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45037",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45087/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45087",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 45037,
+    "number": 45087,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "add missing colon in custom_attention function signature in attention\u2026",
-    "updated_at": "2026-03-27T00:17:55Z"
+    "state": "closed",
+    "title": "Fix PretrainedConfig type checking with mypy",
+    "updated_at": "2026-03-30T12:12:48Z"
   },
   {
-    "additions": 15,
-    "author": "matdou",
+    "additions": 3,
+    "author": "knQzx",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #45030 Configs like `tiny-random/glm-4v` store `rope_theta` at the top level of `config.json` alongside a `rope_scaling` dict (legacy format). For config classes that don't declare `rope_parameters` as a datac\u2026",
+    "body_excerpt": "the function accesses backend_tokenizer.pre_tokenizer but the tokenizer passed is already the raw rust object, so it should be pre_tokenizer directly. fixes #45081",
     "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
+    "cluster_id": "cluster-45081-3",
+    "cluster_ids": [
+      "cluster-45081-3"
+    ],
+    "cluster_role": "member",
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45036",
-    "created_at": "2026-03-26T23:00:35Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45086",
+    "created_at": "2026-03-28T16:37:49Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45036/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45036",
+    "files_url": "https://github.com/huggingface/transformers/pull/45086/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45086",
     "labels": [],
     "merged": false,
-    "number": 45036,
-    "review_comments_count": 1,
-    "state": "closed",
-    "title": "[fix] BC for legacy configs with top-level rope_theta when rope_parameters is set via rope_scaling",
-    "updated_at": "2026-03-28T23:47:33Z"
+    "number": 45086,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "fix AttributeError in _patch_mistral_regex",
+    "updated_at": "2026-03-28T16:37:49Z"
   },
   {
-    "additions": 1,
-    "author": "asuryateja",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "The conversion operations table was missing PermuteForRope. Added it with its reverse (itself), consistent with how other operations are documented. PermuteForRope is self-inverse applying it twice returns the original tensor layout. # Wha\u2026",
-    "changed_files": 1,
+    "additions": 7,
+    "author": "hkc5",
+    "author_association": "NONE",
+    "body_excerpt": "This PR fixes #45072. ## Changes ### SwitchTransformers - Fixed a bug in `SwitchTransformersTop1Router.forward()` where `router_logits` was being reassigned to the max probability values instead of keeping the raw logits from the classifie\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45035",
-    "created_at": "2026-03-26T21:05:17Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45085",
+    "created_at": "2026-03-28T16:28:27Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45035/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45035",
+    "files_url": "https://github.com/huggingface/transformers/pull/45085/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45085",
     "labels": [],
-    "merged": true,
-    "number": 45035,
+    "merged": false,
+    "number": 45085,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "docs: add PermuteForRope to conversion operations reverse table",
-    "updated_at": "2026-03-26T22:09:53Z"
+    "title": "Fix dtype mismatches in SwitchTransformers and TimmWrapperModel for bfloat16",
+    "updated_at": "2026-03-30T11:25:14Z"
   },
   {
-    "additions": 113,
-    "author": "sdharani91",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This is a follow up to https://github.com/huggingface/transformers/pull/44867 This PR fixes Qwen3.5 padding-free packed inputs on the linear-attention fast path by consuming collator-provided packed metadata. The li\u2026",
-    "changed_files": 6,
+    "additions": 143,
+    "author": "NielsRogge",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR updates the conversion script of VidEoMT to convert all remaining checkpoints. Find them here: https://huggingface.co/papers/2602.17807",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 12,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45034",
-    "created_at": "2026-03-26T20:52:51Z",
-    "deletions": 9,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45082",
+    "created_at": "2026-03-28T14:07:12Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45034/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45034",
+    "files_url": "https://github.com/huggingface/transformers/pull/45082/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45082",
     "labels": [],
     "merged": false,
-    "number": 45034,
-    "review_comments_count": 39,
+    "number": 45082,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Pass packed boundary metadata to Qwen3.5 linear-attention fast kernels from data collator",
-    "updated_at": "2026-04-15T13:57:34Z"
+    "title": "[VidEoMT] Update conversion script",
+    "updated_at": "2026-03-28T14:16:54Z"
   },
   {
-    "additions": 3,
-    "author": "hmellor",
-    "author_association": "MEMBER",
-    "body_excerpt": "In https://github.com/huggingface/transformers/pull/43514, `BaseImageProcessorFast` became `BaseImageProcessor` and `_further_process_kwargs` was renamed to `_standardize_kwargs` This PR adds some BC for the old name of this method.",
+    "additions": 13,
+    "author": "joaquinhuigomez",
+    "author_association": "NONE",
+    "body_excerpt": "## Root cause The v5.4.0 release converted `PreTrainedConfig` from a regular class to a `@dataclass`. This changes how Pydantic handles it when used as a field type in a `BaseModel`: instead of treating it as an opaque arbitrary type, Pyda\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45033",
-    "created_at": "2026-03-26T20:36:40Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45080",
+    "created_at": "2026-03-28T12:13:57Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45033/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45033",
+    "files_url": "https://github.com/huggingface/transformers/pull/45080/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45080",
     "labels": [],
-    "merged": true,
-    "number": 45033,
+    "merged": false,
+    "number": 45080,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add BC for `_further_process_kwargs`",
-    "updated_at": "2026-03-26T21:01:32Z"
+    "title": "Fix PreTrainedConfig as Pydantic field type after dataclass conversion",
+    "updated_at": "2026-03-29T19:36:24Z"
   },
   {
-    "additions": 163,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Use multi runners to get new failing tests in a CI run.",
+    "additions": 50,
+    "author": "javierdejesusda",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #35141 When `tie_word_embeddings=False`, calling `resize_token_embeddings()` then `post_init()` overwrites the LM head weights with random values. This happens because `_get_resized_lm_head()` returns a new `\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45032",
-    "created_at": "2026-03-26T18:48:18Z",
-    "deletions": 67,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45079",
+    "created_at": "2026-03-28T00:06:03Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45032/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45032",
+    "files_url": "https://github.com/huggingface/transformers/pull/45079/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45079",
     "labels": [],
     "merged": true,
-    "number": 45032,
+    "number": 45079,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Use multi runners to check new failing tests in a CI run",
-    "updated_at": "2026-03-26T18:59:08Z"
+    "title": "Fix resized LM head weights being overwritten by post_init",
+    "updated_at": "2026-04-02T14:13:31Z"
   },
   {
-    "additions": 6,
-    "author": "tomaarsen",
+    "additions": 33,
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Follow-up for https://github.com/huggingface/transformers/pull/44931, which added weight tying for Camembert. Only the CamembertForMaskedLM class had the right _tied_weights_keys, the CamembertForCausalLM had the in\u2026",
+    "body_excerpt": "fixes fallback https://github.com/huggingface/transformers/issues/44993",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45031",
-    "created_at": "2026-03-26T18:28:54Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45078",
+    "created_at": "2026-03-27T23:06:36Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45031/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45031",
+    "files_url": "https://github.com/huggingface/transformers/pull/45078/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45078",
     "labels": [],
-    "merged": true,
-    "number": 45031,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[`fix`] Use the correct _tied_weights_keys for CamembertForCausalLM",
-    "updated_at": "2026-03-26T18:57:29Z"
+    "merged": false,
+    "number": 45078,
+    "review_comments_count": 5,
+    "state": "open",
+    "title": "throw error when conversion required",
+    "updated_at": "2026-04-16T15:02:23Z"
   },
   {
-    "additions": 478,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Release workflow is failing",
-    "changed_files": 101,
+    "additions": 312,
+    "author": "dagecko",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Re-submission of #45010. Had a problem with my fork and had to delete it, which closed the original PR. Apologies for the noise. @tarekziade @ydshieh I noticed you fixed the critical findings from the original PR, which is great. This resu\u2026",
+    "changed_files": 22,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45029",
-    "created_at": "2026-03-26T18:04:37Z",
-    "deletions": 768,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45077",
+    "created_at": "2026-03-27T22:20:56Z",
+    "deletions": 312,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45029/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45029",
+    "files_url": "https://github.com/huggingface/transformers/pull/45077/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45077",
     "labels": [],
-    "merged": true,
-    "number": 45029,
-    "review_comments_count": 7,
+    "merged": false,
+    "number": 45077,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "fix: pin 50 unpinned actions to commit SHA, extract 1 secret to env var",
+    "updated_at": "2026-03-27T22:20:56Z"
+  },
+  {
+    "additions": 376,
+    "author": "osman-akkawi",
+    "author_association": "NONE",
+    "body_excerpt": "As Osman Akkawi, I am proud to submit this comprehensive Pull Request which introduces two world-first, unique innovations to the transformers library alongside essential codebase maintenance. This PR transforms how users interact with and\u2026",
+    "changed_files": 8,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45076",
+    "created_at": "2026-03-27T20:30:37Z",
+    "deletions": 9,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45076/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45076",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 45076,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix release full",
-    "updated_at": "2026-03-27T06:34:00Z"
+    "title": "Osman-Level Innovations: Hardware-Aware Advisor & Selective Weight Surgery CLI",
+    "updated_at": "2026-04-06T10:49:18Z"
   },
   {
-    "additions": 3258,
-    "author": "3outeille",
-    "author_association": "MEMBER",
-    "body_excerpt": "- TODO - how will dtensor works with quantization ? - how will dtensor works with kernels ? - Needs end to end test (combine `verify_all_loss` -> `training` with saving + loading back for generate ?) - double check Save FSDP + TP - do test\u2026",
-    "changed_files": 218,
+    "additions": 5168,
+    "author": "thisisiron",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Adds the **DeepSeek-OCR2** model. ### Reference - Arxiv Paper: [DeepSeek-OCR 2: Visual Causal Flow](https://arxiv.org/abs/2601.20552) - Huggingface hub: [deepseek-ai/DeepSeek-OCR-2](https://huggingface.co/deepseek-a\u2026",
+    "changed_files": 21,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45028",
-    "created_at": "2026-03-26T17:50:07Z",
-    "deletions": 3591,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/45028/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45028",
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45075",
+    "created_at": "2026-03-27T20:14:27Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45075/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45075",
     "labels": [],
     "merged": false,
-    "number": 45028,
-    "review_comments_count": 26,
+    "number": 45075,
+    "review_comments_count": 57,
     "state": "open",
-    "title": "TP refactor for FSDP + TP integration",
-    "updated_at": "2026-04-15T15:30:41Z"
+    "title": "Add Deepseek-OCR-2 model",
+    "updated_at": "2026-04-15T11:14:06Z"
   },
   {
-    "additions": 714,
-    "author": "remi-or",
-    "author_association": "MEMBER",
-    "body_excerpt": "# Summary This PR adds per-request logits processors and overalls the way CB handles logits processors. It introduces batched logits processing with per-request parameters for continuous batching, enabling each request in a batch to use di\u2026",
-    "changed_files": 11,
+    "additions": 12,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? The following dtype mismatch use cases were identified and fixed in this PR: \u2192 **Switch Transformers:** [7938e91fa](https://github.com/harshaljanjani/transformers/commit/7938e91faabb051f3a001cd39c173d4697c2d81c) r\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45026",
-    "created_at": "2026-03-26T17:00:07Z",
-    "deletions": 202,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45074",
+    "created_at": "2026-03-27T20:02:28Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45026/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45026",
+    "files_url": "https://github.com/huggingface/transformers/pull/45074/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45074",
     "labels": [],
     "merged": true,
-    "number": 45026,
-    "review_comments_count": 32,
+    "number": 45074,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "[CB] Add per-request logits processors",
-    "updated_at": "2026-04-03T16:44:11Z"
+    "title": "fix(models): Fix dtype mismatch in SwitchTransformers and TimmWrapperModel",
+    "updated_at": "2026-04-02T13:59:46Z"
   },
   {
-    "additions": 2,
-    "author": "layla1824",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "This PR adds a clarification comment regarding the behavior of rotary_pct. Currently, rotary_pct may reset to its default value (0.25) after reload due to the use of kwargs.pop. This note helps developers better understand this behavior.",
-    "changed_files": 1,
+    "additions": 1239,
+    "author": "Aravind-11",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "What does this pr do? - Add modular_owlvit.py inheriting CLIP vision/text embeddings, MLP, encoder layer, encoder - Import box IoU helpers from loss_for_object_detection; eager_attention from BERT - Regenerate modeling_owlvit.py via modula\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45025",
-    "created_at": "2026-03-26T16:48:53Z",
-    "deletions": 1,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45073",
+    "created_at": "2026-03-27T20:00:41Z",
+    "deletions": 135,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45025/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45025",
+    "files_url": "https://github.com/huggingface/transformers/pull/45073/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45073",
     "labels": [],
     "merged": false,
-    "number": 45025,
+    "number": 45073,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Update configuration_Clarify rotary_pct reset behavior in GPTNeoXConfiggpt_neox.py",
-    "updated_at": "2026-03-27T09:05:48Z"
+    "state": "open",
+    "title": "Refactor OwlViT to modular Transformers",
+    "updated_at": "2026-04-15T22:29:53Z"
   },
   {
     "additions": 1,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? `tf` and `flax` are long gone (unknown extras generate only warnings so this slipped through the cracks)",
+    "author": "Fr0do",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes a `TypeError` in `_check_received_keys` (line 919 of `modeling_rope_utils.py`) where `received_keys -= ignore_keys` fails when `ignore_keys` is a `list` instead of a `set`. ## Root cause Model configs (Qwen3.\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45024",
-    "created_at": "2026-03-26T16:21:30Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45069",
+    "created_at": "2026-03-27T19:21:01Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45024/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45024",
+    "files_url": "https://github.com/huggingface/transformers/pull/45069/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45069",
     "labels": [],
     "merged": true,
-    "number": 45024,
+    "number": 45069,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "chore: remove old extras",
-    "updated_at": "2026-03-27T11:16:03Z"
-  },
-  {
-    "additions": 2529,
-    "author": "eustlb",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? **Integration notes:** For now, this integration does not load mel filters from the checkpoint. The original model was trained backpropagating gradients in it, but we saw previously (with parakeet-ctc) that this doe\u2026",
-    "changed_files": 19,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45023",
-    "created_at": "2026-03-26T15:58:08Z",
-    "deletions": 8,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45023/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45023",
-    "labels": [
-      "New model",
-      "Audio"
-    ],
-    "merged": true,
-    "number": 45023,
-    "review_comments_count": 13,
-    "state": "closed",
-    "title": "Add cohere asr",
-    "updated_at": "2026-03-26T22:48:16Z"
+    "title": "Fix TypeError in rope validation when ignore_keys is a list",
+    "updated_at": "2026-03-30T11:41:12Z"
   },
   {
-    "additions": 1,
-    "author": "popotest",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "additions": 22,
+    "author": "aarushisingh04",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### what does this PR do? this pr enables `trainer.train(resume_from_checkpoint=...)` to accept hugging face hub repository ids. instead of only local paths, users can now pass `user/repo@revision` and the trainer will automatically downlo\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45022",
-    "created_at": "2026-03-26T15:33:19Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45067",
+    "created_at": "2026-03-27T18:26:03Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45022/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45022",
+    "files_url": "https://github.com/huggingface/transformers/pull/45067/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45067",
     "labels": [],
     "merged": false,
-    "number": 45022,
+    "number": 45067,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Update _config.py",
-    "updated_at": "2026-03-27T13:17:36Z"
+    "state": "open",
+    "title": "feat: trainer resume_from_checkpoint support hub downloads (#43375)",
+    "updated_at": "2026-03-27T18:52:10Z"
   },
   {
-    "additions": 2,
-    "author": "hf-security-analysis[bot]",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Update `.github/workflows/anti-slop.yml` workflow configuration. cc @tarekziade Closes huggingface/tracking-issues#30",
-    "changed_files": 1,
+    "additions": 305,
+    "author": "osman-akkawi",
+    "author_association": "NONE",
+    "body_excerpt": "As **Osman Akkawi**, I am proud to submit this Pull Request which introduces a world-first, unique feature to the `transformers` library alongside essential codebase maintenance. This PR focuses on one goal: making state-of-the-art models\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45021",
-    "created_at": "2026-03-26T13:40:18Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45066",
+    "created_at": "2026-03-27T17:04:14Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45021/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45021",
+    "files_url": "https://github.com/huggingface/transformers/pull/45066/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45066",
     "labels": [],
     "merged": false,
-    "number": 45021,
+    "number": 45066,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "chore: update anti-slop.yml",
-    "updated_at": "2026-03-26T13:53:57Z"
+    "title": "[PR] Unique Enhancement: Transformers Model Advisor & Legacy Cleanup",
+    "updated_at": "2026-03-27T20:31:59Z"
   },
   {
-    "additions": 21,
-    "author": "javierdejesusda",
+    "additions": 0,
+    "author": "Sai-Suraj-27",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #44877 Loading `ibm-granite/granite-4.0-1b-speech` fails with `StrictDataclassFieldValidationError` because its config.json stores `embedding_multiplier` and `logits_scaling` as integers (e.g. `12`, `8`), but\u2026",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Removes TensorFlow's `\"TF_CPP_MIN_LOG_LEVEL\"` env var. This is no longer needed since TF/Jax are gone. This `utils/print_env.py` script is being used in CI running tests to print, req useful env vars. ## Code Agent\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45019",
-    "created_at": "2026-03-26T11:19:19Z",
-    "deletions": 12,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45065",
+    "created_at": "2026-03-27T16:43:30Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45019/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45019",
+    "files_url": "https://github.com/huggingface/transformers/pull/45065/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45065",
     "labels": [],
     "merged": true,
-    "number": 45019,
-    "review_comments_count": 2,
+    "number": 45065,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix GraniteConfig type hints to accept int for multiplier fields",
-    "updated_at": "2026-03-27T09:30:17Z"
+    "title": "Remove unused TensorFlow env var",
+    "updated_at": "2026-03-27T17:24:30Z"
   },
   {
-    "additions": 1814,
-    "author": "zucchini-nlp",
+    "additions": 101,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per title, deletes long lines of auto-mapping and dynamically builds it from code. The whole auto-generation runs with AST and takes ~5 sec on my machine, so we don't infer anything at run-time. Instead it is bui\u2026",
-    "changed_files": 68,
+    "body_excerpt": "# What does this PR do? check modular import can be extremely slow (8mn in CI) we're investigating speeding it up in https://github.com/huggingface/transformers/pull/45046 But we can also shard jobs in CI to mitigate a little bit. This pat\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45018",
-    "created_at": "2026-03-26T11:18:13Z",
-    "deletions": 1437,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45018/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45018",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45064",
+    "created_at": "2026-03-27T16:26:27Z",
+    "deletions": 34,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45064/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45064",
     "labels": [],
     "merged": false,
-    "number": 45018,
-    "review_comments_count": 7,
+    "number": 45064,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Dynamic auto mapping",
-    "updated_at": "2026-04-15T13:28:35Z"
+    "title": "refactor: shard checkers",
+    "updated_at": "2026-03-27T17:05:37Z"
   },
   {
-    "additions": 571,
-    "author": "JaredforReal",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? ### Get the rope operation right Before: NeoX split-half style After: GPT-J/interleaved style(`interleaved=True` same as `is_neox_style=Flase`) the right one ### Get rid of `F.relu` Reason: - `F.relu` works with `ac\u2026",
-    "changed_files": 5,
+    "additions": 176,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR adds some features that makes serving more efficient. It shouldn't impact `generate_batch` at all: - Per-request result delivery via callbacks (replaces shared queue contention). Added `_request_callbacks` d\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45017",
-    "created_at": "2026-03-26T09:21:10Z",
-    "deletions": 49,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45063",
+    "created_at": "2026-03-27T16:07:43Z",
+    "deletions": 30,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45017/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45017",
+    "files_url": "https://github.com/huggingface/transformers/pull/45063/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45063",
     "labels": [],
-    "merged": false,
-    "number": 45017,
-    "review_comments_count": 5,
-    "state": "open",
-    "title": "[WIP][Fix] GLM 5 set `apply_rotary_pos_emb` to `is_neox_style=False` && remove `F.relu()`",
-    "updated_at": "2026-04-14T09:36:01Z"
+    "merged": true,
+    "number": 45063,
+    "review_comments_count": 10,
+    "state": "closed",
+    "title": "CB improvements for serving ",
+    "updated_at": "2026-03-30T18:48:33Z"
   },
   {
-    "additions": 64,
-    "author": "inisis",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "additions": 55,
+    "author": "ErenAta16",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "This PR adds a regression test for Unicode corruption when decoding `added_tokens` with ByteLevel tokenizers (e.g. GPT-2 family). In affected cases, characters such as `\u010d`, `\u0107`, `\u0111` can decode into control characters (`\\r`, `\\x07`, `\\x11`)\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45016",
-    "created_at": "2026-03-26T09:09:41Z",
-    "deletions": 4,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45062",
+    "created_at": "2026-03-27T15:23:38Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45016/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45016",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45062/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45062",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 45016,
-    "review_comments_count": 0,
+    "number": 45062,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "fix: glm5 inference bug",
-    "updated_at": "2026-03-26T12:10:21Z"
+    "title": "Add regression test for ByteLevel added-token Unicode decode corruption",
+    "updated_at": "2026-03-27T23:36:35Z"
   },
   {
-    "additions": 55,
-    "author": "pnehete23",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes a `KeyError` in `_can_set_attn_implementation` and `_can_set_experts_implementation` when a model's module is absent from `sys.modules`. Fixes #45003 ## Root Cause Both `_can_set_attn_implementation` (line 19\u2026",
+    "additions": 12,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "As per title, see https://github.com/huggingface/transformers/pull/42435#issuecomment-4143234736",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45015",
-    "created_at": "2026-03-26T08:52:10Z",
-    "deletions": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45061",
+    "created_at": "2026-03-27T15:09:50Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45015/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45015",
+    "files_url": "https://github.com/huggingface/transformers/pull/45061/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45061",
     "labels": [
-      "Code agent slop"
+      "for patch"
     ],
-    "merged": false,
-    "number": 45015,
+    "merged": true,
+    "number": 45061,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: guard sys.modules access in _can_set_attn/experts_implementation",
-    "updated_at": "2026-03-26T12:15:50Z"
+    "title": "[`FA`] Fix BC support for a few versions + add deprecation cycle",
+    "updated_at": "2026-03-27T15:37:13Z"
   },
   {
-    "additions": 16,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? #30674 refactors the way we obtain CircleCI test files to run for each job. It always puts [\"tests\"] for `tests_hub`, so each commit of each PR will run it, no matter if there is any change to codebase. Let's reduce\u2026",
-    "changed_files": 1,
+    "additions": 54,
+    "author": "ErenAta16",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "This PR fixes a regression where PIL-based image/video processors were incorrectly treated as requiring `torchvision`. As a result, `AutoProcessor` / `AutoImageProcessor` could fail in environments without `torchvision`, even though a vali\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45014",
-    "created_at": "2026-03-26T08:52:04Z",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45060",
+    "created_at": "2026-03-27T13:43:38Z",
     "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45014/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45014",
+    "files_url": "https://github.com/huggingface/transformers/pull/45060/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45060",
     "labels": [],
-    "merged": true,
-    "number": 45014,
-    "review_comments_count": 1,
-    "state": "closed",
-    "title": "Don't run `tests_hub` if no tests found",
-    "updated_at": "2026-03-26T09:32:39Z"
+    "merged": false,
+    "number": 45060,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix PIL backend fallback when torchvision is unavailable",
+    "updated_at": "2026-03-30T13:50:34Z"
   },
   {
-    "additions": 325,
-    "author": "tarekziade",
+    "additions": 9,
+    "author": "hmellor",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - Adds a new checker in `make chek-repo` that will `import transformers` and count the number of imported module. - Lazy import of torch when doing `import transformers` The change will reduce the import time from ~\u2026",
-    "changed_files": 6,
+    "body_excerpt": "Some checkpoints, such as https://huggingface.co/omni-research/Tarsier2-Recap-7b, have the wrong `model_type` in their `config.json`. This PR allows advanced users (vLLM) to pass `model_type` into `AutoConfig.from_pretrained` via `kwargs`\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45013",
-    "created_at": "2026-03-26T07:47:23Z",
-    "deletions": 49,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45058",
+    "created_at": "2026-03-27T13:24:05Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45013/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45013",
+    "files_url": "https://github.com/huggingface/transformers/pull/45058/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45058",
     "labels": [],
     "merged": true,
-    "number": 45013,
-    "review_comments_count": 3,
+    "number": 45058,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "feature: added import complexity checker",
-    "updated_at": "2026-03-31T07:01:08Z"
+    "title": "Allow advanced users to override `model_type` in `AutoConfig.from_pretrained`",
+    "updated_at": "2026-03-27T14:29:53Z"
   },
   {
-    "additions": 155,
-    "author": "tarekziade",
+    "additions": 318,
+    "author": "NathanHB",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Cache to speed up ast walks, and ast tweak",
-    "changed_files": 2,
+    "body_excerpt": "Change model_dump_json() to model_dump() to avoid double JSON encoding. When using continuous batching with stream=false, the response was being double-encoded as a string instead of returning a proper JSON object. Added a UV script to run\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45012",
-    "created_at": "2026-03-26T07:10:45Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45057",
+    "created_at": "2026-03-27T13:02:59Z",
+    "deletions": 37,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45012/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45012",
+    "files_url": "https://github.com/huggingface/transformers/pull/45057/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45057",
     "labels": [],
     "merged": true,
-    "number": 45012,
+    "number": 45057,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "refactor: added cache in check_repo",
-    "updated_at": "2026-03-30T06:44:23Z"
+    "title": "[serving] Fix continuous batching JSON response serialization",
+    "updated_at": "2026-03-31T13:04:32Z"
   },
   {
-    "additions": 64,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@ydshieh Hi, can you help review? Thx! Current error (before this PR): ```bash _ PI0ForConditionalGenerationModelTest.test_flash_attn_2_inference_equivalence _ self = <tests.models.pi0.test_modeling_pi0.PI0ForConditionalGenerationModelTest\u2026",
-    "changed_files": 1,
+    "additions": 331,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This is mega long due I wanted to check benches. Its not super super huge but a win is a win",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45011",
-    "created_at": "2026-03-26T06:27:09Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45056",
+    "created_at": "2026-03-27T11:36:10Z",
+    "deletions": 28,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45011/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45011",
+    "files_url": "https://github.com/huggingface/transformers/pull/45056/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45056",
     "labels": [],
-    "merged": true,
-    "number": 45011,
+    "merged": false,
+    "number": 45056,
     "review_comments_count": 5,
-    "state": "closed",
-    "title": "skip 2 invalid test cases for pi0 model",
-    "updated_at": "2026-04-10T10:41:16Z"
+    "state": "open",
+    "title": "[`auto_docstring`] needs to be only run on __doc__ ",
+    "updated_at": "2026-03-29T12:57:04Z"
   },
   {
-    "additions": 75,
-    "author": "dagecko",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Fix: CI/CD Security Vulnerabilities in GitHub Actions Hi! [Runner Guard](https://github.com/Vigilant-LLC/runner-guard), an open-source CI/CD security scanner by [Vigilant Cyber Security](https://www.vigilantdefense.com), identified secu\u2026",
-    "changed_files": 20,
+    "additions": 3,
+    "author": "vasanthrpjan1-boop",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? When `Trainer` saves a checkpoint for a model that is not a `PreTrainedModel` (e.g. a custom `nn.Module`), it only saves the state dict but not the model config. This means `Model.from_pretrained(ckpt_path)` requir\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45010",
-    "created_at": "2026-03-26T06:19:29Z",
-    "deletions": 71,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45055",
+    "created_at": "2026-03-27T11:31:10Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45010/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45010",
+    "files_url": "https://github.com/huggingface/transformers/pull/45055/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45055",
     "labels": [],
     "merged": false,
-    "number": 45010,
+    "number": 45055,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: pin 69 unpinned action(s),extract 2 unsafe expression(s) to env vars",
-    "updated_at": "2026-03-27T22:20:57Z"
+    "state": "open",
+    "title": "Save model config in Trainer checkpoints for non-PreTrainedModel models",
+    "updated_at": "2026-03-27T11:31:10Z"
   },
   {
-    "additions": 330,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This patch improves the docstring checker implementation (redundant AST walks) and adds cache. For the AST calls, 2.3x speedup check_docstrings.py --check_all on my M1: - before : 29.3s - after: 12.6s",
-    "changed_files": 2,
+    "additions": 3,
+    "author": "hf-security-analysis[bot]",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Update `.github/workflows/update_metdata.yml` workflow configuration. cc @tarekziade @ydshieh Closes huggingface/tracking-issues#33",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45009",
-    "created_at": "2026-03-26T05:31:41Z",
-    "deletions": 47,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45054",
+    "created_at": "2026-03-27T11:16:11Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45009/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45009",
+    "files_url": "https://github.com/huggingface/transformers/pull/45054/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45054",
     "labels": [],
     "merged": true,
-    "number": 45009,
-    "review_comments_count": 12,
+    "number": 45054,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "refactor: speed up docstring checker",
-    "updated_at": "2026-03-27T07:21:11Z"
+    "title": "chore: update update_metdata.yml",
+    "updated_at": "2026-03-27T15:57:14Z"
   },
   {
-    "additions": 380,
-    "author": "Krishnachaitanyakc",
+    "additions": 1,
+    "author": "Sai-Suraj-27",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fix type annotation bugs across config classes that cause `@strict` validation errors from `huggingface_hub`. ### Bool fields mistyped as `int` (22 fixes) Fields with boolean defaults (`True`/`False`) were annotated as `int` ins\u2026",
-    "changed_files": 198,
+    "body_excerpt": "# What does this PR do? Fixes failing [`XCLIPModelIntegrationTests`](https://github.com/huggingface/transformers/actions/runs/23230643883/job/67524760869#step:14:1384). [`self.get_attributes()`](https://github.com/huggingface/transformers/\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45007",
-    "created_at": "2026-03-25T23:12:53Z",
-    "deletions": 380,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45053",
+    "created_at": "2026-03-27T11:11:18Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45007/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45007",
+    "files_url": "https://github.com/huggingface/transformers/pull/45053/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45053",
     "labels": [],
-    "merged": true,
-    "number": 45007,
+    "merged": false,
+    "number": 45053,
     "review_comments_count": 3,
     "state": "closed",
-    "title": "fix: correct type annotations across config classes for @strict validation",
-    "updated_at": "2026-04-01T17:50:09Z"
+    "title": "Fix failing `XCLIPModelIntegrationTest`",
+    "updated_at": "2026-04-13T11:05:45Z"
   },
   {
-    "additions": 3,
-    "author": "Krishnachaitanyakc",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #44869 Adds a bounds check in `_split_tokens_on_unicode()` in `tokenization_whisper.py` to handle trailing Unicode replacement characters (U+FFFD) at the end of decoded token streams without crashing with `IndexError`. ##\u2026",
+    "additions": 1,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? `.gitignore` was not updated when `mlinter` was refactored",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45006",
-    "created_at": "2026-03-25T23:03:00Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45052",
+    "created_at": "2026-03-27T10:16:17Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45006/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45006",
+    "files_url": "https://github.com/huggingface/transformers/pull/45052/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45052",
     "labels": [],
-    "merged": false,
-    "number": 45006,
+    "merged": true,
+    "number": 45052,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "fix: prevent IndexError in Whisper timestamp decode on trailing replacement char",
-    "updated_at": "2026-04-06T01:03:09Z"
+    "state": "closed",
+    "title": "chore: Fix mlinter cache location",
+    "updated_at": "2026-03-27T10:26:38Z"
   },
   {
-    "additions": 10,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following failing tests were identified and fixed in this PR (grouped them together since they share related root causes OR the code changes were extremely minimal and didn't warrant separate PRs): \u2192 **Phi-3**\u2026",
-    "changed_files": 2,
+    "additions": 2,
+    "author": "albertvillanova",
+    "author_association": "MEMBER",
+    "body_excerpt": "Fix NaN weights on non-rank-0 FSDP processes by using `zeros_like` instead of `empty_like` in `_move_missing_keys_from_meta_to_device` Follow-up to: - #44473 See related downstream issue in `trl` : - https://github.com/huggingface/trl/issu\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45004",
-    "created_at": "2026-03-25T19:58:57Z",
-    "deletions": 6,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45050",
+    "created_at": "2026-03-27T09:19:32Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45004/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45004",
+    "files_url": "https://github.com/huggingface/transformers/pull/45050/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45050",
     "labels": [],
     "merged": true,
-    "number": 45004,
-    "review_comments_count": 3,
+    "number": 45050,
+    "review_comments_count": 8,
     "state": "closed",
-    "title": "fix(testing): Fix Parakeet, Evolla, Pi0, and Phi-3 test failures on main CI",
-    "updated_at": "2026-03-27T15:06:52Z"
+    "title": "Fix NaN weights on non-rank-0 FSDP processes",
+    "updated_at": "2026-04-13T13:35:35Z"
   },
   {
-    "additions": 1,
-    "author": "hmellor",
+    "additions": 6,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "`None` is a valid value that can be used to disable chunked attention in `DynamicCache` and Flex Attention. hf.co/morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct is an example of a checkpoint which does this.",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/45030 After the config validation, all validations are now run after config is initialized. So this config has been wrong from the beginning but we didn't com\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45002",
-    "created_at": "2026-03-25T17:40:14Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45049",
+    "created_at": "2026-03-27T08:46:46Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45002/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45002",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45049/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45049",
+    "labels": [
+      "for patch"
+    ],
     "merged": true,
-    "number": 45002,
+    "number": 45049,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix type hint for `attention_chunk_size` in `Llama4TextConfig`",
-    "updated_at": "2026-03-25T20:42:11Z"
+    "title": "Fix when RoPE params are in kwargs",
+    "updated_at": "2026-03-27T16:28:13Z"
   },
   {
-    "additions": 20,
+    "additions": 4,
     "author": "Sai-Suraj-27",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? For [torch>=2.10.0](https://docs.pytorch.org/docs/2.10/generated/torch.nn.functional.grouped_mm.html#torch-nn-functional-grouped-mm), the minimum CUDA compute capability requirement for `torch.nn.functional.grouped_\u2026",
+    "body_excerpt": "# What does this PR do? Fixes failing [`SmolLM3IntegrationTest:test_model_3b_long_prompt`](https://github.com/huggingface/transformers/actions/runs/23629638266/job/68826332952#step:14:216). `SmolLM3` has [`do_sample=True` by default](https\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45001",
-    "created_at": "2026-03-25T17:00:28Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45048",
+    "created_at": "2026-03-27T08:45:26Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45001/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45001",
+    "files_url": "https://github.com/huggingface/transformers/pull/45048/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45048",
     "labels": [],
     "merged": true,
-    "number": 45001,
-    "review_comments_count": 2,
+    "number": 45048,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add cuda compatibility check for using `grouped_mm`",
-    "updated_at": "2026-04-10T07:54:45Z"
+    "title": "Fix failing `SmolLM3IntegrationTest`",
+    "updated_at": "2026-03-27T14:18:05Z"
   },
   {
-    "additions": 22,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per title, after https://github.com/huggingface/transformers/pull/44976 users will be seeing a `missing_weights - lm_head not found` error even though the model doesn't use an lm head On the way also deleted unne\u2026",
-    "changed_files": 8,
+    "additions": 503,
+    "author": "Akshay404error",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "Fixes #44568 \u2014 restores add_special_tokens behavior for mDeBERTa tokenizer and non-persistent buffers in v5 ## What does this PR do? This PR fixes two v5 regressions: 1. `add_special_tokens=True` no longer added BOS/EOS tokens for the `mic\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/45000",
-    "created_at": "2026-03-25T16:28:55Z",
-    "deletions": 109,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45047",
+    "created_at": "2026-03-27T08:33:01Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/45000/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/45000",
-    "labels": [],
-    "merged": true,
-    "number": 45000,
+    "files_url": "https://github.com/huggingface/transformers/pull/45047/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45047",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 45047,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Embedding VLMs don't need a head",
-    "updated_at": "2026-03-27T09:57:18Z"
+    "title": "fix: restore add_special_tokens behavior for mDeBERTa tokenizer and n\u2026",
+    "updated_at": "2026-03-27T13:15:58Z"
   },
   {
-    "additions": 2450,
-    "author": "itazap",
+    "additions": 548,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "## Summary - Auto-generated modular integration for `sarvam` - `modular_sarvam.py` written by Claude Opus 4.6 guided by `modular_model_detector.py` - `modeling_sarvam.py` regenerated from modular via `modular_model_converter.py` ## Test pl\u2026",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Investigate speedups on modular conversion",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44999",
-    "created_at": "2026-03-25T16:21:37Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45046",
+    "created_at": "2026-03-27T08:31:58Z",
+    "deletions": 42,
     "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44999/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44999",
+    "files_url": "https://github.com/huggingface/transformers/pull/45046/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45046",
+    "labels": [],
+    "merged": false,
+    "number": 45046,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "investigate modular conversion speedups",
+    "updated_at": "2026-03-27T16:33:45Z"
+  },
+  {
+    "additions": 5095,
+    "author": "Lidang-Jiang",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Isolate dependencies, make PIL independant from Torchvision backend Fixes #45042 PR #45029 added `@requires(backends=(\"vision\", \"torch\", \"torchvision\"))` to 67 PIL backend `image_processing_pil_*.py` files. This causes PIL backend class\u2026",
+    "changed_files": 188,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45045",
+    "created_at": "2026-03-27T08:19:19Z",
+    "deletions": 1270,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45045/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45045",
+    "labels": [
+      "for patch"
+    ],
+    "merged": true,
+    "number": 45045,
+    "review_comments_count": 24,
+    "state": "closed",
+    "title": "[Bugfix] Remove incorrect torchvision requirement from PIL backend image processors",
+    "updated_at": "2026-03-30T07:25:49Z"
+  },
+  {
+    "additions": 35,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fix issue in https://github.com/huggingface/transformers/issues/44792. @zucchini-nlp @ydshieh pls help review, thx!",
+    "changed_files": 3,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45044",
+    "created_at": "2026-03-27T07:50:21Z",
+    "deletions": 26,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45044/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45044",
     "labels": [],
-    "merged": false,
-    "number": 44999,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 45044,
+    "review_comments_count": 11,
     "state": "closed",
-    "title": "Add sarvam model",
-    "updated_at": "2026-03-30T23:21:49Z"
+    "title": "fix bug for janus model image generation",
+    "updated_at": "2026-04-02T02:46:14Z"
   },
   {
-    "additions": 1179,
-    "author": "itazap",
-    "author_association": "MEMBER",
-    "body_excerpt": "## Summary - Auto-generated modular integration for `sarvam` - `modular_sarvam.py` written by Claude Opus 4.6 guided by `modular_model_detector.py` - `modeling_sarvam.py` regenerated from modular via `modular_model_converter.py` ## Test pl\u2026",
-    "changed_files": 4,
+    "additions": 31,
+    "author": "Lidang-Jiang",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #45003. `_can_set_attn_implementation` and `_can_set_experts_implementation` in `modeling_utils.py` crash with `KeyError` when `cls.__module__` is absent from `sys.modules`. This happens in real-world scenari\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44997",
-    "created_at": "2026-03-25T14:23:13Z",
-    "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44997/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44997",
-    "labels": [],
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45043",
+    "created_at": "2026-03-27T06:07:20Z",
+    "deletions": 4,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45043/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45043",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44997,
+    "number": 45043,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add sarvam model",
-    "updated_at": "2026-03-25T14:35:45Z"
+    "title": "[Bugfix] Use sys.modules.get() to avoid KeyError in modeling_utils",
+    "updated_at": "2026-03-27T13:02:19Z"
   },
   {
-    "additions": 286,
-    "author": "3outeille",
-    "author_association": "MEMBER",
-    "body_excerpt": "- Introduce `DistributedConfig` - `DistributedConfig(tp_size=2, fsdp_size=2) # plans default to \"auto\"` replaces passing separate `tp_plan, tp_size, fsdp_plan kwargs`. Sizes auto-fill (specify one, the other defaults to 1). Plans default t\u2026",
-    "changed_files": 10,
+    "additions": 677,
+    "author": "JJJYmmm",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR addresses the performance issues observed with nn.Conv3d across different PyTorch/cuDNN, such as https://github.com/vllm-project/vllm/pull/27418, https://mp.weixin.qq.com/s/hKRIpB561EdrMY8cbg1hEw. We replace\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44996",
-    "created_at": "2026-03-25T14:20:25Z",
-    "deletions": 283,
+    "comments_count": 14,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45041",
+    "created_at": "2026-03-27T03:50:54Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44996/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44996",
+    "files_url": "https://github.com/huggingface/transformers/pull/45041/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45041",
     "labels": [],
     "merged": true,
-    "number": 44996,
-    "review_comments_count": 0,
+    "number": 45041,
+    "review_comments_count": 61,
     "state": "closed",
-    "title": " from_pretrained distributed refactor (FSDP2 + TP)",
-    "updated_at": "2026-03-26T15:32:27Z"
+    "title": "[inference_fusion] convert conv3d patch embed to linear",
+    "updated_at": "2026-04-13T15:30:39Z"
   },
   {
-    "additions": 3639,
-    "author": "itazap",
-    "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 8,
+    "additions": 13,
+    "author": "sywangyi",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "- vision models: @yonigozlan @molbap - CIs: @ydshieh when running tests/models/video_llama_3/test_modeling_video_llama_3.py::VideoLlama3IntegrationTest all fail cause by lm_head.weight is missing.",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44994",
-    "created_at": "2026-03-25T14:02:50Z",
-    "deletions": 242,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45040",
+    "created_at": "2026-03-27T02:56:58Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44994/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44994",
+    "files_url": "https://github.com/huggingface/transformers/pull/45040/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45040",
     "labels": [],
     "merged": false,
-    "number": 44994,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add sarvam model",
-    "updated_at": "2026-03-25T14:04:38Z"
+    "number": 45040,
+    "review_comments_count": 3,
+    "state": "open",
+    "title": "Llama3 video fix",
+    "updated_at": "2026-03-27T10:55:26Z"
   },
   {
-    "additions": 583,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? `make check-repo` can be quite slow, this patch adds file-level cache to speed up checks. We get up to a 27x speedup - cold cache : 46s - warm cache : 1.6s",
-    "changed_files": 20,
+    "additions": 26,
+    "author": "Lidang-Jiang",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes #44716 - Propagate `interpolate_pos_encoding` parameter through `PixioModel.forward()`, `PixioBackbone.forward()`, and `PixioEmbeddings.forward()` down to `PixioPatchEmbeddings.forward()` - Follows the same pattern used by\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44992",
-    "created_at": "2026-03-25T11:40:46Z",
-    "deletions": 48,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45039",
+    "created_at": "2026-03-27T02:54:31Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44992/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44992",
+    "files_url": "https://github.com/huggingface/transformers/pull/45039/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45039",
     "labels": [],
-    "merged": true,
-    "number": 44992,
-    "review_comments_count": 4,
+    "merged": false,
+    "number": 45039,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "refactoring: speedup static checks with disk cache",
-    "updated_at": "2026-03-31T12:34:20Z"
+    "title": "[Bugfix] Propagate interpolate_pos_encoding through Pixio model",
+    "updated_at": "2026-03-27T12:57:30Z"
   },
   {
-    "additions": 8,
+    "additions": 1,
     "author": "ArthurZucker",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - BC for check model inputs",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44990",
-    "created_at": "2026-03-25T10:26:20Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45038",
+    "created_at": "2026-03-27T00:18:35Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44990/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44990",
+    "files_url": "https://github.com/huggingface/transformers/pull/45038/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45038",
     "labels": [],
     "merged": true,
-    "number": 44990,
+    "number": 45038,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "More small vllm fixes",
-    "updated_at": "2026-03-25T13:05:44Z"
+    "title": "style was missing sorry @ydshieh :)",
+    "updated_at": "2026-03-27T00:28:23Z"
   },
   {
-    "additions": 542,
-    "author": "3outeille",
-    "author_association": "MEMBER",
-    "body_excerpt": "# Distributed Training API ## Goal ```python # torchrun --nproc_per_node=4 train_fsdp_tp.py import os import torch from torch.utils.data import DataLoader from datasets import load_dataset from transformers import AutoModelForCausalLM, Aut\u2026",
-    "changed_files": 6,
+    "additions": 1,
+    "author": "asuryateja",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "\u2026_interface docs The `custom_attention` function definition in the attention_interface documentation was missing a colon at the end of the return type annotation, making it invalid Python syntax. # What does this PR do? <!-- Congratulation\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44989",
-    "created_at": "2026-03-25T09:10:02Z",
-    "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44989/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44989",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45037",
+    "created_at": "2026-03-26T23:48:04Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45037/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45037",
     "labels": [],
     "merged": false,
-    "number": 44989,
+    "number": 45037,
     "review_comments_count": 0,
     "state": "open",
-    "title": "\ud83d\udea8 Distributed training API",
-    "updated_at": "2026-04-13T16:44:35Z"
+    "title": "add missing colon in custom_attention function signature in attention\u2026",
+    "updated_at": "2026-03-27T00:17:55Z"
   },
   {
-    "additions": 730,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Adds Rule 15 ``` if _tied_weights_keys is present and non-empty in modeling -> Config MUST contain the tie_word_embeddings field ```",
-    "changed_files": 16,
+    "additions": 15,
+    "author": "matdou",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #45030 Configs like `tiny-random/glm-4v` store `rope_theta` at the top level of `config.json` alongside a `rope_scaling` dict (legacy format). For config classes that don't declare `rope_parameters` as a datac\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 17,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44988",
-    "created_at": "2026-03-25T07:08:20Z",
-    "deletions": 2,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45036",
+    "created_at": "2026-03-26T23:00:35Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44988/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44988",
+    "files_url": "https://github.com/huggingface/transformers/pull/45036/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45036",
     "labels": [],
     "merged": false,
-    "number": 44988,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "typing: rule 15 - checks for tie_word_embeddings presence",
-    "updated_at": "2026-04-14T07:07:02Z"
+    "number": 45036,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "[fix] BC for legacy configs with top-level rope_theta when rope_parameters is set via rope_scaling",
+    "updated_at": "2026-03-28T23:47:33Z"
   },
   {
-    "additions": 0,
-    "author": "Krishnachaitanyakc",
+    "additions": 1,
+    "author": "asuryateja",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #44855 On Python 3.13, placing a `# Copied from` comment between `@torch.jit.script` and the function definition causes an `IndentationError`. This happens because `torch.jit.script` calls `inspect.getsource()` followed by\u2026",
-    "changed_files": 2,
+    "body_excerpt": "The conversion operations table was missing PermuteForRope. Added it with its reverse (itself), consistent with how other operations are documented. PermuteForRope is self-inverse applying it twice returns the original tensor layout. # Wha\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44986",
-    "created_at": "2026-03-25T03:18:31Z",
-    "deletions": 6,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45035",
+    "created_at": "2026-03-26T21:05:17Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44986/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44986",
+    "files_url": "https://github.com/huggingface/transformers/pull/45035/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45035",
     "labels": [],
     "merged": true,
-    "number": 44986,
+    "number": 45035,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: remove Copied from comments between @torch.jit.script and def for Python 3.13 compat",
-    "updated_at": "2026-03-25T13:39:54Z"
-  },
-  {
-    "additions": 2,
-    "author": "Krishnachaitanyakc",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #44913 When creating a `GPTNeoXConfig` (or `GPTNeoXJapaneseConfig`) with a non-default `rotary_pct`, the value is lost after a `save_pretrained` / `from_pretrained` round-trip. This happens because `convert_rope_params_to_\u2026",
-    "changed_files": 2,
+    "title": "docs: add PermuteForRope to conversion operations reverse table",
+    "updated_at": "2026-03-26T22:09:53Z"
+  },
+  {
+    "additions": 113,
+    "author": "sdharani91",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This is a follow up to https://github.com/huggingface/transformers/pull/44867 This PR fixes Qwen3.5 padding-free packed inputs on the linear-attention fast path by consuming collator-provided packed metadata. The li\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44985",
-    "created_at": "2026-03-25T02:15:04Z",
-    "deletions": 2,
+    "comments_count": 12,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45034",
+    "created_at": "2026-03-26T20:52:51Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44985/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44985",
+    "files_url": "https://github.com/huggingface/transformers/pull/45034/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45034",
     "labels": [],
-    "merged": true,
-    "number": 44985,
-    "review_comments_count": 1,
-    "state": "closed",
-    "title": "fix: preserve rotary_pct across save/load cycle in GPTNeoX configs",
-    "updated_at": "2026-03-27T09:19:46Z"
+    "merged": false,
+    "number": 45034,
+    "review_comments_count": 39,
+    "state": "open",
+    "title": "Pass packed boundary metadata to Qwen3.5 linear-attention fast kernels from data collator",
+    "updated_at": "2026-04-15T13:57:34Z"
   },
   {
-    "additions": 2,
-    "author": "Butanium",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? `maybe_autocast` calls `torch.is_autocast_enabled(device_type)` which raises a `RuntimeError` when `device_type` is `\"meta\"`: ``` RuntimeError: unknown device type for autocast in get_autocast_dispatch_key_from_dev\u2026",
+    "additions": 3,
+    "author": "hmellor",
+    "author_association": "MEMBER",
+    "body_excerpt": "In https://github.com/huggingface/transformers/pull/43514, `BaseImageProcessorFast` became `BaseImageProcessor` and `_further_process_kwargs` was renamed to `_standardize_kwargs` This PR adds some BC for the old name of this method.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44984",
-    "created_at": "2026-03-25T01:39:23Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45033",
+    "created_at": "2026-03-26T20:36:40Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44984/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44984",
+    "files_url": "https://github.com/huggingface/transformers/pull/45033/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45033",
     "labels": [],
     "merged": true,
-    "number": 44984,
+    "number": 45033,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix `maybe_autocast` crashing on meta device tensors",
-    "updated_at": "2026-03-25T17:45:03Z"
+    "title": "Add BC for `_further_process_kwargs`",
+    "updated_at": "2026-03-26T21:01:32Z"
   },
   {
-    "additions": 26,
-    "author": "Hyungkeun-Park-Nota",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes `save_pretrained()` for models loaded with `dequantize=True`. `save_pretrained` calls `reverse_op` on all weight conversion operations from loading. Dequantize ops (`Mxfp4Dequantize`, `Fp8Dequantize`, `MetalD\u2026",
-    "changed_files": 4,
+    "additions": 163,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Use multi runners to get new failing tests in a CI run.",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44983",
-    "created_at": "2026-03-25T01:19:59Z",
-    "deletions": 3,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45032",
+    "created_at": "2026-03-26T18:48:18Z",
+    "deletions": 67,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44983/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44983",
+    "files_url": "https://github.com/huggingface/transformers/pull/45032/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45032",
     "labels": [],
     "merged": true,
-    "number": 44983,
-    "review_comments_count": 6,
+    "number": 45032,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: add identity reverse_op to dequantize ops for save_pretrained",
-    "updated_at": "2026-03-27T17:21:18Z"
+    "title": "Use multi runners to check new failing tests in a CI run",
+    "updated_at": "2026-03-26T18:59:08Z"
   },
   {
-    "additions": 108,
-    "author": "AkshajKashyap",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #43039 ## What does this PR do? When `prediction_loss_only=True` during evaluation and `use_liger_kernel=True`, `Trainer.prediction_step` now passes `skip_logits=True` to the model forward if the forward signature supports it and lab\u2026",
+    "additions": 6,
+    "author": "tomaarsen",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Follow-up for https://github.com/huggingface/transformers/pull/44931, which added weight tying for Camembert. Only the CamembertForMaskedLM class had the right _tied_weights_keys, the CamembertForCausalLM had the in\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44981",
-    "created_at": "2026-03-25T00:38:02Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45031",
+    "created_at": "2026-03-26T18:28:54Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44981/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44981",
+    "files_url": "https://github.com/huggingface/transformers/pull/45031/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45031",
     "labels": [],
-    "merged": false,
-    "number": 44981,
-    "review_comments_count": 6,
-    "state": "open",
-    "title": "Trainer: set skip_logits for loss-only eval when liger enabled",
-    "updated_at": "2026-04-09T15:31:50Z"
+    "merged": true,
+    "number": 45031,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[`fix`] Use the correct _tied_weights_keys for CamembertForCausalLM",
+    "updated_at": "2026-03-26T18:57:29Z"
   },
   {
-    "additions": 6,
-    "author": "kallewoof",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Pre-patch unnecessarily breaks merging a LoRA adapter with a model using CUDA_VISIBLE_DEVICES= e.g. when VRAM is insufficient. It also breaks non-cuda machine operations (such as merging). # What does this PR do? This PR un-breaks `CUDA_VI\u2026",
-    "changed_files": 6,
+    "additions": 478,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Release workflow is failing",
+    "changed_files": 101,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44980",
-    "created_at": "2026-03-24T23:50:07Z",
-    "deletions": 6,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45029",
+    "created_at": "2026-03-26T18:04:37Z",
+    "deletions": 768,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44980/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44980",
+    "files_url": "https://github.com/huggingface/transformers/pull/45029/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45029",
     "labels": [],
-    "merged": false,
-    "number": 44980,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 45029,
+    "review_comments_count": 7,
     "state": "closed",
-    "title": "bug-fix: do not assume torch.cuda is available when setting up norm values, even if flash linear attention is available",
-    "updated_at": "2026-03-27T13:25:18Z"
+    "title": "Fix release full",
+    "updated_at": "2026-03-27T06:34:00Z"
   },
   {
-    "additions": 492,
-    "author": "michaelbenayoun",
+    "additions": 3314,
+    "author": "3outeille",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Introduces `src/transformers/module_fusion.py`, a utility for fusing adjacent submodules in a model into a single FusedModule that executes them as a chain in one forward pass. The key components are: - `RegistryCol\u2026",
-    "changed_files": 2,
+    "body_excerpt": "- TODO - how will dtensor works with quantization ? - how will dtensor works with kernels ? - Needs end to end test (combine `verify_all_loss` -> `training` with saving + loading back for generate ?) - double check Save FSDP + TP - do test\u2026",
+    "changed_files": 218,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44979",
-    "created_at": "2026-03-24T22:33:31Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44979/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44979",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45028",
+    "created_at": "2026-03-26T17:50:07Z",
+    "deletions": 3582,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/45028/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45028",
     "labels": [],
     "merged": false,
-    "number": 44979,
-    "review_comments_count": 0,
+    "number": 45028,
+    "review_comments_count": 26,
     "state": "open",
-    "title": "Module Fusion API",
-    "updated_at": "2026-04-13T19:46:19Z"
+    "title": "TP refactor for FSDP + TP integration",
+    "updated_at": "2026-04-16T15:08:35Z"
   },
   {
-    "additions": 4,
-    "author": "cjkindel",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? `_can_set_attn_implementation` and `_can_set_experts_implementation` both do a direct subscript lookup into `sys.modules`: ```python class_module = sys.modules[cls.__module__] ``` If the module is not registered und\u2026",
-    "changed_files": 1,
+    "additions": 714,
+    "author": "remi-or",
+    "author_association": "MEMBER",
+    "body_excerpt": "# Summary This PR adds per-request logits processors and overalls the way CB handles logits processors. It introduces batched logits processing with per-request parameters for continuous batching, enabling each request in a batch to use di\u2026",
+    "changed_files": 11,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44978",
-    "created_at": "2026-03-24T21:01:11Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45026",
+    "created_at": "2026-03-26T17:00:07Z",
+    "deletions": 202,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44978/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44978",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44978,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/45026/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45026",
+    "labels": [],
+    "merged": true,
+    "number": 45026,
+    "review_comments_count": 32,
     "state": "closed",
-    "title": "fix: handle absent sys.modules entry in modeling_utils",
-    "updated_at": "2026-03-26T12:25:31Z"
+    "title": "[CB] Add per-request logits processors",
+    "updated_at": "2026-04-03T16:44:11Z"
   },
   {
     "additions": 2,
-    "author": "hmellor",
-    "author_association": "MEMBER",
-    "body_excerpt": "- Adds a type hint to `ModernVBertForMaskedLM.__init__` - Removes `tie_word_embeddings` from `Qwen2VLTextConfig` (and therefore also `Qwen2_5_VLTextConfig`) because it's not valid for these models - Remove hack from `ColQwen2Config` (and t\u2026",
-    "changed_files": 6,
+    "author": "layla1824",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "This PR adds a clarification comment regarding the behavior of rotary_pct. Currently, rotary_pct may reset to its default value (0.25) after reload due to the use of kwargs.pop. This note helps developers better understand this behavior.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44976",
-    "created_at": "2026-03-24T19:26:33Z",
-    "deletions": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45025",
+    "created_at": "2026-03-26T16:48:53Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44976/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44976",
+    "files_url": "https://github.com/huggingface/transformers/pull/45025/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45025",
     "labels": [],
-    "merged": true,
-    "number": 44976,
-    "review_comments_count": 3,
+    "merged": false,
+    "number": 45025,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix tie_word_embedding issues with `Qwen2VL`",
-    "updated_at": "2026-03-24T20:55:15Z"
+    "title": "Update configuration_Clarify rotary_pct reset behavior in GPTNeoXConfiggpt_neox.py",
+    "updated_at": "2026-03-27T09:05:48Z"
   },
   {
-    "additions": 6971,
-    "author": "philippguevorguian",
-    "author_association": "NONE",
-    "body_excerpt": null,
-    "changed_files": 20,
+    "additions": 1,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? `tf` and `flax` are long gone (unknown extras generate only warnings so this slipped through the cracks)",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44975",
-    "created_at": "2026-03-24T17:12:31Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45024",
+    "created_at": "2026-03-26T16:21:30Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44975/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44975",
+    "files_url": "https://github.com/huggingface/transformers/pull/45024/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45024",
     "labels": [],
-    "merged": false,
-    "number": 44975,
+    "merged": true,
+    "number": 45024,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: rebase main; clean config reads, ImageProcessor backend, misc cleanup",
-    "updated_at": "2026-03-24T17:13:42Z"
+    "title": "chore: remove old extras",
+    "updated_at": "2026-03-27T11:16:03Z"
   },
   {
-    "additions": 428,
-    "author": "3outeille",
+    "additions": 2529,
+    "author": "eustlb",
     "author_association": "MEMBER",
-    "body_excerpt": "TODO: - Saving seems to take a bit of time tho. Need investigation - Need to check if it works in 1D (FSDP or TP)and 2D (FSDP + TP). Running the script from https://github.com/huggingface/transformers/pull/44996 ``` (env_pr-44974-fsdp-core\u2026",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? **Integration notes:** For now, this integration does not load mel filters from the checkpoint. The original model was trained backpropagating gradients in it, but we saw previously (with parakeet-ctc) that this doe\u2026",
+    "changed_files": 19,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44974",
-    "created_at": "2026-03-24T16:13:25Z",
-    "deletions": 82,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45023",
+    "created_at": "2026-03-26T15:58:08Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44974/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44974",
-    "labels": [],
-    "merged": false,
-    "number": 44974,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "Refactor core_model_loading to support FSDP shard-on-read loading",
-    "updated_at": "2026-04-14T14:38:45Z"
+    "files_url": "https://github.com/huggingface/transformers/pull/45023/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45023",
+    "labels": [
+      "New model",
+      "Audio"
+    ],
+    "merged": true,
+    "number": 45023,
+    "review_comments_count": 13,
+    "state": "closed",
+    "title": "Add cohere asr",
+    "updated_at": "2026-03-26T22:48:16Z"
   },
   {
-    "additions": 22,
-    "author": "andylizf",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Adds `.item()` to `max_seqlen = (cu_seqlens[1:] - cu_seqlens[:-1]).max()` in all vision attention modules that pass this value to `flash_attn_varlen_func`. ### Context On **released versions** (e.g. 4.52.4), using\u2026",
-    "changed_files": 19,
+    "additions": 1,
+    "author": "popotest",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44973",
-    "created_at": "2026-03-24T15:42:32Z",
-    "deletions": 22,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45022",
+    "created_at": "2026-03-26T15:33:19Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44973/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44973",
+    "files_url": "https://github.com/huggingface/transformers/pull/45022/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45022",
     "labels": [],
     "merged": false,
-    "number": 44973,
+    "number": 45022,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix max_seqlen type in vision attention for torch.compile + FA2",
-    "updated_at": "2026-03-25T14:12:50Z"
+    "state": "closed",
+    "title": "Update _config.py",
+    "updated_at": "2026-03-27T13:17:36Z"
   },
   {
-    "additions": 17,
-    "author": "Abdennacer-Badaoui",
-    "author_association": "MEMBER",
-    "body_excerpt": "As per title. Updating Gemma3/Gemma3n expectations.",
-    "changed_files": 3,
+    "additions": 2,
+    "author": "hf-security-analysis[bot]",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Update `.github/workflows/anti-slop.yml` workflow configuration. cc @tarekziade Closes huggingface/tracking-issues#30",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44972",
-    "created_at": "2026-03-24T15:11:50Z",
-    "deletions": 12,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45021",
+    "created_at": "2026-03-26T13:40:18Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44972/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44972",
+    "files_url": "https://github.com/huggingface/transformers/pull/45021/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45021",
     "labels": [],
-    "merged": true,
-    "number": 44972,
-    "review_comments_count": 10,
+    "merged": false,
+    "number": 45021,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[AMD CI] Gemma3/Gemma3n Expectations",
-    "updated_at": "2026-03-24T16:30:03Z"
+    "title": "chore: update anti-slop.yml",
+    "updated_at": "2026-03-26T13:53:57Z"
   },
   {
-    "additions": 0,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Removed the tokenizer_class attr was never there to begin with, and kwargs are now supported. This was failing some test on vllm ci. Fixes https://buildkite.com/vllm/ci/builds/57601/steps/canvas?sid=019d1aec-aa5a-41\u2026",
+    "additions": 21,
+    "author": "javierdejesusda",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #44877 Loading `ibm-granite/granite-4.0-1b-speech` fails with `StrictDataclassFieldValidationError` because its config.json stores `embedding_multiplier` and `logits_scaling` as integers (e.g. `12`, `8`), but\u2026",
     "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44971",
-    "created_at": "2026-03-24T14:59:36Z",
-    "deletions": 11,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45019",
+    "created_at": "2026-03-26T11:19:19Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44971/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44971",
+    "files_url": "https://github.com/huggingface/transformers/pull/45019/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45019",
     "labels": [],
     "merged": true,
-    "number": 44971,
-    "review_comments_count": 1,
+    "number": 45019,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "[ `vllm x v5`] nit",
-    "updated_at": "2026-03-24T17:40:05Z"
+    "title": "Fix GraniteConfig type hints to accept int for multiplier fields",
+    "updated_at": "2026-03-27T09:30:17Z"
   },
   {
-    "additions": 20,
-    "author": "IlyasMoutawwakil",
+    "additions": 1814,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? As per title, deletes long lines of auto-mapping and dynamically builds it from code. The whole auto-generation runs with AST and takes ~5 sec on my machine, so we don't infer anything at run-time. Instead it is bui\u2026",
+    "changed_files": 68,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44970",
-    "created_at": "2026-03-24T13:49:21Z",
-    "deletions": 76,
+    "comments_count": 16,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45018",
+    "created_at": "2026-03-26T11:18:13Z",
+    "deletions": 1437,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44970/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44970",
+    "files_url": "https://github.com/huggingface/transformers/pull/45018/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45018",
     "labels": [],
     "merged": true,
-    "number": 44970,
-    "review_comments_count": 1,
+    "number": 45018,
+    "review_comments_count": 7,
     "state": "closed",
-    "title": "Fix CPU 16 bytes alignment issue using equivalent fallback",
-    "updated_at": "2026-03-25T09:01:03Z"
+    "title": "Dynamic auto mapping",
+    "updated_at": "2026-04-16T12:54:59Z"
   },
   {
-    "additions": 4,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Extends the CI so we can use Make and read toml files",
-    "changed_files": 3,
+    "additions": 571,
+    "author": "JaredforReal",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? ### Get the rope operation right Before: NeoX split-half style After: GPT-J/interleaved style(`interleaved=True` same as `is_neox_style=Flase`) the right one ### Get rid of `F.relu` Reason: - `F.relu` works with `ac\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44968",
-    "created_at": "2026-03-24T11:43:24Z",
-    "deletions": 2,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45017",
+    "created_at": "2026-03-26T09:21:10Z",
+    "deletions": 49,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44968/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44968",
+    "files_url": "https://github.com/huggingface/transformers/pull/45017/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45017",
     "labels": [],
     "merged": false,
-    "number": 44968,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Added Make to the docker and `tomli` to `.[quality]`",
-    "updated_at": "2026-03-24T15:06:29Z"
+    "number": 45017,
+    "review_comments_count": 5,
+    "state": "open",
+    "title": "[WIP][Fix] GLM 5 set `apply_rotary_pos_emb` to `is_neox_style=False` && remove `F.relu()`",
+    "updated_at": "2026-04-14T09:36:01Z"
   },
   {
-    "additions": 87,
-    "author": "Qubitium",
+    "additions": 64,
+    "author": "inisis",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fix: FA kernel launches currently are not thread-safe (nogil) in multi-gpu env. This simple patch fixes the issue. ```py # Set the correct CUDA context before launching the FlashAttention kernel. with torch.cuda.dev\u2026",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44967",
-    "created_at": "2026-03-24T11:33:45Z",
-    "deletions": 84,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44967/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44967",
-    "labels": [],
-    "merged": false,
-    "number": 44967,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[WIP] Fix FA kernel launch needs correct cuda device ctx in multi-gpu env",
-    "updated_at": "2026-04-08T15:10:47Z"
-  },
-  {
-    "additions": 8,
-    "author": "pramilajangid",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #44964 ## Summary This PR restores backward compatibility for `CommonKwargs` in `transformers.processing_utils`, which is still referenced by some remote processor implementations. ## Problem After the typed-dict cleanup (commit `533\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44966",
-    "created_at": "2026-03-24T11:06:57Z",
-    "deletions": 0,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45016",
+    "created_at": "2026-03-26T09:09:41Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44966/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44966",
+    "files_url": "https://github.com/huggingface/transformers/pull/45016/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45016",
     "labels": [],
     "merged": false,
-    "number": 44966,
+    "number": 45016,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix backward compatibility for CommonKwargs in processing_utils (brea\u2026",
-    "updated_at": "2026-03-24T12:48:44Z"
+    "title": "fix: glm5 inference bug",
+    "updated_at": "2026-03-26T12:10:21Z"
   },
   {
-    "additions": 37,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "additions": 55,
+    "author": "pnehete23",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes a `KeyError` in `_can_set_attn_implementation` and `_can_set_experts_implementation` when a model's module is absent from `sys.modules`. Fixes #45003 ## Root Cause Both `_can_set_attn_implementation` (line 19\u2026",
     "changed_files": 2,
     "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44965",
-    "created_at": "2026-03-24T10:59:31Z",
-    "deletions": 32,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45015",
+    "created_at": "2026-03-26T08:52:10Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44965/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44965",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/45015/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45015",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44965,
+    "number": 45015,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "try",
-    "updated_at": "2026-03-24T11:19:27Z"
+    "state": "closed",
+    "title": "fix: guard sys.modules access in _can_set_attn/experts_implementation",
+    "updated_at": "2026-03-26T12:15:50Z"
   },
   {
-    "additions": 3,
-    "author": "josh-kean",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes an import in src/transformers/video_processing_utils.py that was causing the main build to fail Fixes # 44933 ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs\u2026",
+    "additions": 16,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? #30674 refactors the way we obtain CircleCI test files to run for each job. It always puts [\"tests\"] for `tests_hub`, so each commit of each PR will run it, no matter if there is any change to codebase. Let's reduce\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44958",
-    "created_at": "2026-03-23T20:07:09Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45014",
+    "created_at": "2026-03-26T08:52:04Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44958/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44958",
+    "files_url": "https://github.com/huggingface/transformers/pull/45014/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45014",
     "labels": [],
-    "merged": false,
-    "number": 44958,
+    "merged": true,
+    "number": 45014,
     "review_comments_count": 1,
-    "state": "open",
-    "title": "fixed import error with PILImageResampling",
-    "updated_at": "2026-03-24T13:53:00Z"
+    "state": "closed",
+    "title": "Don't run `tests_hub` if no tests found",
+    "updated_at": "2026-03-26T09:32:39Z"
   },
   {
-    "additions": 1272,
-    "author": "bigshanedogg",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds native Transformers support for **[HyperCLOVA X SEED Think 14B](https://huggingface.co/naver-hyperclovax/HyperCLOVAX-SEED-Think-14B)**, a 14.74B-parameter Korean reasoning LLM developed by NAVER Cloud. - relate\u2026",
-    "changed_files": 12,
+    "additions": 325,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? - Adds a new checker in `make chek-repo` that will `import transformers` and count the number of imported module. - Lazy import of torch when doing `import transformers` The change will reduce the import time from ~\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44956",
-    "created_at": "2026-03-23T19:34:30Z",
-    "deletions": 0,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45013",
+    "created_at": "2026-03-26T07:47:23Z",
+    "deletions": 49,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44956/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44956",
+    "files_url": "https://github.com/huggingface/transformers/pull/45013/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45013",
     "labels": [],
-    "merged": false,
-    "number": 44956,
-    "review_comments_count": 24,
-    "state": "open",
-    "title": "Add HyperCLOVAX SEED Think 14B",
-    "updated_at": "2026-04-12T06:17:53Z"
+    "merged": true,
+    "number": 45013,
+    "review_comments_count": 3,
+    "state": "closed",
+    "title": "feature: added import complexity checker",
+    "updated_at": "2026-03-31T07:01:08Z"
   },
   {
-    "additions": 0,
-    "author": "stevhliu",
+    "additions": 155,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "removes outdated qa pipeline reference",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Cache to speed up ast walks, and ast tweak",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44954",
-    "created_at": "2026-03-23T17:20:37Z",
-    "deletions": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45012",
+    "created_at": "2026-03-26T07:10:45Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44954/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44954",
+    "files_url": "https://github.com/huggingface/transformers/pull/45012/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45012",
     "labels": [],
     "merged": true,
-    "number": 44954,
+    "number": 45012,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] pipeline cleanup",
-    "updated_at": "2026-04-08T17:03:36Z"
+    "title": "refactor: added cache in check_repo",
+    "updated_at": "2026-03-30T06:44:23Z"
   },
   {
-    "additions": 861,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Decouples `kwargs` manipulation from hub's strict decorator, and ensures that all subclasses of a `PreTrainedConfig` accept any kwargs which is what we supported prev. Not all remote code has `@strict` or has an `__\u2026",
-    "changed_files": 536,
+    "additions": 64,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@ydshieh Hi, can you help review? Thx! Current error (before this PR): ```bash _ PI0ForConditionalGenerationModelTest.test_flash_attn_2_inference_equivalence _ self = <tests.models.pi0.test_modeling_pi0.PI0ForConditionalGenerationModelTest\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44953",
-    "created_at": "2026-03-23T17:13:39Z",
-    "deletions": 824,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45011",
+    "created_at": "2026-03-26T06:27:09Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44953/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44953",
+    "files_url": "https://github.com/huggingface/transformers/pull/45011/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45011",
     "labels": [],
     "merged": true,
-    "number": 44953,
-    "review_comments_count": 0,
+    "number": 45011,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "Config kwargs",
-    "updated_at": "2026-03-24T14:14:46Z"
+    "title": "skip 2 invalid test cases for pi0 model",
+    "updated_at": "2026-04-10T10:41:16Z"
   },
   {
-    "additions": 10,
-    "author": "Jess-Co-Del",
+    "additions": 75,
+    "author": "dagecko",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes the non existence of output dictionary change, when parameter output_hidden_states=True is passed to models like CLIP or SigLip. This is especially pertinent for the vision model config. According to #42759 no\u2026",
-    "changed_files": 2,
+    "body_excerpt": "## Fix: CI/CD Security Vulnerabilities in GitHub Actions Hi! [Runner Guard](https://github.com/Vigilant-LLC/runner-guard), an open-source CI/CD security scanner by [Vigilant Cyber Security](https://www.vigilantdefense.com), identified secu\u2026",
+    "changed_files": 20,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44952",
-    "created_at": "2026-03-23T17:02:50Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45010",
+    "created_at": "2026-03-26T06:19:29Z",
+    "deletions": 71,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44952/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44952",
+    "files_url": "https://github.com/huggingface/transformers/pull/45010/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45010",
     "labels": [],
     "merged": false,
-    "number": 44952,
+    "number": 45010,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix: Add correct return behaviour when output_hidden_states=True for CLIP and SIGLIP vision models",
-    "updated_at": "2026-03-24T11:19:35Z"
+    "state": "closed",
+    "title": "fix: pin 69 unpinned action(s),extract 2 unsafe expression(s) to env vars",
+    "updated_at": "2026-03-27T22:20:57Z"
   },
   {
-    "additions": 113,
-    "author": "hemantmm",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? This pull request adds routing replay functionality for mixture-of-experts (MoE) model types by giving users the option to override router probabilities while processing a forward pass through their models. <!-- Con\u2026",
-    "changed_files": 4,
+    "additions": 330,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This patch improves the docstring checker implementation (redundant AST walks) and adds cache. For the AST calls, 2.3x speedup check_docstrings.py --check_all on my M1: - before : 29.3s - after: 12.6s",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44951",
-    "created_at": "2026-03-23T16:29:46Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45009",
+    "created_at": "2026-03-26T05:31:41Z",
+    "deletions": 47,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44951/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44951",
+    "files_url": "https://github.com/huggingface/transformers/pull/45009/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45009",
     "labels": [],
-    "merged": false,
-    "number": 44951,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 45009,
+    "review_comments_count": 12,
     "state": "closed",
-    "title": "feat: Add router_logits override to enable Routing Replay for MoE models",
-    "updated_at": "2026-04-14T21:05:38Z"
+    "title": "refactor: speed up docstring checker",
+    "updated_at": "2026-03-27T07:21:11Z"
   },
   {
-    "additions": 1346,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. This PR finally makes mamba layer caches first class citizen, and adds native support for them. It supports the following layers combinations: - all mamba layers - alternating attention layer/mamba\u2026",
-    "changed_files": 64,
+    "additions": 380,
+    "author": "Krishnachaitanyakc",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Fix type annotation bugs across config classes that cause `@strict` validation errors from `huggingface_hub`. ### Bool fields mistyped as `int` (22 fixes) Fields with boolean defaults (`True`/`False`) were annotated as `int` ins\u2026",
+    "changed_files": 198,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44950",
-    "created_at": "2026-03-23T16:25:13Z",
-    "deletions": 4113,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45007",
+    "created_at": "2026-03-25T23:12:53Z",
+    "deletions": 380,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44950/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44950",
+    "files_url": "https://github.com/huggingface/transformers/pull/45007/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45007",
     "labels": [],
     "merged": true,
-    "number": 44950,
-    "review_comments_count": 48,
+    "number": 45007,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "\ud83d\udea8 [Cache] Native mamba & hybrid cache",
-    "updated_at": "2026-03-31T13:09:44Z"
+    "title": "fix: correct type annotations across config classes for @strict validation",
+    "updated_at": "2026-04-01T17:50:09Z"
   },
   {
-    "additions": 101,
-    "author": "Charly21r",
+    "additions": 3,
+    "author": "Krishnachaitanyakc",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes #44869 Adds a bounds check in `_split_tokens_on_unicode()` in `tokenization_whisper.py` to handle trailing Unicode replacement characters (U+FFFD) at the end of decoded token streams without crashing with `IndexError`. ##\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45006",
+    "created_at": "2026-03-25T23:03:00Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/45006/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45006",
+    "labels": [],
+    "merged": false,
+    "number": 45006,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "fix: prevent IndexError in Whisper timestamp decode on trailing replacement char",
+    "updated_at": "2026-04-06T01:03:09Z"
+  },
+  {
+    "additions": 10,
+    "author": "harshaljanjani",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #44936 This PR fixes an issue with `NotebookProgressCallback` in the `Trainer` where calling evaluate() before or after training would crash due to the training tracker being `None`. The callback now properly\u2026",
-    "changed_files": 5,
+    "body_excerpt": "### What does this PR do? The following failing tests were identified and fixed in this PR (grouped them together since they share related root causes OR the code changes were extremely minimal and didn't warrant separate PRs): \u2192 **Phi-3**\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44949",
-    "created_at": "2026-03-23T16:07:50Z",
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45004",
+    "created_at": "2026-03-25T19:58:57Z",
     "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44949/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44949",
+    "files_url": "https://github.com/huggingface/transformers/pull/45004/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45004",
     "labels": [],
     "merged": true,
-    "number": 44949,
-    "review_comments_count": 13,
+    "number": 45004,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Fix: NotebookProgressCallback crash when evaluating with the Trainer",
-    "updated_at": "2026-04-13T13:47:51Z"
+    "title": "fix(testing): Fix Parakeet, Evolla, Pi0, and Phi-3 test failures on main CI",
+    "updated_at": "2026-03-27T15:06:52Z"
   },
   {
     "additions": 1,
-    "author": "heycorgi",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "author": "hmellor",
+    "author_association": "MEMBER",
+    "body_excerpt": "`None` is a valid value that can be used to disable chunked attention in `DynamicCache` and Flex Attention. hf.co/morgendave/EAGLE-Llama-4-Scout-17B-16E-Instruct is an example of a checkpoint which does this.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44948",
-    "created_at": "2026-03-23T15:33:56Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45002",
+    "created_at": "2026-03-25T17:40:14Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44948/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44948",
+    "files_url": "https://github.com/huggingface/transformers/pull/45002/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45002",
     "labels": [],
-    "merged": false,
-    "number": 44948,
+    "merged": true,
+    "number": 45002,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Create aa.py",
-    "updated_at": "2026-03-23T15:34:35Z"
+    "title": "Fix type hint for `attention_chunk_size` in `Llama4TextConfig`",
+    "updated_at": "2026-03-25T20:42:11Z"
   },
   {
-    "additions": 123,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The doc was generated by Claude. I deleted unnecessary repetitions and fixed a few moments to be more precise. We don't really need to merge it now so if you think the text is too LLM, feel free to take this as an i\u2026",
-    "changed_files": 4,
+    "additions": 20,
+    "author": "Sai-Suraj-27",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? For [torch>=2.10.0](https://docs.pytorch.org/docs/2.10/generated/torch.nn.functional.grouped_mm.html#torch-nn-functional-grouped-mm), the minimum CUDA compute capability requirement for `torch.nn.functional.grouped_\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44947",
-    "created_at": "2026-03-23T13:23:04Z",
-    "deletions": 2,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45001",
+    "created_at": "2026-03-25T17:00:28Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44947/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44947",
+    "files_url": "https://github.com/huggingface/transformers/pull/45001/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45001",
     "labels": [],
     "merged": true,
-    "number": 44947,
-    "review_comments_count": 16,
+    "number": 45001,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Add doc page for capturing outputs",
-    "updated_at": "2026-03-26T13:08:46Z"
+    "title": "Add cuda compatibility check for using `grouped_mm`",
+    "updated_at": "2026-04-10T07:54:45Z"
   },
   {
-    "additions": 14,
-    "author": "BSchilperoort",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 13,
+    "additions": 22,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per title, after https://github.com/huggingface/transformers/pull/44976 users will be seeing a `missing_weights - lm_head not found` error even though the model doesn't use an lm head On the way also deleted unne\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44946",
-    "created_at": "2026-03-23T12:18:34Z",
-    "deletions": 14,
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/45000",
+    "created_at": "2026-03-25T16:28:55Z",
+    "deletions": 109,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44946/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44946",
+    "files_url": "https://github.com/huggingface/transformers/pull/45000/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/45000",
     "labels": [],
     "merged": true,
-    "number": 44946,
+    "number": 45000,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Correct docstrings for `from_pretrained` (url input deprecated)",
-    "updated_at": "2026-03-23T13:05:16Z"
+    "title": "Embedding VLMs don't need a head",
+    "updated_at": "2026-03-27T09:57:18Z"
   },
   {
-    "additions": 71,
-    "author": "zucchini-nlp",
+    "additions": 2450,
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? @hmellor",
-    "changed_files": 5,
+    "body_excerpt": "## Summary - Auto-generated modular integration for `sarvam` - `modular_sarvam.py` written by Claude Opus 4.6 guided by `modular_model_detector.py` - `modeling_sarvam.py` regenerated from modular via `modular_model_converter.py` ## Test pl\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44943",
-    "created_at": "2026-03-23T10:58:40Z",
-    "deletions": 9,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44943/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44943",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44999",
+    "created_at": "2026-03-25T16:21:37Z",
+    "deletions": 0,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44999/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44999",
     "labels": [],
-    "merged": true,
-    "number": 44943,
-    "review_comments_count": 1,
+    "merged": false,
+    "number": 44999,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Clearer type hints and fix rope validation in configs",
-    "updated_at": "2026-03-23T13:32:11Z"
+    "title": "Add sarvam model",
+    "updated_at": "2026-03-30T23:21:49Z"
   },
   {
-    "additions": 220,
-    "author": "hmellor",
+    "additions": 1179,
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": null,
-    "changed_files": 3,
+    "body_excerpt": "## Summary - Auto-generated modular integration for `sarvam` - `modular_sarvam.py` written by Claude Opus 4.6 guided by `modular_model_detector.py` - `modeling_sarvam.py` regenerated from modular via `modular_model_converter.py` ## Test pl\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44942",
-    "created_at": "2026-03-23T10:46:23Z",
-    "deletions": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44997",
+    "created_at": "2026-03-25T14:23:13Z",
+    "deletions": 0,
     "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44942/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44942",
+    "files_url": "https://github.com/huggingface/transformers/pull/44997/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44997",
     "labels": [],
     "merged": false,
-    "number": 44942,
+    "number": 44997,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add inference time layer fusion optimisations via `PreTrainedModel.from_pretrained(fuse_layers=True)`",
-    "updated_at": "2026-03-25T16:14:19Z"
+    "state": "closed",
+    "title": "Add sarvam model",
+    "updated_at": "2026-03-25T14:35:45Z"
   },
   {
-    "additions": 4,
-    "author": "ydshieh",
+    "additions": 286,
+    "author": "3outeille",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix the failing job after #43514 (the fix is effefctive, see [here](https://github.com/huggingface/transformers/actions/runs/23433395911/job/68165255513?pr=44941)) [Update Transformers metadata](https://github.com/h\u2026",
-    "changed_files": 1,
+    "body_excerpt": "- Introduce `DistributedConfig` - `DistributedConfig(tp_size=2, fsdp_size=2) # plans default to \"auto\"` replaces passing separate `tp_plan, tp_size, fsdp_plan kwargs`. Sizes auto-fill (specify one, the other defaults to 1). Plans default t\u2026",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44941",
-    "created_at": "2026-03-23T10:42:09Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44996",
+    "created_at": "2026-03-25T14:20:25Z",
+    "deletions": 283,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44941/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44941",
+    "files_url": "https://github.com/huggingface/transformers/pull/44996/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44996",
     "labels": [],
     "merged": true,
-    "number": 44941,
-    "review_comments_count": 1,
+    "number": 44996,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix failing job `Update Transformers metadata` after #43514",
-    "updated_at": "2026-03-23T13:41:39Z"
+    "title": " from_pretrained distributed refactor (FSDP2 + TP)",
+    "updated_at": "2026-03-26T15:32:27Z"
   },
   {
-    "additions": 138,
-    "author": "Qubitium",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Model loading of same model path but 2 different threads (2 different instances) have meta device tensor issues: unloaded meta/empty embedding/lm-head when it should not be empty post model load. Cause: `tie_weight(\u2026",
-    "changed_files": 3,
+    "additions": 3639,
+    "author": "itazap",
+    "author_association": "MEMBER",
+    "body_excerpt": null,
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44940",
-    "created_at": "2026-03-23T09:55:57Z",
-    "deletions": 10,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44994",
+    "created_at": "2026-03-25T14:02:50Z",
+    "deletions": 242,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44940/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44940",
+    "files_url": "https://github.com/huggingface/transformers/pull/44994/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44994",
     "labels": [],
     "merged": false,
-    "number": 44940,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "Fix tie_weights skipping logic is not tied to model thread scope",
-    "updated_at": "2026-04-07T02:01:50Z"
+    "number": 44994,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Add sarvam model",
+    "updated_at": "2026-03-25T14:04:38Z"
   },
   {
-    "additions": 2038,
+    "additions": 583,
     "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Refactored and cleaned up model linter - separated package - one rule per module - refactored legacy checks into their own rules - simplified pattern, duplication removal",
-    "changed_files": 25,
+    "body_excerpt": "# What does this PR do? `make check-repo` can be quite slow, this patch adds file-level cache to speed up checks. We get up to a 27x speedup - cold cache : 46s - warm cache : 1.6s",
+    "changed_files": 20,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44939",
-    "created_at": "2026-03-23T08:45:36Z",
-    "deletions": 1446,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44992",
+    "created_at": "2026-03-25T11:40:46Z",
+    "deletions": 48,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44939/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44939",
+    "files_url": "https://github.com/huggingface/transformers/pull/44992/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44992",
     "labels": [],
     "merged": true,
-    "number": 44939,
-    "review_comments_count": 5,
+    "number": 44992,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "refactor: mlinter as its own package",
-    "updated_at": "2026-03-24T07:56:15Z"
+    "title": "refactoring: speedup static checks with disk cache",
+    "updated_at": "2026-03-31T12:34:20Z"
   },
   {
-    "additions": 2,
-    "author": "VanshikaSohal",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes two small but impactful bugs in the BART documentation: 1. **Variable shadowing bug**: In the Pipeline example, the variable was named `pipeline` which shadows the imported `pipeline` function. Renamed to `fi\u2026",
+    "additions": 8,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? - BC for check model inputs",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44935",
-    "created_at": "2026-03-22T18:45:01Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44990",
+    "created_at": "2026-03-25T10:26:20Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44935/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44935",
+    "files_url": "https://github.com/huggingface/transformers/pull/44990/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44990",
     "labels": [],
     "merged": true,
-    "number": 44935,
+    "number": 44990,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix variable shadowing in pipeline example and typo in BART docs (BERT \u2192 BART)",
-    "updated_at": "2026-03-23T14:28:04Z"
+    "title": "More small vllm fixes",
+    "updated_at": "2026-03-25T13:05:44Z"
   },
   {
-    "additions": 9,
-    "author": "Sai-Suraj-27",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes this failing [T5ModelIntegrationTest](https://github.com/huggingface/transformers/actions/runs/23230643883/job/67524758706#step:14:1449) & this [Qwen2IntegrationTest](https://github.com/huggingface/transformer\u2026",
-    "changed_files": 2,
+    "additions": 542,
+    "author": "3outeille",
+    "author_association": "MEMBER",
+    "body_excerpt": "# Distributed Training API ## Goal ```python # torchrun --nproc_per_node=4 train_fsdp_tp.py import os import torch from torch.utils.data import DataLoader from datasets import load_dataset from transformers import AutoModelForCausalLM, Aut\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44934",
-    "created_at": "2026-03-22T18:03:34Z",
-    "deletions": 7,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44934/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44934",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44989",
+    "created_at": "2026-03-25T09:10:02Z",
+    "deletions": 0,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44989/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44989",
     "labels": [],
-    "merged": true,
-    "number": 44934,
+    "merged": false,
+    "number": 44989,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix failing `T5ModelIntegrationTest`",
-    "updated_at": "2026-03-24T14:50:10Z"
+    "state": "open",
+    "title": "\ud83d\udea8 Distributed training API",
+    "updated_at": "2026-04-13T16:44:35Z"
   },
   {
-    "additions": 1,
-    "author": "r266-tech",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #44908 The `get_inverse_sqrt_schedule` function accepts `timescale` and `last_epoch` parameters, but `get_scheduler` was not forwarding `scheduler_specific_kwargs` to it. This caused user-provided kwargs like\u2026",
-    "changed_files": 1,
+    "additions": 730,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Adds Rule 15 ``` if _tied_weights_keys is present and non-empty in modeling -> Config MUST contain the tie_word_embeddings field ```",
+    "changed_files": 16,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44932",
-    "created_at": "2026-03-22T17:30:56Z",
-    "deletions": 1,
+    "comments_count": 17,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44988",
+    "created_at": "2026-03-25T07:08:20Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44932/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44932",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44988/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44988",
+    "labels": [],
     "merged": false,
-    "number": 44932,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix: Pass scheduler_specific_kwargs to inverse_sqrt scheduler",
-    "updated_at": "2026-03-23T12:44:16Z"
+    "number": 44988,
+    "review_comments_count": 3,
+    "state": "open",
+    "title": "typing: rule 15 - checks for tie_word_embeddings presence",
+    "updated_at": "2026-04-14T07:07:02Z"
   },
   {
-    "additions": 1,
-    "author": "r266-tech",
+    "additions": 0,
+    "author": "Krishnachaitanyakc",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes a v5 regression where `CamembertForMaskedLM` (and all CamemBERT masked-LM tasks) produces near-zero, near-uniform logits, making the model completely non-functional. ### Root cause In v5, `modeling_utils.get_\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary Fixes #44855 On Python 3.13, placing a `# Copied from` comment between `@torch.jit.script` and the function definition causes an `IndentationError`. This happens because `torch.jit.script` calls `inspect.getsource()` followed by\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44931",
-    "created_at": "2026-03-22T17:28:57Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44986",
+    "created_at": "2026-03-25T03:18:31Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44931/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44931",
+    "files_url": "https://github.com/huggingface/transformers/pull/44986/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44986",
     "labels": [],
     "merged": true,
-    "number": 44931,
+    "number": 44986,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(camembert): add tie_word_embeddings=True to CamembertConfig",
-    "updated_at": "2026-03-25T07:09:37Z"
+    "title": "fix: remove Copied from comments between @torch.jit.script and def for Python 3.13 compat",
+    "updated_at": "2026-03-25T13:39:54Z"
   },
   {
-    "additions": 103,
-    "author": "javierdejesusda",
+    "additions": 2,
+    "author": "Krishnachaitanyakc",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary - Fixes #44912 \u2014 MXFP4 quantization error messages combine `is_triton_available()` and `is_kernels_available()` into a single `kernels_available` boolean, making it impossible to identify which dependency is missing - Split the\u2026",
+    "body_excerpt": "## Summary Fixes #44913 When creating a `GPTNeoXConfig` (or `GPTNeoXJapaneseConfig`) with a non-default `rotary_pct`, the value is lost after a `save_pretrained` / `from_pretrained` round-trip. This happens because `convert_rope_params_to_\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44930",
-    "created_at": "2026-03-22T17:27:20Z",
-    "deletions": 13,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44985",
+    "created_at": "2026-03-25T02:15:04Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44930/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44930",
+    "files_url": "https://github.com/huggingface/transformers/pull/44985/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44985",
     "labels": [],
     "merged": true,
-    "number": 44930,
-    "review_comments_count": 3,
+    "number": 44985,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "fix: split MXFP4 dependency checks for specific error messages",
-    "updated_at": "2026-03-24T15:33:14Z"
+    "title": "fix: preserve rotary_pct across save/load cycle in GPTNeoX configs",
+    "updated_at": "2026-03-27T09:19:46Z"
   },
   {
-    "additions": 26,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "## Problem In `TokenizersBackend.convert_to_native_format()`, when a tokenizer has a custom `__init__` (the `elif` branch), `tokenizer.json` was parsed **twice**: 1. `TokenizerFast.from_file(fast_tokenizer_file)` \u2014 full Rust parse includin\u2026",
+    "additions": 2,
+    "author": "Butanium",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? `maybe_autocast` calls `torch.is_autocast_enabled(device_type)` which raises a `RuntimeError` when `device_type` is `\"meta\"`: ``` RuntimeError: unknown device type for autocast in get_autocast_dispatch_key_from_dev\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44927",
-    "created_at": "2026-03-22T15:33:23Z",
-    "deletions": 4,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44984",
+    "created_at": "2026-03-25T01:39:23Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44927/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44927",
+    "files_url": "https://github.com/huggingface/transformers/pull/44984/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44984",
     "labels": [],
     "merged": true,
-    "number": 44927,
-    "review_comments_count": 6,
+    "number": 44984,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: improve processor loading performance by avoiding redundant tokenizer parsing",
-    "updated_at": "2026-03-23T11:03:52Z"
+    "title": "Fix `maybe_autocast` crashing on meta device tensors",
+    "updated_at": "2026-03-25T17:45:03Z"
   },
   {
-    "additions": 25,
-    "author": "yonigozlan",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Solve import errors when trying to import `from transformers.models.llama4.image_processing_llama4_fast import Llama4ImageProcessorFast` for example",
-    "changed_files": 2,
+    "additions": 26,
+    "author": "Hyungkeun-Park-Nota",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes `save_pretrained()` for models loaded with `dequantize=True`. `save_pretrained` calls `reverse_op` on all weight conversion operations from loading. Dequantize ops (`Mxfp4Dequantize`, `Fp8Dequantize`, `MetalD\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44926",
-    "created_at": "2026-03-22T14:46:17Z",
-    "deletions": 0,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44983",
+    "created_at": "2026-03-25T01:19:59Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44926/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44926",
+    "files_url": "https://github.com/huggingface/transformers/pull/44983/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44983",
     "labels": [],
     "merged": true,
-    "number": 44926,
-    "review_comments_count": 1,
+    "number": 44983,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "Fix backward compatibility for full path imports of Fast Image Processors",
-    "updated_at": "2026-03-23T14:16:49Z"
+    "title": "fix: add identity reverse_op to dequantize ops for save_pretrained",
+    "updated_at": "2026-03-27T17:21:18Z"
   },
   {
-    "additions": 96,
-    "author": "kashif",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds a first-class MoE routing capture/replay API for Qwen2Moe and introduces shared MoE routing helpers for reuse by other MoE model families. It adds: - a structured `MoERouting` payload in modeling output\u2026",
-    "changed_files": 1,
+    "additions": 108,
+    "author": "AkshajKashyap",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #43039 ## What does this PR do? When `prediction_loss_only=True` during evaluation and `use_liger_kernel=True`, `Trainer.prediction_step` now passes `skip_logits=True` to the model forward if the forward signature supports it and lab\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44925",
-    "created_at": "2026-03-22T14:04:40Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44981",
+    "created_at": "2026-03-25T00:38:02Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44925/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44925",
+    "files_url": "https://github.com/huggingface/transformers/pull/44981/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44981",
     "labels": [],
-    "merged": true,
-    "number": 44925,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[Doc]  MoE routing capture and replay recipe ",
-    "updated_at": "2026-04-14T08:09:16Z"
+    "merged": false,
+    "number": 44981,
+    "review_comments_count": 6,
+    "state": "open",
+    "title": "Trainer: set skip_logits for loss-only eval when liger enabled",
+    "updated_at": "2026-04-09T15:31:50Z"
   },
   {
-    "additions": 9,
-    "author": "Qubitium",
+    "additions": 6,
+    "author": "kallewoof",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fix two `nogil` threading bugs (reproduced on 3.14) : 1. Continus Batching crashes with torch graph errors with 2 threads on 2 separate model instances (same model path, but two distinct instances). Cause is missing\u2026",
-    "changed_files": 1,
+    "body_excerpt": "Pre-patch unnecessarily breaks merging a LoRA adapter with a model using CUDA_VISIBLE_DEVICES= e.g. when VRAM is insufficient. It also breaks non-cuda machine operations (such as merging). # What does this PR do? This PR un-breaks `CUDA_VI\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 13,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44924",
-    "created_at": "2026-03-22T11:46:49Z",
-    "deletions": 1,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44980",
+    "created_at": "2026-03-24T23:50:07Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44924/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44924",
+    "files_url": "https://github.com/huggingface/transformers/pull/44980/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44980",
     "labels": [],
-    "merged": true,
-    "number": 44924,
-    "review_comments_count": 1,
+    "merged": false,
+    "number": 44980,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Continuous batching thread safety",
-    "updated_at": "2026-03-24T05:42:56Z"
+    "title": "bug-fix: do not assume torch.cuda is available when setting up norm values, even if flash linear attention is available",
+    "updated_at": "2026-03-27T13:25:18Z"
   },
   {
-    "additions": 3,
-    "author": "prakhar-agarwal",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Addresses issue #44843. Verified with isolated repro logic. Changes made: Updated the logic to properly identify local and offline scenarios upfront. Now, is_local is correctly set to True if: 1. is_offline_mode() is active. 2. The local_f\u2026",
-    "changed_files": 1,
+    "additions": 492,
+    "author": "michaelbenayoun",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Introduces `src/transformers/module_fusion.py`, a utility for fusing adjacent submodules in a model into a single FusedModule that executes them as a chain in one forward pass. The key components are: - `RegistryCol\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44923",
-    "created_at": "2026-03-22T05:20:22Z",
-    "deletions": 1,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44923/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44923",
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44979",
+    "created_at": "2026-03-24T22:33:31Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44979/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44979",
     "labels": [],
     "merged": false,
-    "number": 44923,
+    "number": 44979,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fix: avoid unconditional model_info call in _patch_mistral_regex",
-    "updated_at": "2026-03-22T05:24:11Z"
+    "title": "Module Fusion API",
+    "updated_at": "2026-04-13T19:46:19Z"
   },
   {
-    "additions": 10,
-    "author": "s-zx",
+    "additions": 4,
+    "author": "cjkindel",
     "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes #44849. When `output_hidden_states=True` (or `output_attentions=True`) is passed to `model.generate()`, the `@capture_outputs` decorator reads the flag value but leaves it in `**kwargs`. These flags then prop\u2026",
+    "body_excerpt": "# What does this PR do? `_can_set_attn_implementation` and `_can_set_experts_implementation` both do a direct subscript lookup into `sys.modules`: ```python class_module = sys.modules[cls.__module__] ``` If the module is not registered und\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44922",
-    "created_at": "2026-03-22T01:21:22Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44978",
+    "created_at": "2026-03-24T21:01:11Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44922/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44922",
+    "files_url": "https://github.com/huggingface/transformers/pull/44978/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44978",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44922,
+    "number": 44978,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: pop output_* flags from kwargs in capture_outputs to prevent submodule leakage",
-    "updated_at": "2026-03-23T12:38:56Z"
+    "title": "fix: handle absent sys.modules entry in modeling_utils",
+    "updated_at": "2026-03-26T12:25:31Z"
   },
   {
-    "additions": 4,
-    "author": "s-zx",
+    "additions": 2,
+    "author": "hmellor",
+    "author_association": "MEMBER",
+    "body_excerpt": "- Adds a type hint to `ModernVBertForMaskedLM.__init__` - Removes `tie_word_embeddings` from `Qwen2VLTextConfig` (and therefore also `Qwen2_5_VLTextConfig`) because it's not valid for these models - Remove hack from `ColQwen2Config` (and t\u2026",
+    "changed_files": 6,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44976",
+    "created_at": "2026-03-24T19:26:33Z",
+    "deletions": 10,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44976/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44976",
+    "labels": [],
+    "merged": true,
+    "number": 44976,
+    "review_comments_count": 3,
+    "state": "closed",
+    "title": "Fix tie_word_embedding issues with `Qwen2VL`",
+    "updated_at": "2026-03-24T20:55:15Z"
+  },
+  {
+    "additions": 6971,
+    "author": "philippguevorguian",
     "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes #44918. `compute_3d_position_ids` in the Qwen2.5-VL / Qwen3-VL / Qwen3.5 model families destructures `inputs_embeds.shape` into exactly three variables: ```python batch_size, seq_length, _ = inputs_embeds.sha\u2026",
-    "changed_files": 4,
+    "body_excerpt": null,
+    "changed_files": 20,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44921",
-    "created_at": "2026-03-22T00:39:01Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44975",
+    "created_at": "2026-03-24T17:12:31Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44921/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44921",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44975/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44975",
+    "labels": [],
     "merged": false,
-    "number": 44921,
+    "number": 44975,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: use shape index access in compute_3d_position_ids for Qwen VL models",
-    "updated_at": "2026-03-23T10:00:51Z"
+    "title": "fix: rebase main; clean config reads, ImageProcessor backend, misc cleanup",
+    "updated_at": "2026-03-24T17:13:42Z"
   },
   {
-    "additions": 15,
-    "author": "s-zx",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes `num_labels` not being propagated from `Qwen3_5Config` to its `text_config` when loading via `AutoConfig.from_pretrained(model, num_labels=N)`. **Root cause:** `Qwen3_5Config.__post_init__` initializes `text_\u2026",
-    "changed_files": 2,
+    "additions": 428,
+    "author": "3outeille",
+    "author_association": "MEMBER",
+    "body_excerpt": "TODO: - Saving seems to take a bit of time tho. Need investigation - Need to check if it works in 1D (FSDP or TP)and 2D (FSDP + TP). Running the script from https://github.com/huggingface/transformers/pull/44996 ``` (env_pr-44974-fsdp-core\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44920",
-    "created_at": "2026-03-22T00:01:59Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44974",
+    "created_at": "2026-03-24T16:13:25Z",
+    "deletions": 82,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44920/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44920",
+    "files_url": "https://github.com/huggingface/transformers/pull/44974/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44974",
     "labels": [],
     "merged": false,
-    "number": 44920,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: propagate num_labels/id2label to text_config in Qwen3_5Config",
-    "updated_at": "2026-03-23T12:06:04Z"
+    "number": 44974,
+    "review_comments_count": 3,
+    "state": "open",
+    "title": "Refactor core_model_loading to support FSDP shard-on-read loading",
+    "updated_at": "2026-04-14T14:38:45Z"
   },
   {
-    "additions": 18,
-    "author": "s-zx",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes a crash in `Qwen2_5_VLProcessor.__call__` when processing batched inputs without padding (`padding=False`). **Root cause:** When the tokenizer returns sequences of different lengths (ragged list), `np.array(t\u2026",
-    "changed_files": 1,
+    "additions": 22,
+    "author": "andylizf",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Adds `.item()` to `max_seqlen = (cu_seqlens[1:] - cu_seqlens[:-1]).max()` in all vision attention modules that pass this value to `flash_attn_varlen_func`. ### Context On **released versions** (e.g. 4.52.4), using\u2026",
+    "changed_files": 19,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44919",
-    "created_at": "2026-03-21T23:57:37Z",
-    "deletions": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44973",
+    "created_at": "2026-03-24T15:42:32Z",
+    "deletions": 22,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44919/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44919",
+    "files_url": "https://github.com/huggingface/transformers/pull/44973/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44973",
     "labels": [],
     "merged": false,
-    "number": 44919,
+    "number": 44973,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: handle ragged batch inputs in Qwen2_5_VLProcessor mm_token_type_ids computation",
-    "updated_at": "2026-03-23T10:38:30Z"
+    "state": "open",
+    "title": "Fix max_seqlen type in vision attention for torch.compile + FA2",
+    "updated_at": "2026-03-25T14:12:50Z"
   },
   {
-    "additions": 5,
-    "author": "s-zx",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary `GPTNeoXConfig.convert_rope_params_to_dict` unconditionally overwrote `rope_parameters[\"partial_rotary_factor\"]` with the default `0.25` when `rotary_pct` was absent from kwargs. On every `from_pretrained` call, `rotary_pct` is\u2026",
-    "changed_files": 1,
+    "additions": 17,
+    "author": "Abdennacer-Badaoui",
+    "author_association": "MEMBER",
+    "body_excerpt": "As per title. Updating Gemma3/Gemma3n expectations.",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44917",
-    "created_at": "2026-03-21T23:34:32Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44972",
+    "created_at": "2026-03-24T15:11:50Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44917/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44917",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44917,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44972/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44972",
+    "labels": [],
+    "merged": true,
+    "number": 44972,
+    "review_comments_count": 10,
     "state": "closed",
-    "title": "fix(gpt-neox): preserve rotary_pct across save/load cycle",
-    "updated_at": "2026-03-23T12:37:48Z"
+    "title": "[AMD CI] Gemma3/Gemma3n Expectations",
+    "updated_at": "2026-03-24T16:30:03Z"
   },
   {
-    "additions": 8,
-    "author": "s-zx",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Importing `DebertaV2Model` (or anything that depends on it, e.g. `gliner`) raises `IndentationError` on Python 3.13 because `torch.jit.script` calls `inspect.getsource()`, dedents the snippet, and passes it to `ast.parse()`. Pyt\u2026",
-    "changed_files": 2,
+    "additions": 0,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Removed the tokenizer_class attr was never there to begin with, and kwargs are now supported. This was failing some test on vllm ci. Fixes https://buildkite.com/vllm/ci/builds/57601/steps/canvas?sid=019d1aec-aa5a-41\u2026",
+    "changed_files": 4,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44971",
+    "created_at": "2026-03-24T14:59:36Z",
+    "deletions": 11,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44971/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44971",
+    "labels": [],
+    "merged": true,
+    "number": 44971,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "[ `vllm x v5`] nit",
+    "updated_at": "2026-03-24T17:40:05Z"
+  },
+  {
+    "additions": 20,
+    "author": "IlyasMoutawwakil",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44916",
-    "created_at": "2026-03-21T23:34:07Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44970",
+    "created_at": "2026-03-24T13:49:21Z",
+    "deletions": 76,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44916/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44916",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44916,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44970/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44970",
+    "labels": [],
+    "merged": true,
+    "number": 44970,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "fix(deberta-v2): move \"Copied from\" comments above @torch.jit.script for Python 3.13 compat",
-    "updated_at": "2026-03-23T12:34:24Z"
+    "title": "Fix CPU 16 bytes alignment issue using equivalent fallback",
+    "updated_at": "2026-03-25T09:01:03Z"
   },
   {
-    "additions": 90,
-    "author": "maxsloef-goodfire",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? `clean_up_tokenization` applies English-specific string replacements (` .` \u2192 `.`, ` ?` \u2192 `?`, ` ,` \u2192 `,`, etc.) to decoded text. This was designed for BERT-era WordPiece tokenizers where decoding produced artifacts\u2026",
-    "changed_files": 4,
+    "additions": 4,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Extends the CI so we can use Make and read toml files",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44915",
-    "created_at": "2026-03-21T20:45:03Z",
-    "deletions": 6,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44968",
+    "created_at": "2026-03-24T11:43:24Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44915/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44915",
+    "files_url": "https://github.com/huggingface/transformers/pull/44968/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44968",
     "labels": [],
     "merged": false,
-    "number": 44915,
-    "review_comments_count": 1,
-    "state": "open",
-    "title": "fix: skip `clean_up_tokenization` for BPE tokenizers in `PreTrainedTokenizerFast`",
-    "updated_at": "2026-04-09T13:39:02Z"
+    "number": 44968,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Added Make to the docker and `tomli` to `.[quality]`",
+    "updated_at": "2026-03-24T15:06:29Z"
   },
   {
-    "additions": 1,
-    "author": "maxsloef-goodfire",
+    "additions": 87,
+    "author": "Qubitium",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? The `Llama3Converter` in `convert_llama_weights_to_hf.py` hardcodes `clean_up_tokenization_spaces=True` (line 468). This causes `tokenizer.decode()` to silently strip spaces before punctuation for all converted Lla\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Fix: FA kernel launches currently are not thread-safe (nogil) in multi-gpu env. This simple patch fixes the issue. ```py # Set the correct CUDA context before launching the FlashAttention kernel. with torch.cuda.dev\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44914",
-    "created_at": "2026-03-21T20:25:51Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44914/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44914",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44967",
+    "created_at": "2026-03-24T11:33:45Z",
+    "deletions": 84,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44967/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44967",
     "labels": [],
-    "merged": true,
-    "number": 44914,
+    "merged": false,
+    "number": 44967,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: set `clean_up_tokenization_spaces=False` in Llama 3 tokenizer conversion",
-    "updated_at": "2026-03-23T08:38:18Z"
+    "title": "[WIP] Fix FA kernel launch needs correct cuda device ctx in multi-gpu env",
+    "updated_at": "2026-04-08T15:10:47Z"
   },
   {
     "additions": 8,
-    "author": "ouroborosscr",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "Qwen3.5 uses 3D position_ids [3, batch, seq_len] for multi-dimensional rotary embedding. _is_packed_sequence() misinterprets this as a packed sequence, causing cu_seqlens to be constructed with 3x the actual token count. Flash attention th\u2026",
+    "author": "pramilajangid",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #44964 ## Summary This PR restores backward compatibility for `CommonKwargs` in `transformers.processing_utils`, which is still referenced by some remote processor implementations. ## Problem After the typed-dict cleanup (commit `533\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44911",
-    "created_at": "2026-03-21T15:42:57Z",
-    "deletions": 4,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44966",
+    "created_at": "2026-03-24T11:06:57Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44911/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44911",
+    "files_url": "https://github.com/huggingface/transformers/pull/44966/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44966",
     "labels": [],
     "merged": false,
-    "number": 44911,
+    "number": 44966,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix flash attention crash with 3D position_ids (Qwen3.5)",
-    "updated_at": "2026-03-24T14:35:57Z"
+    "title": "Fix backward compatibility for CommonKwargs in processing_utils (brea\u2026",
+    "updated_at": "2026-03-24T12:48:44Z"
   },
   {
-    "additions": 1,
-    "author": "anshuS1310",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "The `get_scheduler` function was identifying the `inverse_sqrt` scheduler type but failing to pass `**scheduler_specific_kwargs` to the underlying `get_inverse_sqrt_schedule` function. This caused user-defined parameters like `timescale` t\u2026",
-    "changed_files": 1,
+    "additions": 37,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44909",
-    "created_at": "2026-03-21T09:59:07Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44965",
+    "created_at": "2026-03-24T10:59:31Z",
+    "deletions": 32,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44909/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44909",
+    "files_url": "https://github.com/huggingface/transformers/pull/44965/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44965",
     "labels": [],
-    "merged": true,
-    "number": 44909,
+    "merged": false,
+    "number": 44965,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix: Update optimization.py",
-    "updated_at": "2026-03-24T13:06:15Z"
+    "state": "open",
+    "title": "try",
+    "updated_at": "2026-03-24T11:19:27Z"
   },
   {
-    "additions": 200,
-    "author": "syncdoth",
+    "additions": 3,
+    "author": "josh-kean",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #44906 ## Summary - Remove `.expand_as(inputs_embeds)` from placeholder mask creation in `get_placeholder_mask` and equivalent inline patterns across all VLM models. `masked_scatter` natively broadcasts `(B, S, 1)` \u2192 `(B, S, H)`, mak\u2026",
-    "changed_files": 71,
+    "body_excerpt": "# What does this PR do? Fixes an import in src/transformers/video_processing_utils.py that was causing the main build to fail Fixes # 44933 ## Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44907",
-    "created_at": "2026-03-21T06:07:35Z",
-    "deletions": 222,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44958",
+    "created_at": "2026-03-23T20:07:09Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44907/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44907",
+    "files_url": "https://github.com/huggingface/transformers/pull/44958/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44958",
     "labels": [],
     "merged": false,
-    "number": 44907,
-    "review_comments_count": 0,
+    "number": 44958,
+    "review_comments_count": 1,
     "state": "open",
-    "title": "Remove unnecessary expand_as in get_placeholder_mask across VLMs",
-    "updated_at": "2026-03-23T12:20:03Z"
+    "title": "fixed import error with PILImageResampling",
+    "updated_at": "2026-03-24T13:53:00Z"
   },
   {
-    "additions": 13,
-    "author": "NicoleRobin",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary - 13 i18n README files used `./awesome-transformers.md` which resolves relative to the `i18n/` directory and leads to a 404 - Replace with the absolute GitHub URL so links work from any location - `README_ko.md` was already corr\u2026",
-    "changed_files": 13,
+    "additions": 1272,
+    "author": "bigshanedogg",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Adds native Transformers support for **[HyperCLOVA X SEED Think 14B](https://huggingface.co/naver-hyperclovax/HyperCLOVAX-SEED-Think-14B)**, a 14.74B-parameter Korean reasoning LLM developed by NAVER Cloud. - relate\u2026",
+    "changed_files": 12,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44905",
-    "created_at": "2026-03-21T03:25:56Z",
-    "deletions": 13,
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44956",
+    "created_at": "2026-03-23T19:34:30Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44905/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44905",
+    "files_url": "https://github.com/huggingface/transformers/pull/44956/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44956",
     "labels": [],
-    "merged": true,
-    "number": 44905,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(i18n): replace broken relative links to awesome-transformers.md with absolute URLs",
-    "updated_at": "2026-03-23T12:47:56Z"
+    "merged": false,
+    "number": 44956,
+    "review_comments_count": 24,
+    "state": "open",
+    "title": "Add HyperCLOVAX SEED Think 14B",
+    "updated_at": "2026-04-12T06:17:53Z"
   },
   {
-    "additions": 101,
-    "author": "vivekvar-dl",
-    "author_association": "NONE",
-    "body_excerpt": "# Fix granite_speech config loading failure with int multiplier fields ## Fixes #44877 ### Problem Loading `granite_speech` configs fails with `StrictDataclassFieldValidationError` when multiplier fields (e.g., `embedding_multiplier`) are\u2026",
-    "changed_files": 3,
+    "additions": 0,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "removes outdated qa pipeline reference",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44904",
-    "created_at": "2026-03-21T03:12:37Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44954",
+    "created_at": "2026-03-23T17:20:37Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44904/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44904",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44904,
+    "files_url": "https://github.com/huggingface/transformers/pull/44954/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44954",
+    "labels": [],
+    "merged": true,
+    "number": 44954,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(granite_speech): convert int to float for multiplier fields in text_config",
-    "updated_at": "2026-03-23T10:37:38Z"
+    "title": "[docs] pipeline cleanup",
+    "updated_at": "2026-04-08T17:03:36Z"
   },
   {
-    "additions": 16,
-    "author": "yonigozlan",
+    "additions": 861,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Some remote code models are using `get_size_dict` directly, and now that size is converted to SizeDict in init, we need to support it as input in `get_size_dict`",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Decouples `kwargs` manipulation from hub's strict decorator, and ensures that all subclasses of a `PreTrainedConfig` accept any kwargs which is what we supported prev. Not all remote code has `@strict` or has an `__\u2026",
+    "changed_files": 536,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44903",
-    "created_at": "2026-03-21T01:25:53Z",
-    "deletions": 7,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44953",
+    "created_at": "2026-03-23T17:13:39Z",
+    "deletions": 824,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44903/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44903",
+    "files_url": "https://github.com/huggingface/transformers/pull/44953/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44953",
     "labels": [],
     "merged": true,
-    "number": 44903,
-    "review_comments_count": 5,
+    "number": 44953,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Support SizeDict import in get_size_dict",
-    "updated_at": "2026-03-23T10:28:52Z"
+    "title": "Config kwargs",
+    "updated_at": "2026-03-24T14:14:46Z"
   },
   {
-    "additions": 3,
-    "author": "guoyangzhen",
-    "author_association": "NONE",
-    "body_excerpt": "## Problem `_split_tokens_on_unicode()` crashes with `IndexError: string index out of range` when the decoded token stream ends with a dangling Unicode replacement character (\\uFFFD). The computed index `unicode_offset + decoded.index(repl\u2026",
-    "changed_files": 1,
+    "additions": 10,
+    "author": "Jess-Co-Del",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes the non existence of output dictionary change, when parameter output_hidden_states=True is passed to models like CLIP or SigLip. This is especially pertinent for the vision model config. According to #42759 no\u2026",
+    "changed_files": 2,
     "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44902",
-    "created_at": "2026-03-20T22:08:49Z",
-    "deletions": 1,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44952",
+    "created_at": "2026-03-23T17:02:50Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44902/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44902",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44952/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44952",
+    "labels": [],
     "merged": false,
-    "number": 44902,
+    "number": 44952,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: Whisper word timestamp OOB access on trailing replacement char",
-    "updated_at": "2026-03-23T11:59:14Z"
+    "state": "open",
+    "title": "Fix: Add correct return behaviour when output_hidden_states=True for CLIP and SIGLIP vision models",
+    "updated_at": "2026-03-24T11:19:35Z"
   },
   {
-    "additions": 19,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following failing Perceiver use case was identified and fixed in this PR: \u2192 c6d2848a23 ([\ud83d\udea8 Fix torch.jit.trace for interpolate_pos_encoding in all vision models](https://github.com/huggingface/transformers/pul\u2026",
-    "changed_files": 2,
+    "additions": 113,
+    "author": "hemantmm",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? This pull request adds routing replay functionality for mixture-of-experts (MoE) model types by giving users the option to override router probabilities while processing a forward pass through their models. <!-- Con\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44899",
-    "created_at": "2026-03-20T20:02:10Z",
-    "deletions": 1,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44951",
+    "created_at": "2026-03-23T16:29:46Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44899/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44899",
+    "files_url": "https://github.com/huggingface/transformers/pull/44951/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44951",
     "labels": [],
-    "merged": true,
-    "number": 44899,
-    "review_comments_count": 2,
+    "merged": false,
+    "number": 44951,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(models): Fix Perceiver interpolate_pos_encoding interpolating to the source size",
-    "updated_at": "2026-03-25T11:54:23Z"
+    "title": "feat: Add router_logits override to enable Routing Replay for MoE models",
+    "updated_at": "2026-04-14T21:05:38Z"
   },
   {
-    "additions": 14,
-    "author": "yonigozlan",
+    "additions": 1346,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Add compatibility with remote code importing image_processing_utils_fast modules and methods using `from transformers.image_processing_utils_fast import ...`",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? As per the title. This PR finally makes mamba layer caches first class citizen, and adds native support for them. It supports the following layers combinations: - all mamba layers - alternating attention layer/mamba\u2026",
+    "changed_files": 64,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44897",
-    "created_at": "2026-03-20T19:30:32Z",
-    "deletions": 5,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44950",
+    "created_at": "2026-03-23T16:25:13Z",
+    "deletions": 4113,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44897/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44897",
+    "files_url": "https://github.com/huggingface/transformers/pull/44950/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44950",
     "labels": [],
     "merged": true,
-    "number": 44897,
-    "review_comments_count": 0,
+    "number": 44950,
+    "review_comments_count": 48,
     "state": "closed",
-    "title": "Add backward compatibility for direct imports from legacy `image_processing_utils_fast`",
-    "updated_at": "2026-03-20T20:00:12Z"
+    "title": "\ud83d\udea8 [Cache] Native mamba & hybrid cache",
+    "updated_at": "2026-03-31T13:09:44Z"
   },
   {
-    "additions": 354,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "updates the continuous batching docs - new page for the API reference - adds sections for new features like CUDA graphs, async batching, prefix caching, logprobs (depending on when its merged) - clearer example of generation with varying l\u2026",
-    "changed_files": 4,
+    "additions": 101,
+    "author": "Charly21r",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #44936 This PR fixes an issue with `NotebookProgressCallback` in the `Trainer` where calling evaluate() before or after training would crash due to the training tracker being `None`. The callback now properly\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44896",
-    "created_at": "2026-03-20T19:09:41Z",
-    "deletions": 81,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44949",
+    "created_at": "2026-03-23T16:07:50Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44896/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44896",
+    "files_url": "https://github.com/huggingface/transformers/pull/44949/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44949",
     "labels": [],
     "merged": true,
-    "number": 44896,
-    "review_comments_count": 2,
+    "number": 44949,
+    "review_comments_count": 13,
     "state": "closed",
-    "title": "[docs] continuous batching",
-    "updated_at": "2026-03-30T17:17:13Z"
+    "title": "Fix: NotebookProgressCallback crash when evaluating with the Trainer",
+    "updated_at": "2026-04-13T13:47:51Z"
   },
   {
-    "additions": 57,
-    "author": "SunMarc",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR enables static FP8 experts. This also works on multi-gpu with device-map. A fix for that was to set was to set `torch.cuda.set_device()`. Triton's JIT compiler uses he active device context to determine whic\u2026",
+    "additions": 1,
+    "author": "heycorgi",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44895",
-    "created_at": "2026-03-20T19:01:35Z",
-    "deletions": 10,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44948",
+    "created_at": "2026-03-23T15:33:56Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44895/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44895",
+    "files_url": "https://github.com/huggingface/transformers/pull/44948/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44948",
     "labels": [],
-    "merged": true,
-    "number": 44895,
-    "review_comments_count": 4,
+    "merged": false,
+    "number": 44948,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add static FP8 expert support ",
-    "updated_at": "2026-03-24T14:27:31Z"
+    "title": "Create aa.py",
+    "updated_at": "2026-03-23T15:34:35Z"
   },
   {
-    "additions": 10,
-    "author": "ydshieh",
+    "additions": 123,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "## Problem `ProcessorMixin.to_dict()` was calling `copy.deepcopy(self.__dict__)` on the entire processor, including the tokenizer, even though the tokenizer is always deleted from the output immediately after (since tokenizers are saved se\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? The doc was generated by Claude. I deleted unnecessary repetitions and fixed a few moments to be more precise. We don't really need to merge it now so if you think the text is too LLM, feel free to take this as an i\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44894",
-    "created_at": "2026-03-20T18:57:53Z",
-    "deletions": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44947",
+    "created_at": "2026-03-23T13:23:04Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44894/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44894",
+    "files_url": "https://github.com/huggingface/transformers/pull/44947/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44947",
     "labels": [],
     "merged": true,
-    "number": 44894,
-    "review_comments_count": 0,
+    "number": 44947,
+    "review_comments_count": 16,
     "state": "closed",
-    "title": "fix `processing_utils.py`: avoid deepcopying tokenizer in `ProcessorMixin` to improve performance",
-    "updated_at": "2026-03-23T10:09:02Z"
+    "title": "Add doc page for capturing outputs",
+    "updated_at": "2026-03-26T13:08:46Z"
   },
   {
-    "additions": 18,
-    "author": "ai-man-codes",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #43011 `StaticLayer` was missing a `.crop(max_length)` method, so implemented that according to the discussion of the issue. Added `StaticLayer.crop(max_length)` to match the API of StaticCache with the Dynami\u2026",
-    "changed_files": 1,
+    "additions": 14,
+    "author": "BSchilperoort",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44893",
-    "created_at": "2026-03-20T17:48:23Z",
-    "deletions": 0,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44946",
+    "created_at": "2026-03-23T12:18:34Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44893/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44893",
+    "files_url": "https://github.com/huggingface/transformers/pull/44946/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44946",
     "labels": [],
-    "merged": false,
-    "number": 44893,
+    "merged": true,
+    "number": 44946,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "add `StaticLayer.crop()` to match `DynamicLayer` API",
-    "updated_at": "2026-03-28T01:41:22Z"
+    "state": "closed",
+    "title": "Correct docstrings for `from_pretrained` (url input deprecated)",
+    "updated_at": "2026-03-23T13:05:16Z"
   },
   {
-    "additions": 51,
-    "author": "he-yufeng",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #44821 The `elif is_remote_url(...)` / `download_url(...)` branch in `get_image_processor_dict` was accidentally removed during the image processor refactor in #43514. This caused `AutoImageProcessor.from_pretrained(url)` to break wi\u2026",
+    "additions": 71,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? @hmellor",
     "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44892",
-    "created_at": "2026-03-20T16:21:25Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44943",
+    "created_at": "2026-03-23T10:58:40Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44892/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44892",
+    "files_url": "https://github.com/huggingface/transformers/pull/44943/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44943",
     "labels": [],
-    "merged": false,
-    "number": 44892,
-    "review_comments_count": 4,
+    "merged": true,
+    "number": 44943,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Fix AutoImageProcessor.from_pretrained failing on URL input",
-    "updated_at": "2026-03-24T13:30:38Z"
+    "title": "Clearer type hints and fix rope validation in configs",
+    "updated_at": "2026-03-23T13:32:11Z"
   },
   {
-    "additions": 507,
-    "author": "kashif",
+    "additions": 220,
+    "author": "hmellor",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Add a MoERouterHealthCallback to log MoE router-health metrics. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the\u2026",
-    "changed_files": 7,
+    "body_excerpt": null,
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44891",
-    "created_at": "2026-03-20T16:17:05Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44891/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44891",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44942",
+    "created_at": "2026-03-23T10:46:23Z",
+    "deletions": 5,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44942/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44942",
     "labels": [],
     "merged": false,
-    "number": 44891,
+    "number": 44942,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[Trainer] add MoERouterHealthCallback Callback",
-    "updated_at": "2026-03-20T16:28:43Z"
+    "title": "Add inference time layer fusion optimisations via `PreTrainedModel.from_pretrained(fuse_layers=True)`",
+    "updated_at": "2026-03-25T16:14:19Z"
   },
   {
-    "additions": 72,
-    "author": "Rocketknight1",
+    "additions": 4,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "As discussed on Slack, this is the first phase of our approach to controlling the code agent epidemic. This PR places large warnings in both the pull request template and `CONTRIBUTING.md`, which should hopefully be seen by most contributo\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Fix the failing job after #43514 (the fix is effefctive, see [here](https://github.com/huggingface/transformers/actions/runs/23433395911/job/68165255513?pr=44941)) [Update Transformers metadata](https://github.com/h\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44890",
-    "created_at": "2026-03-20T16:12:45Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44941",
+    "created_at": "2026-03-23T10:42:09Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44890/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44890",
+    "files_url": "https://github.com/huggingface/transformers/pull/44941/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44941",
     "labels": [],
     "merged": true,
-    "number": 44890,
-    "review_comments_count": 7,
+    "number": 44941,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Add big angry code agent warnings!",
-    "updated_at": "2026-03-23T11:54:48Z"
+    "title": "Fix failing job `Update Transformers metadata` after #43514",
+    "updated_at": "2026-03-23T13:41:39Z"
   },
   {
-    "additions": 86,
-    "author": "roycho96",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Calling `trainer.evaluate()` before `trainer.train()` with DeepSpeed is broken in three ways: 1. **ZeRO-3 stale state crash:** `evaluate()` creates an inference engine. `train()` starts with `accelerator.free_memor\u2026",
-    "changed_files": 2,
+    "additions": 138,
+    "author": "Qubitium",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Model loading of same model path but 2 different threads (2 different instances) have meta device tensor issues: unloaded meta/empty embedding/lm-head when it should not be empty post model load. Cause: `tie_weight(\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44889",
-    "created_at": "2026-03-20T15:08:32Z",
-    "deletions": 21,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44940",
+    "created_at": "2026-03-23T09:55:57Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44889/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44889",
+    "files_url": "https://github.com/huggingface/transformers/pull/44940/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44940",
     "labels": [],
     "merged": false,
-    "number": 44889,
-    "review_comments_count": 0,
+    "number": 44940,
+    "review_comments_count": 3,
     "state": "open",
-    "title": "[DeepSpeed] Fix evaluate()/predict() before train()",
-    "updated_at": "2026-03-21T11:06:07Z"
+    "title": "Fix tie_weights skipping logic is not tied to model thread scope",
+    "updated_at": "2026-04-07T02:01:50Z"
   },
   {
-    "additions": 2,
-    "author": "Cyrilvallez",
+    "additions": 2038,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? In general, it should be much better to let the kernel do what it wants for perfs! There's no reasons to have troubles from it!",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Refactored and cleaned up model linter - separated package - one rule per module - refactored legacy checks into their own rules - simplified pattern, duplication removal",
+    "changed_files": 25,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44888",
-    "created_at": "2026-03-20T14:45:28Z",
-    "deletions": 22,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44939",
+    "created_at": "2026-03-23T08:45:36Z",
+    "deletions": 1446,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44888/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44888",
+    "files_url": "https://github.com/huggingface/transformers/pull/44939/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44939",
     "labels": [],
-    "merged": false,
-    "number": 44888,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Remove explicit cuda stream in nemotron_h",
-    "updated_at": "2026-03-26T19:15:29Z"
+    "merged": true,
+    "number": 44939,
+    "review_comments_count": 5,
+    "state": "closed",
+    "title": "refactor: mlinter as its own package",
+    "updated_at": "2026-03-24T07:56:15Z"
   },
   {
     "additions": 2,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. On currently pinned version, when we run this small snippet (which is called on some model's `__init__` functions \ud83d\ude05): ```python from transformers.integrations.hub_kernels import lazy_load_kernel ca\u2026",
-    "changed_files": 2,
+    "author": "VanshikaSohal",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes two small but impactful bugs in the BART documentation: 1. **Variable shadowing bug**: In the Pipeline example, the variable was named `pipeline` which shadows the imported `pipeline` function. Renamed to `fi\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44887",
-    "created_at": "2026-03-20T14:00:33Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44935",
+    "created_at": "2026-03-22T18:45:01Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44887/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44887",
+    "files_url": "https://github.com/huggingface/transformers/pull/44935/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44935",
     "labels": [],
     "merged": true,
-    "number": 44887,
-    "review_comments_count": 1,
+    "number": 44935,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Bump kernels version dependency to avoid crashes",
-    "updated_at": "2026-03-20T19:01:51Z"
+    "title": "Fix variable shadowing in pipeline example and typo in BART docs (BERT \u2192 BART)",
+    "updated_at": "2026-03-23T14:28:04Z"
   },
   {
-    "additions": 14,
-    "author": "m-matthias",
+    "additions": 9,
+    "author": "Sai-Suraj-27",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Prevent crash in class LwDetrImageLoss when using it with float16 automatic mixed precision on a Cuda device. torch.pow causes an autocast to float32 when used with Cuda, which caused a type mismatch at ``` pos_weig\u2026",
+    "body_excerpt": "# What does this PR do? Fixes this failing [T5ModelIntegrationTest](https://github.com/huggingface/transformers/actions/runs/23230643883/job/67524758706#step:14:1449) & this [Qwen2IntegrationTest](https://github.com/huggingface/transformer\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44886",
-    "created_at": "2026-03-20T13:56:08Z",
-    "deletions": 12,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44934",
+    "created_at": "2026-03-22T18:03:34Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44886/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44886",
+    "files_url": "https://github.com/huggingface/transformers/pull/44934/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44934",
     "labels": [],
     "merged": true,
-    "number": 44886,
-    "review_comments_count": 4,
+    "number": 44934,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "LwDetrImageLoss: Fix dtype casting to prevent crash when using amp on cuda device",
-    "updated_at": "2026-03-24T17:02:32Z"
+    "title": "Fix failing `T5ModelIntegrationTest`",
+    "updated_at": "2026-03-24T14:50:10Z"
   },
   {
-    "additions": 2,
-    "author": "guoyangzhen",
-    "author_association": "NONE",
-    "body_excerpt": "## Problem In _split_tokens_on_unicode(), when the decoded token stream ends with a dangling Unicode replacement character (U+FFFD), the computed index can equal len(decoded_full), causing IndexError: string index out of range. The failing\u2026",
+    "additions": 1,
+    "author": "r266-tech",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #44908 The `get_inverse_sqrt_schedule` function accepts `timescale` and `last_epoch` parameters, but `get_scheduler` was not forwarding `scheduler_specific_kwargs` to it. This caused user-provided kwargs like\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44885",
-    "created_at": "2026-03-20T13:03:54Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44932",
+    "created_at": "2026-03-22T17:30:56Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44885/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44885",
+    "files_url": "https://github.com/huggingface/transformers/pull/44932/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44932",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44885,
+    "number": 44932,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: prevent IndexError in Whisper word timestamp decode",
-    "updated_at": "2026-03-23T12:01:50Z"
+    "title": "Fix: Pass scheduler_specific_kwargs to inverse_sqrt scheduler",
+    "updated_at": "2026-03-23T12:44:16Z"
   },
   {
-    "additions": 14,
-    "author": "hmellor",
-    "author_association": "MEMBER",
-    "body_excerpt": "Some libraries that use Transformers (i.e. vLLM) use `|` on the `size` config. This PR adds `__or__` and `__ror__` so that the following works: ```console $ {\"longest_edge\": 20} | SizeDict(height=10, width=20) {'longest_edge': 20, 'height'\u2026",
+    "additions": 1,
+    "author": "r266-tech",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes a v5 regression where `CamembertForMaskedLM` (and all CamemBERT masked-LM tasks) produces near-zero, near-uniform logits, making the model completely non-functional. ### Root cause In v5, `modeling_utils.get_\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44884",
-    "created_at": "2026-03-20T11:52:15Z",
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44931",
+    "created_at": "2026-03-22T17:28:57Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44884/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44884",
+    "files_url": "https://github.com/huggingface/transformers/pull/44931/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44931",
     "labels": [],
     "merged": true,
-    "number": 44884,
-    "review_comments_count": 2,
+    "number": 44931,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add missing dunder methods to `SizeDict`",
-    "updated_at": "2026-03-20T12:21:12Z"
+    "title": "fix(camembert): add tie_word_embeddings=True to CamembertConfig",
+    "updated_at": "2026-03-25T07:09:37Z"
   },
   {
-    "additions": 2,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/transformers/issues/44589.",
-    "changed_files": 1,
+    "additions": 103,
+    "author": "javierdejesusda",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary - Fixes #44912 \u2014 MXFP4 quantization error messages combine `is_triton_available()` and `is_kernels_available()` into a single `kernels_available` boolean, making it impossible to identify which dependency is missing - Split the\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44883",
-    "created_at": "2026-03-20T11:43:13Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44930",
+    "created_at": "2026-03-22T17:27:20Z",
+    "deletions": 13,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44883/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44883",
+    "files_url": "https://github.com/huggingface/transformers/pull/44930/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44930",
     "labels": [],
     "merged": true,
-    "number": 44883,
-    "review_comments_count": 2,
+    "number": 44930,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Fix dtype guessing from state dict",
-    "updated_at": "2026-03-20T13:12:34Z"
+    "title": "fix: split MXFP4 dependency checks for specific error messages",
+    "updated_at": "2026-03-24T15:33:14Z"
   },
   {
-    "additions": 1,
-    "author": "itazap",
+    "additions": 26,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "fixes ```python model = \"meta-llama/Llama-4-Maverick-17B-128E-Instruct\" tok_auto = AutoTokenizer.from_pretrained(model) print(f\"AutoTokenizer: {tok_auto('hello')}\") ``` ``` The above exception was the direct cause of the following exceptio\u2026",
+    "body_excerpt": "## Problem In `TokenizersBackend.convert_to_native_format()`, when a tokenizer has a custom `__init__` (the `elif` branch), `tokenizer.json` was parsed **twice**: 1. `TokenizerFast.from_file(fast_tokenizer_file)` \u2014 full Rust parse includin\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44882",
-    "created_at": "2026-03-20T11:31:20Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44927",
+    "created_at": "2026-03-22T15:33:23Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44882/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44882",
+    "files_url": "https://github.com/huggingface/transformers/pull/44927/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44927",
     "labels": [],
-    "merged": false,
-    "number": 44882,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44927,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "fix config type",
-    "updated_at": "2026-03-20T16:34:20Z"
+    "title": "fix: improve processor loading performance by avoiding redundant tokenizer parsing",
+    "updated_at": "2026-03-23T11:03:52Z"
   },
   {
-    "additions": 142,
-    "author": "zucchini-nlp",
+    "additions": 25,
+    "author": "yonigozlan",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per title, we don't need a weird way to filter out kwargs anymore because now we don't rely on `tokenizer.apply_chat_template`. I didn't delete the unused `TypedDict` yet and will deprecate for at least 3 minor r\u2026",
-    "changed_files": 6,
+    "body_excerpt": "# What does this PR do? Solve import errors when trying to import `from transformers.models.llama4.image_processing_llama4_fast import Llama4ImageProcessorFast` for example",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44881",
-    "created_at": "2026-03-20T10:44:06Z",
-    "deletions": 82,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44926",
+    "created_at": "2026-03-22T14:46:17Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44881/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44881",
+    "files_url": "https://github.com/huggingface/transformers/pull/44926/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44926",
     "labels": [],
     "merged": true,
-    "number": 44881,
-    "review_comments_count": 12,
+    "number": 44926,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Allow arbitrary template kwargs in processors",
-    "updated_at": "2026-03-27T11:07:08Z"
+    "title": "Fix backward compatibility for full path imports of Fast Image Processors",
+    "updated_at": "2026-03-23T14:16:49Z"
   },
   {
-    "additions": 34,
-    "author": "itazap",
+    "additions": 96,
+    "author": "kashif",
     "author_association": "MEMBER",
-    "body_excerpt": "incorrect model list update",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? This PR adds a first-class MoE routing capture/replay API for Qwen2Moe and introduces shared MoE routing helpers for reuse by other MoE model families. It adds: - a structured `MoERouting` payload in modeling output\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44880",
-    "created_at": "2026-03-20T10:37:13Z",
-    "deletions": 5,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44925",
+    "created_at": "2026-03-22T14:04:40Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44880/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44880",
+    "files_url": "https://github.com/huggingface/transformers/pull/44925/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44925",
     "labels": [],
     "merged": true,
-    "number": 44880,
+    "number": 44925,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "incorrect model list update",
-    "updated_at": "2026-03-24T09:27:24Z"
+    "title": "[Doc]  MoE routing capture and replay recipe ",
+    "updated_at": "2026-04-14T08:09:16Z"
   },
   {
-    "additions": 448,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - uses the Makefile as a single source of truth for running QA checks - adds `tomli` so `make` commands can read the `toml` file when needed - adds a `checkers` Python module that wraps and orchestrates all `checks`\u2026",
-    "changed_files": 7,
+    "additions": 9,
+    "author": "Qubitium",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fix two `nogil` threading bugs (reproduced on 3.14) : 1. Continus Batching crashes with torch graph errors with 2 threads on 2 separate model instances (same model path, but two distinct instances). Cause is missing\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44879",
-    "created_at": "2026-03-20T10:24:29Z",
-    "deletions": 90,
+    "comments_count": 13,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44924",
+    "created_at": "2026-03-22T11:46:49Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44879/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44879",
+    "files_url": "https://github.com/huggingface/transformers/pull/44924/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44924",
     "labels": [],
     "merged": true,
-    "number": 44879,
-    "review_comments_count": 6,
+    "number": 44924,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "refactor: unify QA calls",
-    "updated_at": "2026-03-25T08:51:30Z"
+    "title": "Continuous batching thread safety",
+    "updated_at": "2026-03-24T05:42:56Z"
   },
   {
-    "additions": 8,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? `check_docstrings` has been complaining for a while about those.",
+    "additions": 3,
+    "author": "prakhar-agarwal",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Addresses issue #44843. Verified with isolated repro logic. Changes made: Updated the logic to properly identify local and offline scenarios upfront. Now, is_local is correctly set to True if: 1. is_offline_mode() is active. 2. The local_f\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44878",
-    "created_at": "2026-03-20T10:01:08Z",
-    "deletions": 8,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44878/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44878",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44923",
+    "created_at": "2026-03-22T05:20:22Z",
+    "deletions": 1,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44923/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44923",
     "labels": [],
-    "merged": true,
-    "number": 44878,
+    "merged": false,
+    "number": 44923,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix nemotron config docstrings",
-    "updated_at": "2026-03-20T10:11:04Z"
+    "state": "open",
+    "title": "fix: avoid unconditional model_info call in _patch_mistral_regex",
+    "updated_at": "2026-03-22T05:24:11Z"
   },
   {
-    "additions": 1,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do?",
-    "changed_files": 2,
+    "additions": 10,
+    "author": "s-zx",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes #44849. When `output_hidden_states=True` (or `output_attentions=True`) is passed to `model.generate()`, the `@capture_outputs` decorator reads the flag value but leaves it in `**kwargs`. These flags then prop\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44876",
-    "created_at": "2026-03-20T09:49:54Z",
-    "deletions": 7,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44922",
+    "created_at": "2026-03-22T01:21:22Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44876/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44876",
-    "labels": [],
-    "merged": true,
-    "number": 44876,
+    "files_url": "https://github.com/huggingface/transformers/pull/44922/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44922",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44922,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix nemotron_h modular",
-    "updated_at": "2026-03-20T10:00:35Z"
+    "title": "fix: pop output_* flags from kwargs in capture_outputs to prevent submodule leakage",
+    "updated_at": "2026-03-23T12:38:56Z"
   },
   {
-    "additions": 872,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Refactors `src/transformers/cli/serve.py` to reduce nesting depth, eliminate code duplication, and improve maintainability. No behavioral changes and the public API is unchanged. Also added a module docstring to exp\u2026",
-    "changed_files": 1,
+    "additions": 4,
+    "author": "s-zx",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes #44918. `compute_3d_position_ids` in the Qwen2.5-VL / Qwen3-VL / Qwen3.5 model families destructures `inputs_embeds.shape` into exactly three variables: ```python batch_size, seq_length, _ = inputs_embeds.sha\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44875",
-    "created_at": "2026-03-20T09:06:34Z",
-    "deletions": 701,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44875/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44875",
-    "labels": [],
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44921",
+    "created_at": "2026-03-22T00:39:01Z",
+    "deletions": 4,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44921/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44921",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44875,
+    "number": 44921,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "refactor: improved the cli server module code organization",
-    "updated_at": "2026-03-23T08:08:17Z"
+    "state": "closed",
+    "title": "fix: use shape index access in compute_3d_position_ids for Qwen VL models",
+    "updated_at": "2026-03-23T10:00:51Z"
   },
   {
-    "additions": 2,
-    "author": "hmellor",
-    "author_association": "MEMBER",
-    "body_excerpt": "`Llama4`'s was incorrect and causing `StrictDataclassFieldValidationErrors`. `AFMoE`'s was was fine but now it's more specific.",
+    "additions": 15,
+    "author": "s-zx",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes `num_labels` not being propagated from `Qwen3_5Config` to its `text_config` when loading via `AutoConfig.from_pretrained(model, num_labels=N)`. **Root cause:** `Qwen3_5Config.__post_init__` initializes `text_\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44874",
-    "created_at": "2026-03-20T09:05:02Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44920",
+    "created_at": "2026-03-22T00:01:59Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44874/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44874",
+    "files_url": "https://github.com/huggingface/transformers/pull/44920/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44920",
     "labels": [],
-    "merged": true,
-    "number": 44874,
+    "merged": false,
+    "number": 44920,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix `layer_types` type hint for `AFMoE` and `Llama4`",
-    "updated_at": "2026-03-20T12:03:58Z"
+    "title": "fix: propagate num_labels/id2label to text_config in Qwen3_5Config",
+    "updated_at": "2026-03-23T12:06:04Z"
   },
   {
-    "additions": 75,
-    "author": "sergiopaniego",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? ## Problem Online RL training (GRPO, RLOO, PPO) with all VL models using MRoPE with rope_deltas (Qwen2-VL, Qwen2.5-VL, Qwen3-VL, Qwen3.5, GLM4V, PaddleOCR-VL, Ernie4.5-VL-MoE, etc.) crashes with `RuntimeError: Sizes\u2026",
-    "changed_files": 15,
+    "additions": 18,
+    "author": "s-zx",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes a crash in `Qwen2_5_VLProcessor.__call__` when processing batched inputs without padding (`padding=False`). **Root cause:** When the tokenizer returns sequences of different lengths (ragged list), `np.array(t\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44873",
-    "created_at": "2026-03-20T08:38:03Z",
-    "deletions": 30,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44919",
+    "created_at": "2026-03-21T23:57:37Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44873/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44873",
+    "files_url": "https://github.com/huggingface/transformers/pull/44919/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44919",
     "labels": [],
-    "merged": true,
-    "number": 44873,
+    "merged": false,
+    "number": 44919,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix VL model rope_deltas batch size mismatch in online RL training",
-    "updated_at": "2026-03-20T13:51:08Z"
+    "title": "fix: handle ragged batch inputs in Qwen2_5_VLProcessor mm_token_type_ids computation",
+    "updated_at": "2026-03-23T10:38:30Z"
   },
   {
-    "additions": 2,
-    "author": "IvanFan-Van",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Description Update outdated comment that references non-existent file `generation_utils_samplers.py` ## Changes Detail - The comment on line 1200 states \"all samplers can be found in `generation_utils_samplers.py`\" - In reality, all sam\u2026",
+    "additions": 5,
+    "author": "s-zx",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary `GPTNeoXConfig.convert_rope_params_to_dict` unconditionally overwrote `rope_parameters[\"partial_rotary_factor\"]` with the default `0.25` when `rotary_pct` was absent from kwargs. On every `from_pretrained` call, `rotary_pct` is\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44872",
-    "created_at": "2026-03-20T05:45:46Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44917",
+    "created_at": "2026-03-21T23:34:32Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44872/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44872",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44917/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44917",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44872,
+    "number": 44917,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix: Update outdated sampler comment in generation/utils.py",
-    "updated_at": "2026-03-20T05:45:46Z"
+    "state": "closed",
+    "title": "fix(gpt-neox): preserve rotary_pct across save/load cycle",
+    "updated_at": "2026-03-23T12:37:48Z"
   },
   {
-    "additions": 666,
-    "author": "JonusClapshaw",
+    "additions": 8,
+    "author": "s-zx",
     "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Fixes #42200 `prediction_step` is type-hinted to return `Optional[torch.Tensor]` for logits, but when no `preprocess_logits_for_metrics` is provided it could return a tuple instead of a tensor. This caused `torch_pa\u2026",
-    "changed_files": 33,
+    "body_excerpt": "## Summary Importing `DebertaV2Model` (or anything that depends on it, e.g. `gliner`) raises `IndentationError` on Python 3.13 because `torch.jit.script` calls `inspect.getsource()`, dedents the snippet, and passes it to `ast.parse()`. Pyt\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44870",
-    "created_at": "2026-03-20T02:28:27Z",
-    "deletions": 3,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44916",
+    "created_at": "2026-03-21T23:34:07Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44870/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44870",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44916/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44916",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44870,
+    "number": 44916,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: ensure prediction_step returns tensor for logits, not tuple #42200",
-    "updated_at": "2026-03-20T17:51:19Z"
+    "title": "fix(deberta-v2): move \"Copied from\" comments above @torch.jit.script for Python 3.13 compat",
+    "updated_at": "2026-03-23T12:34:24Z"
   },
   {
-    "additions": 98,
-    "author": "sdharani91",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #44717 This PR fixes packed-sequence handling for the Qwen3.5 linear-attention fast path. Before this change, Qwen3.5 produced different outputs for: a padded representation of multiple sequences a packed repr\u2026",
-    "changed_files": 3,
+    "additions": 90,
+    "author": "maxsloef-goodfire",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? `clean_up_tokenization` applies English-specific string replacements (` .` \u2192 `.`, ` ?` \u2192 `?`, ` ,` \u2192 `,`, etc.) to decoded text. This was designed for BERT-era WordPiece tokenizers where decoding produced artifacts\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44867",
-    "created_at": "2026-03-19T17:31:45Z",
-    "deletions": 5,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44915",
+    "created_at": "2026-03-21T20:45:03Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44867/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44867",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44915/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44915",
+    "labels": [],
     "merged": false,
-    "number": 44867,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Pass packed boundary metadata to Qwen3.5 linear-attention fast kernels",
-    "updated_at": "2026-03-26T21:02:58Z"
+    "number": 44915,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "fix: skip `clean_up_tokenization` for BPE tokenizers in `PreTrainedTokenizerFast`",
+    "updated_at": "2026-04-09T13:39:02Z"
   },
   {
-    "additions": 78,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title",
-    "changed_files": 3,
+    "additions": 1,
+    "author": "maxsloef-goodfire",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? The `Llama3Converter` in `convert_llama_weights_to_hf.py` hardcodes `clean_up_tokenization_spaces=True` (line 468). This causes `tokenizer.decode()` to silently strip spaces before punctuation for all converted Lla\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44866",
-    "created_at": "2026-03-19T17:27:58Z",
-    "deletions": 75,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44914",
+    "created_at": "2026-03-21T20:25:51Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44866/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44866",
+    "files_url": "https://github.com/huggingface/transformers/pull/44914/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44914",
     "labels": [],
     "merged": true,
-    "number": 44866,
+    "number": 44914,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Align lfm2 cache to other mamba caches",
-    "updated_at": "2026-03-20T10:50:28Z"
+    "title": "fix: set `clean_up_tokenization_spaces=False` in Llama 3 tokenizer conversion",
+    "updated_at": "2026-03-23T08:38:18Z"
   },
   {
-    "additions": 496,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Added Rule 11 forward() must not access non-nn.Module attributes on submodules (breaks pipeline parallelism with Identity replacement). we want to make sure we just use metadata in config and elesewere when in that\u2026",
-    "changed_files": 10,
+    "additions": 8,
+    "author": "ouroborosscr",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "Qwen3.5 uses 3D position_ids [3, batch, seq_len] for multi-dimensional rotary embedding. _is_packed_sequence() misinterprets this as a packed sequence, causing cu_seqlens to be constructed with 3x the actual token count. Flash attention th\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44865",
-    "created_at": "2026-03-19T16:39:59Z",
-    "deletions": 26,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44911",
+    "created_at": "2026-03-21T15:42:57Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44865/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44865",
+    "files_url": "https://github.com/huggingface/transformers/pull/44911/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44911",
     "labels": [],
-    "merged": true,
-    "number": 44865,
-    "review_comments_count": 1,
+    "merged": false,
+    "number": 44911,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "chore(typing): added rule 11",
-    "updated_at": "2026-03-23T12:29:21Z"
+    "title": "Fix flash attention crash with 3D position_ids (Qwen3.5)",
+    "updated_at": "2026-03-24T14:35:57Z"
   },
   {
-    "additions": 99,
-    "author": "SunMarc",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR switches FP8 per-tensor implementation to rely on the official torch impl `torch._scaled_mm`. Note that `torch._scaled_mm` don't explicitly support per tensor. We hack the api a bit as it only support per ro\u2026",
+    "additions": 1,
+    "author": "anshuS1310",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "The `get_scheduler` function was identifying the `inverse_sqrt` scheduler type but failing to pass `**scheduler_specific_kwargs` to the underlying `get_inverse_sqrt_schedule` function. This caused user-defined parameters like `timescale` t\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44864",
-    "created_at": "2026-03-19T16:19:53Z",
-    "deletions": 12,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44909",
+    "created_at": "2026-03-21T09:59:07Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44864/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44864",
+    "files_url": "https://github.com/huggingface/transformers/pull/44909/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44909",
     "labels": [],
-    "merged": false,
-    "number": 44864,
+    "merged": true,
+    "number": 44909,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Switch FP8 per tensor quant to use `torch._scaled_mm`",
-    "updated_at": "2026-03-20T19:05:05Z"
+    "title": "Fix: Update optimization.py",
+    "updated_at": "2026-03-24T13:06:15Z"
   },
   {
-    "additions": 19,
-    "author": "gh-wf",
-    "author_association": "NONE",
-    "body_excerpt": "Some models (e.g. Nemotron-H) define `_tied_weights_keys` as a list, which caused `AttributeError: 'list' object has no attribute 'keys'` when calling `save_pretrained` during full finetuning. # What does this PR do? `_get_tied_weight_keys\u2026",
-    "changed_files": 2,
+    "additions": 200,
+    "author": "syncdoth",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #44906 ## Summary - Remove `.expand_as(inputs_embeds)` from placeholder mask creation in `get_placeholder_mask` and equivalent inline patterns across all VLM models. `masked_scatter` natively broadcasts `(B, S, 1)` \u2192 `(B, S, H)`, mak\u2026",
+    "changed_files": 71,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44862",
-    "created_at": "2026-03-19T15:14:12Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44907",
+    "created_at": "2026-03-21T06:07:35Z",
+    "deletions": 222,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44862/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44862",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44907/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44907",
+    "labels": [],
     "merged": false,
-    "number": 44862,
+    "number": 44907,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: handle list-type _tied_weights_keys in _get_tied_weight_keys",
-    "updated_at": "2026-03-20T09:47:09Z"
+    "state": "open",
+    "title": "Remove unnecessary expand_as in get_placeholder_mask across VLMs",
+    "updated_at": "2026-03-23T12:20:03Z"
   },
   {
-    "additions": 11,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. As discussed offline with the Mistral team, the scaling applied to the query should not be the absolute one (old `cache_position`), but the actual `position_ids`, taking into account padding, packe\u2026",
-    "changed_files": 4,
+    "additions": 13,
+    "author": "NicoleRobin",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - 13 i18n README files used `./awesome-transformers.md` which resolves relative to the `i18n/` directory and leads to a 404 - Replace with the absolute GitHub URL so links work from any location - `README_ko.md` was already corr\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44860",
-    "created_at": "2026-03-19T14:27:33Z",
-    "deletions": 17,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44905",
+    "created_at": "2026-03-21T03:25:56Z",
+    "deletions": 13,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44860/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44860",
+    "files_url": "https://github.com/huggingface/transformers/pull/44905/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44905",
     "labels": [],
     "merged": true,
-    "number": 44860,
+    "number": 44905,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Mistral] Fix query scaling for Mistral4 and Ministral3",
-    "updated_at": "2026-03-19T18:02:06Z"
+    "title": "fix(i18n): replace broken relative links to awesome-transformers.md with absolute URLs",
+    "updated_at": "2026-03-23T12:47:56Z"
   },
   {
-    "additions": 7001,
-    "author": "philippguevorguian",
+    "additions": 101,
+    "author": "vivekvar-dl",
     "author_association": "NONE",
-    "body_excerpt": null,
-    "changed_files": 19,
+    "body_excerpt": "# Fix granite_speech config loading failure with int multiplier fields ## Fixes #44877 ### Problem Loading `granite_speech` configs fails with `StrictDataclassFieldValidationError` when multiplier fields (e.g., `embedding_multiplier`) are\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44859",
-    "created_at": "2026-03-19T13:54:19Z",
-    "deletions": 138,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44904",
+    "created_at": "2026-03-21T03:12:37Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44859/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44859",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44904/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44904",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44859,
+    "number": 44904,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "refactor: rope in model, flatten vision, rely on qwen3 backone, misc changes",
-    "updated_at": "2026-03-19T14:08:01Z"
+    "title": "fix(granite_speech): convert int to float for multiplier fields in text_config",
+    "updated_at": "2026-03-23T10:37:38Z"
   },
   {
-    "additions": 111,
-    "author": "remi-or",
+    "additions": 16,
+    "author": "yonigozlan",
     "author_association": "MEMBER",
-    "body_excerpt": "Right now, the continuous batching tests all use similar mechanisms, namely: 1. loading a model and a tokenizer 2. preparing data for generate or generate_batch 3. running generate to compare its outputs with generate_batch This PR adds 3\u2026",
+    "body_excerpt": "# What does this PR do? Some remote code models are using `get_size_dict` directly, and now that size is converted to SizeDict in init, we need to support it as input in `get_size_dict`",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44858",
-    "created_at": "2026-03-19T13:22:04Z",
-    "deletions": 188,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44903",
+    "created_at": "2026-03-21T01:25:53Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44858/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44858",
+    "files_url": "https://github.com/huggingface/transformers/pull/44903/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44903",
     "labels": [],
     "merged": true,
-    "number": 44858,
-    "review_comments_count": 0,
+    "number": 44903,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "[CB] [Minor] Simplify test suite",
-    "updated_at": "2026-03-24T11:44:39Z"
+    "title": "Support SizeDict import in get_size_dict",
+    "updated_at": "2026-03-23T10:28:52Z"
   },
   {
     "additions": 3,
-    "author": "hkc5",
+    "author": "guoyangzhen",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Python 3.13's stricter parser fails when there's a comment between the `@torch.jit.script` decorator and the function definition, causing an IndentationError when importing DebertaV2Model. ## Changes - Moved comments before the\u2026",
+    "body_excerpt": "## Problem `_split_tokens_on_unicode()` crashes with `IndexError: string index out of range` when the decoded token stream ends with a dangling Unicode replacement character (\\uFFFD). The computed index `unicode_offset + decoded.index(repl\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44856",
-    "created_at": "2026-03-19T12:33:00Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44902",
+    "created_at": "2026-03-20T22:08:49Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44856/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44856",
+    "files_url": "https://github.com/huggingface/transformers/pull/44902/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44902",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44856,
+    "number": 44902,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: move comments before @torch.jit.script decorator for Python 3.13 compatibility",
-    "updated_at": "2026-03-19T13:11:44Z"
+    "title": "fix: Whisper word timestamp OOB access on trailing replacement char",
+    "updated_at": "2026-03-23T11:59:14Z"
   },
   {
-    "additions": 63,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? We had (flaky) ```bash tests/models/nemotron_h/test_modeling_nemotron_h.py::NemotronHModelTest::test_sdpa_can_compile_dynamic Fatal Python error: Segmentation fault ``` `NemotronHBlock.forward` creates a temporary `\u2026",
+    "additions": 19,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? The following failing Perceiver use case was identified and fixed in this PR: \u2192 c6d2848a23 ([\ud83d\udea8 Fix torch.jit.trace for interpolate_pos_encoding in all vision models](https://github.com/huggingface/transformers/pul\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44854",
-    "created_at": "2026-03-19T10:54:36Z",
-    "deletions": 56,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44899",
+    "created_at": "2026-03-20T20:02:10Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44854/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44854",
+    "files_url": "https://github.com/huggingface/transformers/pull/44899/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44899",
     "labels": [],
     "merged": true,
-    "number": 44854,
-    "review_comments_count": 5,
+    "number": 44899,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix core dumped when `NemotronH` is torch compiled",
-    "updated_at": "2026-03-20T14:29:16Z"
+    "title": "fix(models): Fix Perceiver interpolate_pos_encoding interpolating to the source size",
+    "updated_at": "2026-03-25T11:54:23Z"
   },
   {
-    "additions": 102,
-    "author": "sergiopaniego",
+    "additions": 14,
+    "author": "yonigozlan",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? `Zamba2MambaMixer.__init__` calls `lazy_load_kernel(\"mamba-ssm\")` and `lazy_load_kernel(\"causal-conv1d\")` unconditionally. Models that inherit from it (like NemotronH) and set `use_mamba_kernels=False` in their conf\u2026",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Add compatibility with remote code importing image_processing_utils_fast modules and methods using `from transformers.image_processing_utils_fast import ...`",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44853",
-    "created_at": "2026-03-19T10:22:40Z",
-    "deletions": 72,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44897",
+    "created_at": "2026-03-20T19:30:32Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44853/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44853",
+    "files_url": "https://github.com/huggingface/transformers/pull/44897/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44897",
     "labels": [],
-    "merged": false,
-    "number": 44853,
+    "merged": true,
+    "number": 44897,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix Zamba2MambaMixer ignoring use_mamba_kernels=False",
-    "updated_at": "2026-04-10T12:18:45Z"
+    "state": "closed",
+    "title": "Add backward compatibility for direct imports from legacy `image_processing_utils_fast`",
+    "updated_at": "2026-03-20T20:00:12Z"
   },
   {
-    "additions": 117,
-    "author": "zucchini-nlp",
+    "additions": 354,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "An eos token can also be a list on most recent models, so this PR allows all `EOS` in config be a list as well. Same for q-lora-rank which apparently can be an explicit `None` for some model Also bring back `layer_type_validation` and add\u2026",
-    "changed_files": 92,
+    "body_excerpt": "updates the continuous batching docs - new page for the API reference - adds sections for new features like CUDA graphs, async batching, prefix caching, logprobs (depending on when its merged) - clearer example of generation with varying l\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44851",
-    "created_at": "2026-03-19T09:53:31Z",
-    "deletions": 101,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44896",
+    "created_at": "2026-03-20T19:09:41Z",
+    "deletions": 81,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44851/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44851",
+    "files_url": "https://github.com/huggingface/transformers/pull/44896/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44896",
     "labels": [],
     "merged": true,
-    "number": 44851,
+    "number": 44896,
     "review_comments_count": 2,
     "state": "closed",
-    "title": "Update some type hints",
-    "updated_at": "2026-03-19T16:30:32Z"
+    "title": "[docs] continuous batching",
+    "updated_at": "2026-03-30T17:17:13Z"
   },
   {
-    "additions": 5,
-    "author": "vasqu",
+    "additions": 57,
+    "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "See #44458 This is a deep issue tbh - the cross attentions are reshaped into a different shape than the text input leading to a mismatch between batch sizes. This only gets noticed during compile as it is more strict about the concrete sha\u2026",
+    "body_excerpt": "# What does this PR do? This PR enables static FP8 experts. This also works on multi-gpu with device-map. A fix for that was to set was to set `torch.cuda.set_device()`. Triton's JIT compiler uses he active device context to determine whic\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44850",
-    "created_at": "2026-03-19T08:36:18Z",
-    "deletions": 6,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44850/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44850",
-    "labels": [],
-    "merged": false,
-    "number": 44850,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[`Mllama`] Fix workaround compile",
-    "updated_at": "2026-03-26T13:02:24Z"
-  },
-  {
-    "additions": 58,
-    "author": "Sai-Suraj-27",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Update Qwen3_Omni_Moe, to fix these attribute errors [Qwen3OmniModelIntegrationTests](https://github.com/huggingface/transformers/actions/runs/23230643883/job/67524756897#step:14:1131) <img width=\"2292\" height=\"161\"\u2026",
-    "changed_files": 5,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 46,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44848",
-    "created_at": "2026-03-19T07:30:39Z",
-    "deletions": 38,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44895",
+    "created_at": "2026-03-20T19:01:35Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44848/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44848",
+    "files_url": "https://github.com/huggingface/transformers/pull/44895/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44895",
     "labels": [],
     "merged": true,
-    "number": 44848,
-    "review_comments_count": 5,
+    "number": 44895,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Fix few issues in Qwen_3_Omni_Moe",
-    "updated_at": "2026-03-30T16:43:00Z"
+    "title": "Add static FP8 expert support ",
+    "updated_at": "2026-03-24T14:27:31Z"
   },
   {
-    "additions": 68,
-    "author": "tarekziade",
+    "additions": 10,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Activated `anti-slop` action. Enabled checks: - `min-account-age: 30` to catch brand-new throwaway accounts, which are common in automated spam waves. - `max-daily-forks: 7` to catch accounts that fork many reposito\u2026",
+    "body_excerpt": "## Problem `ProcessorMixin.to_dict()` was calling `copy.deepcopy(self.__dict__)` on the entire processor, including the tokenizer, even though the tokenizer is always deleted from the output immediately after (since tokenizers are saved se\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44847",
-    "created_at": "2026-03-19T07:15:38Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44894",
+    "created_at": "2026-03-20T18:57:53Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44847/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44847",
+    "files_url": "https://github.com/huggingface/transformers/pull/44894/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44894",
     "labels": [],
     "merged": true,
-    "number": 44847,
-    "review_comments_count": 3,
+    "number": 44894,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "ci: add anti-slop action",
-    "updated_at": "2026-03-26T13:39:58Z"
+    "title": "fix `processing_utils.py`: avoid deepcopying tokenizer in `ProcessorMixin` to improve performance",
+    "updated_at": "2026-03-23T10:09:02Z"
   },
   {
-    "additions": 64,
-    "author": "RicardoLee510520",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "# What does this PR do? Updated the DeiT model card to follow the new standardized format: - Replaced verbose paper abstract with concise model description - Added Pipeline and AutoModel usage examples - Renamed \"Usage tips\" to \"Notes\" - U\u2026",
+    "additions": 18,
+    "author": "ai-man-codes",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #43011 `StaticLayer` was missing a `.crop(max_length)` method, so implemented that according to the discussion of the issue. Added `StaticLayer.crop(max_length)` to match the API of StaticCache with the Dynami\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44846",
-    "created_at": "2026-03-19T06:30:53Z",
-    "deletions": 90,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44846/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44846",
-    "labels": [],
-    "merged": false,
-    "number": 44846,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[Docs] Update DeiT model card to new format",
-    "updated_at": "2026-03-20T05:30:17Z"
-  },
-  {
-    "additions": 15,
-    "author": "jiqing-feng",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes `torch.compile` failure for Mllama after #42848 introduced a new unified attention mask creation path. The root cause is a **torch inductor C++ codegen bug**: when `padding_mask_function` uses advanced tensor\u2026",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44845",
-    "created_at": "2026-03-19T06:14:54Z",
-    "deletions": 5,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44845/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44845",
-    "labels": [],
-    "merged": false,
-    "number": 44845,
-    "review_comments_count": 1,
-    "state": "open",
-    "title": "Fix Mllama torch.compile failure caused by new attention mask logic",
-    "updated_at": "2026-03-26T13:01:50Z"
-  },
-  {
-    "additions": 8468,
-    "author": "sahilleth",
-    "author_association": "NONE",
-    "body_excerpt": "This PR makes a few small fixes on top of #37875 for the DEIM model: - Ensure `DeimConfig` / `DEIMConfig` and `DeimModel` / `DeimForObjectDetection` are correctly exposed from the `transformers` package. - Fix a configuration docstring lin\u2026",
-    "changed_files": 15,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44844",
-    "created_at": "2026-03-19T05:50:29Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44893",
+    "created_at": "2026-03-20T17:48:23Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44844/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44844",
+    "files_url": "https://github.com/huggingface/transformers/pull/44893/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44893",
     "labels": [],
     "merged": false,
-    "number": 44844,
+    "number": 44893,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix DEIM config export and public API",
-    "updated_at": "2026-03-19T13:18:59Z"
+    "state": "open",
+    "title": "add `StaticLayer.crop()` to match `DynamicLayer` API",
+    "updated_at": "2026-03-28T01:41:22Z"
   },
   {
-    "additions": 26,
-    "author": "omyaaa1",
-    "author_association": "NONE",
-    "body_excerpt": "Reintroduce handling for remote URLs using download_url, which was accidentally removed in recent versions. This restores support for loading image processor configs directly from URLs. Fixes #44821 # What does this PR do? <!-- Congratulat\u2026",
-    "changed_files": 1,
+    "additions": 51,
+    "author": "he-yufeng",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #44821 The `elif is_remote_url(...)` / `download_url(...)` branch in `get_image_processor_dict` was accidentally removed during the image processor refactor in #43514. This caused `AutoImageProcessor.from_pretrained(url)` to break wi\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44842",
-    "created_at": "2026-03-19T04:48:58Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44892",
+    "created_at": "2026-03-20T16:21:25Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44842/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44842",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44892/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44892",
+    "labels": [],
     "merged": false,
-    "number": 44842,
-    "review_comments_count": 0,
+    "number": 44892,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Fix AutoImageProcessor URL loading regression",
-    "updated_at": "2026-03-19T11:57:50Z"
+    "title": "Fix AutoImageProcessor.from_pretrained failing on URL input",
+    "updated_at": "2026-03-24T13:30:38Z"
   },
   {
-    "additions": 1,
-    "author": "zhulinchng",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fix formatting of code block in weightconverter.md # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set\u2026",
-    "changed_files": 1,
+    "additions": 507,
+    "author": "kashif",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Add a MoERouterHealthCallback to log MoE router-health metrics. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44839",
-    "created_at": "2026-03-19T01:40:58Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44891",
+    "created_at": "2026-03-20T16:17:05Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44839/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44839",
+    "files_url": "https://github.com/huggingface/transformers/pull/44891/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44891",
     "labels": [],
-    "merged": true,
-    "number": 44839,
+    "merged": false,
+    "number": 44891,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Correct code block formatting in weightconverter.md",
-    "updated_at": "2026-03-19T07:07:13Z"
+    "state": "open",
+    "title": "[Trainer] add MoERouterHealthCallback Callback",
+    "updated_at": "2026-03-20T16:28:43Z"
   },
   {
-    "additions": 41,
-    "author": "xr843",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary - Fixes #44821 \u2014 `AutoImageProcessor.from_pretrained` fails with `OSError: Repo id must be in the form 'repo_name' or 'namespace/repo_name'` when given a URL - The URL handling branch (`is_remote_url` check) in `get_image_proces\u2026",
+    "additions": 72,
+    "author": "Rocketknight1",
+    "author_association": "MEMBER",
+    "body_excerpt": "As discussed on Slack, this is the first phase of our approach to controlling the code agent epidemic. This PR places large warnings in both the pull request template and `CONTRIBUTING.md`, which should hopefully be seen by most contributo\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44838",
-    "created_at": "2026-03-18T23:53:54Z",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44890",
+    "created_at": "2026-03-20T16:12:45Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44838/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44838",
+    "files_url": "https://github.com/huggingface/transformers/pull/44890/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44890",
     "labels": [],
-    "merged": false,
-    "number": 44838,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44890,
+    "review_comments_count": 7,
     "state": "closed",
-    "title": "Fix AutoImageProcessor.from_pretrained failing with URL input",
-    "updated_at": "2026-03-19T10:43:53Z"
+    "title": "Add big angry code agent warnings!",
+    "updated_at": "2026-03-23T11:54:48Z"
   },
   {
-    "additions": 482,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "backfills empty model cards like gptoss and nemotronh with a `model-card.md` skill i created. its pretty minimal at the moment and just includes a brief intro and code examples. let me know if there is anything else we should add!",
-    "changed_files": 12,
+    "additions": 86,
+    "author": "roycho96",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Calling `trainer.evaluate()` before `trainer.train()` with DeepSpeed is broken in three ways: 1. **ZeRO-3 stale state crash:** `evaluate()` creates an inference engine. `train()` starts with `accelerator.free_memor\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44837",
-    "created_at": "2026-03-18T21:45:31Z",
-    "deletions": 102,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44889",
+    "created_at": "2026-03-20T15:08:32Z",
+    "deletions": 21,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44837/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44837",
+    "files_url": "https://github.com/huggingface/transformers/pull/44889/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44889",
     "labels": [],
-    "merged": true,
-    "number": 44837,
+    "merged": false,
+    "number": 44889,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[docs] model cards",
-    "updated_at": "2026-03-20T22:40:41Z"
+    "state": "open",
+    "title": "[DeepSpeed] Fix evaluate()/predict() before train()",
+    "updated_at": "2026-03-21T11:06:07Z"
   },
   {
-    "additions": 96,
-    "author": "tyler-romero",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Pass cu_seqlens derived from packed attention masks to FLA's ShortConvolution and chunk_gated_delta_rule kernels, preventing recurrent state from leaking across sequence boundaries during packed-sequence training. F\u2026",
+    "additions": 2,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? In general, it should be much better to let the kernel do what it wants for perfs! There's no reasons to have troubles from it!",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44836",
-    "created_at": "2026-03-18T20:24:58Z",
-    "deletions": 20,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44836/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44836",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44888",
+    "created_at": "2026-03-20T14:45:28Z",
+    "deletions": 22,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44888/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44888",
     "labels": [],
     "merged": false,
-    "number": 44836,
-    "review_comments_count": 3,
+    "number": 44888,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Add cu_seqlens support to OlmoHybridGatedDeltaNet for packed sequences",
-    "updated_at": "2026-03-19T05:34:43Z"
+    "title": "Remove explicit cuda stream in nemotron_h",
+    "updated_at": "2026-03-26T19:15:29Z"
   },
   {
-    "additions": 187,
-    "author": "remi-or",
+    "additions": 2,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "## Summary This PR adds the `return_logprobs` flag to the continuous batching, enabling the user to retrieve the log probabilites of the tokens generated. # Tests Added a test to compare with regular generate and it passes. All tests pass.\u2026",
-    "changed_files": 6,
+    "body_excerpt": "# What does this PR do? As per the title. On currently pinned version, when we run this small snippet (which is called on some model's `__init__` functions \ud83d\ude05): ```python from transformers.integrations.hub_kernels import lazy_load_kernel ca\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44835",
-    "created_at": "2026-03-18T17:48:15Z",
-    "deletions": 83,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44887",
+    "created_at": "2026-03-20T14:00:33Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44835/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44835",
+    "files_url": "https://github.com/huggingface/transformers/pull/44887/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44887",
     "labels": [],
     "merged": true,
-    "number": 44835,
-    "review_comments_count": 0,
+    "number": 44887,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "[CB] Add an option to return logprobs",
-    "updated_at": "2026-03-23T18:35:31Z"
+    "title": "Bump kernels version dependency to avoid crashes",
+    "updated_at": "2026-03-20T19:01:51Z"
   },
   {
-    "additions": 0,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Oups this slipped through in https://github.com/huggingface/transformers/pull/44833",
-    "changed_files": 7,
+    "additions": 14,
+    "author": "m-matthias",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Prevent crash in class LwDetrImageLoss when using it with float16 automatic mixed precision on a Cuda device. torch.pow causes an autocast to float32 when used with Cuda, which caused a type mismatch at ``` pos_weig\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44834",
-    "created_at": "2026-03-18T17:07:11Z",
-    "deletions": 11,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44886",
+    "created_at": "2026-03-20T13:56:08Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44834/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44834",
+    "files_url": "https://github.com/huggingface/transformers/pull/44886/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44886",
     "labels": [],
     "merged": true,
-    "number": 44834,
+    "number": 44886,
+    "review_comments_count": 4,
+    "state": "closed",
+    "title": "LwDetrImageLoss: Fix dtype casting to prevent crash when using amp on cuda device",
+    "updated_at": "2026-03-24T17:02:32Z"
+  },
+  {
+    "additions": 2,
+    "author": "guoyangzhen",
+    "author_association": "NONE",
+    "body_excerpt": "## Problem In _split_tokens_on_unicode(), when the decoded token stream ends with a dangling Unicode replacement character (U+FFFD), the computed index can equal len(decoded_full), causing IndexError: string index out of range. The failing\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44885",
+    "created_at": "2026-03-20T13:03:54Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44885/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44885",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44885,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Update more modular examples",
-    "updated_at": "2026-03-18T17:18:54Z"
+    "title": "fix: prevent IndexError in Whisper word timestamp decode",
+    "updated_at": "2026-03-23T12:01:50Z"
   },
   {
-    "additions": 299,
-    "author": "Cyrilvallez",
+    "additions": 14,
+    "author": "hmellor",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix the examples after the config change (https://github.com/huggingface/transformers/pull/41250), and re-run conversion as in general modelings changed quite a bit in the lib.",
-    "changed_files": 19,
+    "body_excerpt": "Some libraries that use Transformers (i.e. vLLM) use `|` on the `size` config. This PR adds `__or__` and `__ror__` so that the following works: ```console $ {\"longest_edge\": 20} | SizeDict(height=10, width=20) {'longest_edge': 20, 'height'\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44833",
-    "created_at": "2026-03-18T16:35:34Z",
-    "deletions": 590,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44884",
+    "created_at": "2026-03-20T11:52:15Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44833/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44833",
+    "files_url": "https://github.com/huggingface/transformers/pull/44884/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44884",
     "labels": [],
     "merged": true,
-    "number": 44833,
-    "review_comments_count": 0,
+    "number": 44884,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix and re-run modular converter on examples",
-    "updated_at": "2026-03-18T17:00:44Z"
+    "title": "Add missing dunder methods to `SizeDict`",
+    "updated_at": "2026-03-20T12:21:12Z"
   },
   {
-    "additions": 346,
-    "author": "IlyasMoutawwakil",
+    "additions": 2,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 6,
+    "body_excerpt": "# What does this PR do? Fix https://github.com/huggingface/transformers/issues/44589.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44832",
-    "created_at": "2026-03-18T15:33:15Z",
-    "deletions": 176,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44883",
+    "created_at": "2026-03-20T11:43:13Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44832/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44832",
+    "files_url": "https://github.com/huggingface/transformers/pull/44883/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44883",
     "labels": [],
     "merged": true,
-    "number": 44832,
-    "review_comments_count": 37,
+    "number": 44883,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "DeepGEMM",
-    "updated_at": "2026-03-31T15:04:06Z"
+    "title": "Fix dtype guessing from state dict",
+    "updated_at": "2026-03-20T13:12:34Z"
   },
   {
-    "additions": 2,
-    "author": "zucchini-nlp",
+    "additions": 1,
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix loading in Sam3 which currently doesn't match the state dict keys from checkpoint. Adding a correct base model prefix will add it to all state dict keys, making the ckpt load-able <!-- Congratulations! You've ma\u2026",
-    "changed_files": 2,
+    "body_excerpt": "fixes ```python model = \"meta-llama/Llama-4-Maverick-17B-128E-Instruct\" tok_auto = AutoTokenizer.from_pretrained(model) print(f\"AutoTokenizer: {tok_auto('hello')}\") ``` ``` The above exception was the direct cause of the following exceptio\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44831",
-    "created_at": "2026-03-18T14:50:48Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44882",
+    "created_at": "2026-03-20T11:31:20Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44831/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44831",
+    "files_url": "https://github.com/huggingface/transformers/pull/44882/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44882",
     "labels": [],
-    "merged": true,
-    "number": 44831,
+    "merged": false,
+    "number": 44882,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix loading issue in Sam3",
-    "updated_at": "2026-03-18T15:44:01Z"
+    "title": "fix config type",
+    "updated_at": "2026-03-20T16:34:20Z"
   },
   {
-    "additions": 2103,
-    "author": "lashahub",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR adds `AudioFlamingoNext` as a separate model name that inherits directly from `MusicFlamingo` #43538 and keeps the same architecture and behavior. Changes: - add `audioflamingonext` model files - register it in the auto mappings -\u2026",
-    "changed_files": 27,
+    "additions": 142,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per title, we don't need a weird way to filter out kwargs anymore because now we don't rely on `tokenizer.apply_chat_template`. I didn't delete the unused `TypedDict` yet and will deprecate for at least 3 minor r\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44830",
-    "created_at": "2026-03-18T14:31:45Z",
-    "deletions": 48,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44881",
+    "created_at": "2026-03-20T10:44:06Z",
+    "deletions": 82,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44830/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44830",
+    "files_url": "https://github.com/huggingface/transformers/pull/44881/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44881",
     "labels": [],
-    "merged": false,
-    "number": 44830,
-    "review_comments_count": 2,
-    "state": "open",
-    "title": "Add AudioFlamingoNext model",
-    "updated_at": "2026-04-15T21:31:08Z"
+    "merged": true,
+    "number": 44881,
+    "review_comments_count": 12,
+    "state": "closed",
+    "title": "Allow arbitrary template kwargs in processors",
+    "updated_at": "2026-03-27T11:07:08Z"
   },
   {
-    "additions": 101,
-    "author": "Cyrilvallez",
+    "additions": 34,
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Last batch finally! Follow up of https://github.com/huggingface/transformers/pull/44759 and many other",
-    "changed_files": 33,
+    "body_excerpt": "incorrect model list update",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44828",
-    "created_at": "2026-03-18T13:52:32Z",
-    "deletions": 512,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44880",
+    "created_at": "2026-03-20T10:37:13Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44828/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44828",
+    "files_url": "https://github.com/huggingface/transformers/pull/44880/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44880",
     "labels": [],
     "merged": true,
-    "number": 44828,
+    "number": 44880,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Remove cache_position in more models (4 and last one)",
-    "updated_at": "2026-03-18T16:02:46Z"
+    "title": "incorrect model list update",
+    "updated_at": "2026-03-24T09:27:24Z"
   },
   {
-    "additions": 80,
-    "author": "3outeille",
+    "additions": 448,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "https://github.com/huggingface/transformers/pull/44825",
-    "changed_files": 6,
+    "body_excerpt": "# What does this PR do? - uses the Makefile as a single source of truth for running QA checks - adds `tomli` so `make` commands can read the `toml` file when needed - adds a `checkers` Python module that wraps and orchestrates all `checks`\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44827",
-    "created_at": "2026-03-18T13:36:53Z",
-    "deletions": 14,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44879",
+    "created_at": "2026-03-20T10:24:29Z",
+    "deletions": 90,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44827/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44827",
+    "files_url": "https://github.com/huggingface/transformers/pull/44879/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44879",
     "labels": [],
-    "merged": false,
-    "number": 44827,
+    "merged": true,
+    "number": 44879,
     "review_comments_count": 6,
-    "state": "open",
-    "title": "Fix Mistral4 tests",
-    "updated_at": "2026-03-26T16:04:20Z"
+    "state": "closed",
+    "title": "refactor: unify QA calls",
+    "updated_at": "2026-03-25T08:51:30Z"
   },
   {
-    "additions": 55,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #44821 This PR fixes the issue where `AutoImageProcessor.from_pretrained()` was unable to load from a URL (e.g., `https://huggingface.co/.../raw/main/config.json`). The bug was introduced in transformers>=5.3.0. Prior versions (e.g.,\u2026",
-    "changed_files": 2,
+    "additions": 8,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? `check_docstrings` has been complaining for a while about those.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44826",
-    "created_at": "2026-03-18T12:08:35Z",
-    "deletions": 5,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44878",
+    "created_at": "2026-03-20T10:01:08Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44826/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44826",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44826,
+    "files_url": "https://github.com/huggingface/transformers/pull/44878/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44878",
+    "labels": [],
+    "merged": true,
+    "number": 44878,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: allow AutoImageProcessor to load from URL",
-    "updated_at": "2026-03-19T13:28:04Z"
+    "title": "Fix nemotron config docstrings",
+    "updated_at": "2026-03-20T10:11:04Z"
   },
   {
     "additions": 1,
     "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. cc @3outeille as I know you're looking into it",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do?",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44825",
-    "created_at": "2026-03-18T12:05:35Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44876",
+    "created_at": "2026-03-20T09:49:54Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44825/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44825",
+    "files_url": "https://github.com/huggingface/transformers/pull/44876/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44876",
     "labels": [],
     "merged": true,
-    "number": 44825,
+    "number": 44876,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[CI] Temporarily skip Mistral4 tests as they almost all fail",
-    "updated_at": "2026-03-18T12:15:34Z"
+    "title": "Fix nemotron_h modular",
+    "updated_at": "2026-03-20T10:00:35Z"
   },
   {
-    "additions": 5,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Description Fixes #44737 The `relative_positional_encoding` function in XLNet was computing all positional encodings on CPU every forward pass because the `torch.arange` calls were missing the `device` parameter. ## Changes - Added devi\u2026",
+    "additions": 872,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Refactors `src/transformers/cli/serve.py` to reduce nesting depth, eliminate code duplication, and improve maintainability. No behavioral changes and the public API is unchanged. Also added a module docstring to exp\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44824",
-    "created_at": "2026-03-18T11:55:01Z",
-    "deletions": 4,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44824/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44824",
-    "labels": [
-      "Code agent slop"
-    ],
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44875",
+    "created_at": "2026-03-20T09:06:34Z",
+    "deletions": 701,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44875/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44875",
+    "labels": [],
     "merged": false,
-    "number": 44824,
+    "number": 44875,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(xlnet): add device parameter to relative_positional_encoding",
-    "updated_at": "2026-03-18T13:17:38Z"
+    "state": "open",
+    "title": "refactor: improved the cli server module code organization",
+    "updated_at": "2026-03-23T08:08:17Z"
   },
   {
-    "additions": 41,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "This PR fixes #44821 where `AutoImageProcessor.from_pretrained()` couldn't load from a direct URL to a config file. ## Problem When passing a URL like `https://huggingface.co/jinfengxie/BFMS_1014/raw/main/config.json` to `AutoImageProcesso\u2026",
+    "additions": 2,
+    "author": "hmellor",
+    "author_association": "MEMBER",
+    "body_excerpt": "`Llama4`'s was incorrect and causing `StrictDataclassFieldValidationErrors`. `AFMoE`'s was was fine but now it's more specific.",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44823",
-    "created_at": "2026-03-18T11:54:24Z",
-    "deletions": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44874",
+    "created_at": "2026-03-20T09:05:02Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44823/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44823",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44823,
+    "files_url": "https://github.com/huggingface/transformers/pull/44874/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44874",
+    "labels": [],
+    "merged": true,
+    "number": 44874,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: AutoImageProcessor from URL loading",
-    "updated_at": "2026-03-18T13:17:48Z"
+    "title": "Fix `layer_types` type hint for `AFMoE` and `Llama4`",
+    "updated_at": "2026-03-20T12:03:58Z"
   },
   {
-    "additions": 4,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #44737 The relative_positional_encoding method in XLNetModel was creating tensors using torch.arange() without specifying device=, causing the entire sinusoidal positional encoding computation to run on CPU every forward pass. Only t\u2026",
-    "changed_files": 1,
+    "additions": 75,
+    "author": "sergiopaniego",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? ## Problem Online RL training (GRPO, RLOO, PPO) with all VL models using MRoPE with rope_deltas (Qwen2-VL, Qwen2.5-VL, Qwen3-VL, Qwen3.5, GLM4V, PaddleOCR-VL, Ernie4.5-VL-MoE, etc.) crashes with `RuntimeError: Sizes\u2026",
+    "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44822",
-    "created_at": "2026-03-18T11:48:28Z",
-    "deletions": 4,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44873",
+    "created_at": "2026-03-20T08:38:03Z",
+    "deletions": 30,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44822/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44822",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44822,
+    "files_url": "https://github.com/huggingface/transformers/pull/44873/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44873",
+    "labels": [],
+    "merged": true,
+    "number": 44873,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: XLNet relative_positional_encoding device placement",
-    "updated_at": "2026-03-18T13:17:30Z"
+    "title": "Fix VL model rope_deltas batch size mismatch in online RL training",
+    "updated_at": "2026-03-20T13:51:08Z"
   },
   {
-    "additions": 14,
-    "author": "BillionClaw",
+    "additions": 2,
+    "author": "IvanFan-Van",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #44811 ## Problem When calling `processor.batch_decode(predicted_ids, skip_special_tokens=False)` with the output from `model.generate()` (without `return_dict_in_generate=True`), the `skip_special_tokens` parameter was being ignored\u2026",
+    "body_excerpt": "## Description Update outdated comment that references non-existent file `generation_utils_samplers.py` ## Changes Detail - The comment on line 1200 states \"all samplers can be found in `generation_utils_samplers.py`\" - In reality, all sam\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44820",
-    "created_at": "2026-03-18T10:57:12Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44872",
+    "created_at": "2026-03-20T05:45:46Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44820/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44820",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44872/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44872",
+    "labels": [],
     "merged": false,
-    "number": 44820,
+    "number": 44872,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(whisper): respect skip_special_tokens in batch_decode",
-    "updated_at": "2026-03-18T13:17:20Z"
+    "state": "open",
+    "title": "Fix: Update outdated sampler comment in generation/utils.py",
+    "updated_at": "2026-03-20T05:45:46Z"
   },
   {
-    "additions": 2,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Description This PR fixes the DeepSeek tokenizer issue (#44779) where tokenization produces incorrect results in v5. ### Problem In transformers v5, the DeepSeek tokenizer (DeepSeek-R1) was producing incorrect results: - Input: \"How are\u2026",
-    "changed_files": 1,
+    "additions": 666,
+    "author": "JonusClapshaw",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Fixes #42200 `prediction_step` is type-hinted to return `Optional[torch.Tensor]` for logits, but when no `preprocess_logits_for_metrics` is provided it could return a tuple instead of a tensor. This caused `torch_pa\u2026",
+    "changed_files": 33,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44819",
-    "created_at": "2026-03-18T10:55:22Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44870",
+    "created_at": "2026-03-20T02:28:27Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44819/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44819",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44870/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44870",
+    "labels": [],
     "merged": false,
-    "number": 44819,
+    "number": 44870,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(tokenizer): add deepseek_v2 and deepseek_v3 to incorrect hub tokenizer class list",
-    "updated_at": "2026-03-18T14:11:16Z"
+    "title": "fix: ensure prediction_step returns tensor for logits, not tuple #42200",
+    "updated_at": "2026-03-20T17:51:19Z"
   },
   {
-    "additions": 64,
-    "author": "BillionClaw",
+    "additions": 98,
+    "author": "sdharani91",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Description Fixes #44805 When training multimodal models (Qwen3-VL, GLM-4.6V, Qwen3-VL-MoE) with LoRA adapters, the `attention_mask` and `mm_token_type_ids` tensors can have different shapes. This causes an IndexError when the `get_rope\u2026",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? Fixes #44717 This PR fixes packed-sequence handling for the Qwen3.5 linear-attention fast path. Before this change, Qwen3.5 produced different outputs for: a padded representation of multiple sequences a packed repr\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44818",
-    "created_at": "2026-03-18T10:46:22Z",
-    "deletions": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44867",
+    "created_at": "2026-03-19T17:31:45Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44818/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44818",
+    "files_url": "https://github.com/huggingface/transformers/pull/44867/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44867",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44818,
+    "number": 44867,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: resolve mask shape mismatch IndexError in multimodal VL models",
-    "updated_at": "2026-03-18T10:51:43Z"
+    "title": "Pass packed boundary metadata to Qwen3.5 linear-attention fast kernels",
+    "updated_at": "2026-03-26T21:02:58Z"
   },
   {
-    "additions": 28,
-    "author": "JJJYmmm",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/QwenLM/Qwen3.5/issues/97. This PR adds `enable_thinking` to the chat-template kwargs. With this change, `enable_thinking` is treated as a template-level argument in the tokenize=True path, so\u2026",
-    "changed_files": 2,
+    "additions": 78,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44817",
-    "created_at": "2026-03-18T10:44:11Z",
-    "deletions": 6,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44866",
+    "created_at": "2026-03-19T17:27:58Z",
+    "deletions": 75,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44817/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44817",
+    "files_url": "https://github.com/huggingface/transformers/pull/44866/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44866",
     "labels": [],
-    "merged": false,
-    "number": 44817,
+    "merged": true,
+    "number": 44866,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Misc] add enable_thinking to template kwargs",
-    "updated_at": "2026-03-20T14:56:04Z"
+    "title": "Align lfm2 cache to other mamba caches",
+    "updated_at": "2026-03-20T10:50:28Z"
   },
   {
-    "additions": 98,
-    "author": "Cyrilvallez",
+    "additions": 496,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. This is the last of many PR to remove the `cache_position`. At this point, all the models were already updated to not use them, and they are fully ignored in all the modelings. So this removes thei\u2026",
-    "changed_files": 57,
+    "body_excerpt": "# What does this PR do? Added Rule 11 forward() must not access non-nn.Module attributes on submodules (breaks pipeline parallelism with Identity replacement). we want to make sure we just use metadata in config and elesewere when in that\u2026",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44816",
-    "created_at": "2026-03-18T10:32:04Z",
-    "deletions": 375,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44865",
+    "created_at": "2026-03-19T16:39:59Z",
+    "deletions": 26,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44816/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44816",
+    "files_url": "https://github.com/huggingface/transformers/pull/44865/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44865",
     "labels": [],
     "merged": true,
-    "number": 44816,
-    "review_comments_count": 14,
+    "number": 44865,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "[generate] Never use `cache_position` anymore in generation",
-    "updated_at": "2026-03-19T14:18:28Z"
+    "title": "chore(typing): added rule 11",
+    "updated_at": "2026-03-23T12:29:21Z"
   },
   {
-    "additions": 135,
-    "author": "ArthurZucker",
+    "additions": 99,
+    "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 6,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44815",
-    "created_at": "2026-03-18T09:54:18Z",
-    "deletions": 23,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44815/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44815",
-    "labels": [],
-    "merged": false,
-    "number": 44815,
-    "review_comments_count": 2,
-    "state": "open",
-    "title": "Dequant fix",
-    "updated_at": "2026-03-24T14:39:52Z"
-  },
-  {
-    "additions": 2,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #44792 This PR fixes the failing test `test_model_generate_images` for the Janus model. ## Problem When generating images with the Janus model, `generation_config.num_return_sequences` and `generation_config.max_length` can be `None`\u2026",
+    "body_excerpt": "# What does this PR do? This PR switches FP8 per-tensor implementation to rely on the official torch impl `torch._scaled_mm`. Note that `torch._scaled_mm` don't explicitly support per tensor. We hack the api a bit as it only support per ro\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44814",
-    "created_at": "2026-03-18T09:51:34Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44864",
+    "created_at": "2026-03-19T16:19:53Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44814/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44814",
+    "files_url": "https://github.com/huggingface/transformers/pull/44864/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44864",
     "labels": [],
     "merged": false,
-    "number": 44814,
+    "number": 44864,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(janus): handle None values in image generation mode",
-    "updated_at": "2026-03-18T10:42:50Z"
+    "title": "Switch FP8 per tensor quant to use `torch._scaled_mm`",
+    "updated_at": "2026-03-20T19:05:05Z"
   },
   {
-    "additions": 20,
-    "author": "hf-security-analysis[bot]",
+    "additions": 19,
+    "author": "gh-wf",
     "author_association": "NONE",
-    "body_excerpt": "Update `.github/workflows/pr-repo-consistency-bot.yml` workflow configuration. cc @ydshieh Closes huggingface/tracking-issues#26",
-    "changed_files": 1,
+    "body_excerpt": "Some models (e.g. Nemotron-H) define `_tied_weights_keys` as a list, which caused `AttributeError: 'list' object has no attribute 'keys'` when calling `save_pretrained` during full finetuning. # What does this PR do? `_get_tied_weight_keys\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44813",
-    "created_at": "2026-03-18T09:49:16Z",
-    "deletions": 46,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44862",
+    "created_at": "2026-03-19T15:14:12Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44813/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44813",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44862/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44862",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44813,
+    "number": 44862,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "chore: update pr-repo-consistency-bot.yml",
-    "updated_at": "2026-03-18T10:08:04Z"
+    "title": "fix: handle list-type _tied_weights_keys in _get_tied_weight_keys",
+    "updated_at": "2026-03-20T09:47:09Z"
   },
   {
-    "additions": 7,
-    "author": "ydshieh",
+    "additions": 11,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Some checks (for example, modular checks) really require the installation from PR branch.",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? As per the title. As discussed offline with the Mistral team, the scaling applied to the query should not be the absolute one (old `cache_position`), but the actual `position_ids`, taking into account padding, packe\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44812",
-    "created_at": "2026-03-18T09:40:36Z",
-    "deletions": 0,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44860",
+    "created_at": "2026-03-19T14:27:33Z",
+    "deletions": 17,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44812/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44812",
+    "files_url": "https://github.com/huggingface/transformers/pull/44860/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44860",
     "labels": [],
     "merged": true,
-    "number": 44812,
+    "number": 44860,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix repo-check bot",
-    "updated_at": "2026-03-18T09:49:50Z"
+    "title": "[Mistral] Fix query scaling for Mistral4 and Ministral3",
+    "updated_at": "2026-03-19T18:02:06Z"
   },
   {
-    "additions": 7,
-    "author": "LincolnBurrows2017",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixed bare except clause in _safe_convert_tensor function to catch only Exception type.",
-    "changed_files": 2,
+    "additions": 7001,
+    "author": "philippguevorguian",
+    "author_association": "NONE",
+    "body_excerpt": null,
+    "changed_files": 19,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44809",
-    "created_at": "2026-03-18T05:47:03Z",
-    "deletions": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44859",
+    "created_at": "2026-03-19T13:54:19Z",
+    "deletions": 138,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44809/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44809",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44859/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44859",
+    "labels": [],
     "merged": false,
-    "number": 44809,
+    "number": 44859,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: replace bare except with Exception in Fuyu image processing",
-    "updated_at": "2026-03-18T13:17:11Z"
+    "title": "refactor: rope in model, flatten vision, rely on qwen3 backone, misc changes",
+    "updated_at": "2026-03-19T14:08:01Z"
   },
   {
-    "additions": 2759,
-    "author": "zhang-prog",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": null,
-    "changed_files": 31,
+    "additions": 111,
+    "author": "remi-or",
+    "author_association": "MEMBER",
+    "body_excerpt": "Right now, the continuous batching tests all use similar mechanisms, namely: 1. loading a model and a tokenizer 2. preparing data for generate or generate_batch 3. running generate to compare its outputs with generate_batch This PR adds 3\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 12,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44808",
-    "created_at": "2026-03-18T04:29:07Z",
-    "deletions": 21,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44858",
+    "created_at": "2026-03-19T13:22:04Z",
+    "deletions": 188,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44808/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44808",
-    "labels": [
-      "New model"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44858/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44858",
+    "labels": [],
     "merged": true,
-    "number": 44808,
-    "review_comments_count": 53,
-    "state": "closed",
-    "title": "[Model] Add PP-OCRv5_server_rec and  PP-OCRv5_mobile_rec models Support",
-    "updated_at": "2026-03-18T20:24:50Z"
-  },
-  {
-    "additions": 0,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "The question-answering pipeline was removed in v5.3 per the migration guide, but the Chinese, Korean, and French quicktour docs still listed it as an available pipeline task. This removes those outdated references to avoid confusing users\u2026",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44807",
-    "created_at": "2026-03-18T03:41:52Z",
-    "deletions": 3,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44807/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44807",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44807,
+    "number": 44858,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "docs(quicktour): remove question-answering pipeline from quicktour tables",
-    "updated_at": "2026-03-18T15:38:09Z"
+    "title": "[CB] [Minor] Simplify test suite",
+    "updated_at": "2026-03-24T11:44:39Z"
   },
   {
-    "additions": 23,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "This PR fixes OOM errors when loading models on AMD Strix Halo APUs. ## Problem AMD Strix Halo (Radeon 8060S/8050S) uses unified memory architecture where memory-mapped file loading doesn't work well with the current amdgpu driver. This ca\u2026",
+    "additions": 3,
+    "author": "hkc5",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Python 3.13's stricter parser fails when there's a comment between the `@torch.jit.script` decorator and the function definition, causing an IndentationError when importing DebertaV2Model. ## Changes - Moved comments before the\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44806",
-    "created_at": "2026-03-18T03:33:59Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44856",
+    "created_at": "2026-03-19T12:33:00Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44806/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44806",
+    "files_url": "https://github.com/huggingface/transformers/pull/44856/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44856",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44806,
+    "number": 44856,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(core_model_loading): disable mmap on Strix Halo to avoid OOM",
-    "updated_at": "2026-03-18T12:30:21Z"
+    "title": "fix: move comments before @torch.jit.script decorator for Python 3.13 compatibility",
+    "updated_at": "2026-03-19T13:11:44Z"
   },
   {
-    "additions": 137,
-    "author": "stevhliu",
+    "additions": 63,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "updates the peft docs: - a more complete training section with a full code snippet, describe saving behavior, resuming from a checkpoint, and distributed training - adds some undocumented API methods (`delete_adapter`, `active_adapters`) -\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? We had (flaky) ```bash tests/models/nemotron_h/test_modeling_nemotron_h.py::NemotronHModelTest::test_sdpa_can_compile_dynamic Fatal Python error: Segmentation fault ``` `NemotronHBlock.forward` creates a temporary `\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44804",
-    "created_at": "2026-03-18T00:08:54Z",
-    "deletions": 89,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44854",
+    "created_at": "2026-03-19T10:54:36Z",
+    "deletions": 56,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44804/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44804",
+    "files_url": "https://github.com/huggingface/transformers/pull/44854/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44854",
     "labels": [],
     "merged": true,
-    "number": 44804,
-    "review_comments_count": 3,
+    "number": 44854,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "[docs] peft",
-    "updated_at": "2026-03-23T17:14:58Z"
+    "title": "Fix core dumped when `NemotronH` is torch compiled",
+    "updated_at": "2026-03-20T14:29:16Z"
   },
   {
-    "additions": 1341,
-    "author": "yonigozlan",
+    "additions": 102,
+    "author": "sergiopaniego",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? (Finally) add support for checking+fixing both generated files and modular files in `check_auto_docstrings`. Also `auto_docstring` was recently added to configs, and this PR updates `check_auto_docstrings` to suppor\u2026",
-    "changed_files": 244,
+    "body_excerpt": "# What does this PR do? `Zamba2MambaMixer.__init__` calls `lazy_load_kernel(\"mamba-ssm\")` and `lazy_load_kernel(\"causal-conv1d\")` unconditionally. Models that inherit from it (like NemotronH) and set `use_mamba_kernels=False` in their conf\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44803",
-    "created_at": "2026-03-17T22:40:45Z",
-    "deletions": 1105,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44853",
+    "created_at": "2026-03-19T10:22:40Z",
+    "deletions": 72,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44803/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44803",
+    "files_url": "https://github.com/huggingface/transformers/pull/44853/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44853",
     "labels": [],
-    "merged": true,
-    "number": 44803,
-    "review_comments_count": 24,
-    "state": "closed",
-    "title": "Support Modular (!!) + Configs in `check_auto_docstrings`",
-    "updated_at": "2026-03-24T17:59:12Z"
+    "merged": false,
+    "number": 44853,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix Zamba2MambaMixer ignoring use_mamba_kernels=False",
+    "updated_at": "2026-04-10T12:18:45Z"
   },
   {
-    "additions": 12,
-    "author": "itazap",
+    "additions": 117,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "deepseek_v2, deepseek_v3, and modernbert fix for having incorrect tokenizer class on the hub fixes: https://github.com/huggingface/transformers/issues/44779, https://github.com/huggingface/transformers/pull/44783",
-    "changed_files": 2,
+    "body_excerpt": "An eos token can also be a list on most recent models, so this PR allows all `EOS` in config be a list as well. Same for q-lora-rank which apparently can be an explicit `None` for some model Also bring back `layer_type_validation` and add\u2026",
+    "changed_files": 92,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44801",
-    "created_at": "2026-03-17T17:40:25Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44851",
+    "created_at": "2026-03-19T09:53:31Z",
+    "deletions": 101,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44801/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44801",
+    "files_url": "https://github.com/huggingface/transformers/pull/44851/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44851",
     "labels": [],
     "merged": true,
-    "number": 44801,
-    "review_comments_count": 0,
+    "number": 44851,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "deepseek_v2, deepseek_v3, and modernbert fix for having incorrect tokenizer class on the hub",
-    "updated_at": "2026-03-19T13:11:54Z"
+    "title": "Update some type hints",
+    "updated_at": "2026-03-19T16:30:32Z"
   },
   {
-    "additions": 36,
-    "author": "aayushbaluni",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44488 `CamembertTokenizer` raised `ValueError: too many values to unpack (expected 2)` when loading models like `cjvt/sleng-bert` that provide vocab as a dict `{token: id}` from `tokenizer.json` (BPE format). The tokenize\u2026",
-    "changed_files": 2,
+    "additions": 5,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "See #44458 This is a deep issue tbh - the cross attentions are reshaped into a different shape than the text input leading to a mismatch between batch sizes. This only gets noticed during compile as it is more strict about the concrete sha\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44850",
+    "created_at": "2026-03-19T08:36:18Z",
+    "deletions": 6,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44850/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44850",
+    "labels": [],
+    "merged": false,
+    "number": 44850,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[`Mllama`] Fix workaround compile",
+    "updated_at": "2026-03-26T13:02:24Z"
+  },
+  {
+    "additions": 58,
+    "author": "Sai-Suraj-27",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Update Qwen3_Omni_Moe, to fix these attribute errors [Qwen3OmniModelIntegrationTests](https://github.com/huggingface/transformers/actions/runs/23230643883/job/67524756897#step:14:1131) <img width=\"2292\" height=\"161\"\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44800",
-    "created_at": "2026-03-17T17:20:35Z",
-    "deletions": 2,
+    "comments_count": 46,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44848",
+    "created_at": "2026-03-19T07:30:39Z",
+    "deletions": 38,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44800/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44800",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44800,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44848/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44848",
+    "labels": [],
+    "merged": true,
+    "number": 44848,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "fix: handle dict vocab in CamembertTokenizer for tokenizer.json (#44488)",
-    "updated_at": "2026-03-18T15:37:54Z"
+    "title": "Fix few issues in Qwen_3_Omni_Moe",
+    "updated_at": "2026-03-30T16:43:00Z"
   },
   {
-    "additions": 327,
-    "author": "stevhliu",
+    "additions": 68,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "updates the Hardware section of the docs for training: - combined CPU/Distributed CPU into a single doc - add more info to the Gaudi doc (mixed precision, torch.compile, distributed training) - add more info to the MPS doc (mixed precision\u2026",
-    "changed_files": 10,
+    "body_excerpt": "# What does this PR do? Activated `anti-slop` action. Enabled checks: - `min-account-age: 30` to catch brand-new throwaway accounts, which are common in automated spam waves. - `max-daily-forks: 7` to catch accounts that fork many reposito\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44799",
-    "created_at": "2026-03-17T17:19:51Z",
-    "deletions": 574,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44847",
+    "created_at": "2026-03-19T07:15:38Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44799/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44799",
+    "files_url": "https://github.com/huggingface/transformers/pull/44847/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44847",
     "labels": [],
     "merged": true,
-    "number": 44799,
-    "review_comments_count": 24,
+    "number": 44847,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "[docs] training on specific hardware",
-    "updated_at": "2026-04-13T07:12:08Z"
+    "title": "ci: add anti-slop action",
+    "updated_at": "2026-03-26T13:39:58Z"
   },
   {
-    "additions": 17,
-    "author": "divyanks",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "additions": 64,
+    "author": "RicardoLee510520",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "# What does this PR do? Updated the DeiT model card to follow the new standardized format: - Replaced verbose paper abstract with concise model description - Added Pipeline and AutoModel usage examples - Renamed \"Usage tips\" to \"Notes\" - U\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44798",
-    "created_at": "2026-03-17T16:51:46Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44846",
+    "created_at": "2026-03-19T06:30:53Z",
+    "deletions": 90,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44798/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44798",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44846/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44846",
+    "labels": [],
     "merged": false,
-    "number": 44798,
+    "number": 44846,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add MPS (Apple Silicon) example and documentation",
-    "updated_at": "2026-03-18T15:37:09Z"
+    "title": "[Docs] Update DeiT model card to new format",
+    "updated_at": "2026-03-20T05:30:17Z"
   },
   {
-    "additions": 1,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "Depends on #44887 and kernels being version `12.3` Works OOB with little changes! Example script for demonstration: ```python from transformers import AutoModelForCausalLM, AutoTokenizer fa_version = 4 #model_id = \"openai/gpt-oss-20b\" mode\u2026",
+    "additions": 15,
+    "author": "jiqing-feng",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes `torch.compile` failure for Mllama after #42848 introduced a new unified attention mask creation path. The root cause is a **torch inductor C++ codegen bug**: when `padding_mask_function` uses advanced tensor\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44797",
-    "created_at": "2026-03-17T15:35:59Z",
-    "deletions": 4,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44797/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44797",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44845",
+    "created_at": "2026-03-19T06:14:54Z",
+    "deletions": 5,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44845/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44845",
     "labels": [],
-    "merged": true,
-    "number": 44797,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[`FA4`] Add kernels fallback",
-    "updated_at": "2026-03-20T19:03:24Z"
+    "merged": false,
+    "number": 44845,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "Fix Mllama torch.compile failure caused by new attention mask logic",
+    "updated_at": "2026-03-26T13:01:50Z"
   },
   {
-    "additions": 4522,
-    "author": "SunMarc",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR refactors transformers serve so that it is not in a single file. We split it into multiple files with clear responsabilities. There were 2,293 lines initially in the serve.py file. ``` \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2026",
+    "additions": 8468,
+    "author": "sahilleth",
+    "author_association": "NONE",
+    "body_excerpt": "This PR makes a few small fixes on top of #37875 for the DEIM model: - Ensure `DeimConfig` / `DEIMConfig` and `DeimModel` / `DeimForObjectDetection` are correctly exposed from the `transformers` package. - Fix a configuration docstring lin\u2026",
     "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 15,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44796",
-    "created_at": "2026-03-17T13:04:06Z",
-    "deletions": 3100,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44844",
+    "created_at": "2026-03-19T05:50:29Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44796/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44796",
+    "files_url": "https://github.com/huggingface/transformers/pull/44844/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44844",
     "labels": [],
-    "merged": true,
-    "number": 44796,
-    "review_comments_count": 40,
+    "merged": false,
+    "number": 44844,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[refactor] Serving into proper modules",
-    "updated_at": "2026-04-01T16:03:37Z"
+    "title": "Fix DEIM config export and public API",
+    "updated_at": "2026-03-19T13:18:59Z"
   },
   {
-    "additions": 105726,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Adds support for a more generic path, aligned with the rest of the loading! model | PR | main ----|-----|---------- \"gdax/Qwen1.5-MoE-A2.7B_gguf\"| 1min 5s |1min 18s",
-    "changed_files": 1980,
+    "additions": 26,
+    "author": "omyaaa1",
+    "author_association": "NONE",
+    "body_excerpt": "Reintroduce handling for remote URLs using download_url, which was accidentally removed in recent versions. This restores support for loading image processor configs directly from URLs. Fixes #44821 # What does this PR do? <!-- Congratulat\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44794",
-    "created_at": "2026-03-17T11:35:33Z",
-    "deletions": 82149,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44794/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44794",
-    "labels": [],
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44842",
+    "created_at": "2026-03-19T04:48:58Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44842/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44842",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44794,
+    "number": 44842,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Refacto GGUF weight conversion",
-    "updated_at": "2026-04-15T09:58:59Z"
+    "state": "closed",
+    "title": "Fix AutoImageProcessor URL loading regression",
+    "updated_at": "2026-03-19T11:57:50Z"
   },
   {
-    "additions": 40,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #44792 - Handles None values in Janus model's image generation mode. The `generate()` method for image generation had several places where it assumed certain config values would always be set, causing failure\u2026",
-    "changed_files": 2,
+    "additions": 1,
+    "author": "zhulinchng",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fix formatting of code block in weightconverter.md # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44793",
-    "created_at": "2026-03-17T11:29:47Z",
-    "deletions": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44839",
+    "created_at": "2026-03-19T01:40:58Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44793/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44793",
+    "files_url": "https://github.com/huggingface/transformers/pull/44839/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44839",
     "labels": [],
-    "merged": false,
-    "number": 44793,
-    "review_comments_count": 2,
-    "state": "open",
-    "title": "fix(janus): Handle None values in image generation mode",
-    "updated_at": "2026-03-18T10:43:24Z"
+    "merged": true,
+    "number": 44839,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Correct code block formatting in weightconverter.md",
+    "updated_at": "2026-03-19T07:07:13Z"
   },
   {
-    "additions": 20,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? AMD Strix Halo APUs (gfx1151) have a driver bug where safetensors mmap doesn't release memory properly with the unified memory architecture. This causes OOM errors when loading models that should fit in memory (e.g.\u2026",
-    "changed_files": 1,
+    "additions": 41,
+    "author": "xr843",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Fixes #44821 \u2014 `AutoImageProcessor.from_pretrained` fails with `OSError: Repo id must be in the form 'repo_name' or 'namespace/repo_name'` when given a URL - The URL handling branch (`is_remote_url` check) in `get_image_proces\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44791",
-    "created_at": "2026-03-17T10:33:22Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44838",
+    "created_at": "2026-03-18T23:53:54Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44791/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44791",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44838/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44838",
+    "labels": [],
     "merged": false,
-    "number": 44791,
+    "number": 44838,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: disable mmap on Strix Halo APUs to avoid OOM",
-    "updated_at": "2026-03-18T12:33:05Z"
+    "title": "Fix AutoImageProcessor.from_pretrained failing with URL input",
+    "updated_at": "2026-03-19T10:43:53Z"
   },
   {
-    "additions": 72,
-    "author": "tarekziade",
+    "additions": 482,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This patch - adds a simple cache to the model linter so we skip files that did not change and were valid - reworks `Makefile` targets",
-    "changed_files": 6,
+    "body_excerpt": "backfills empty model cards like gptoss and nemotronh with a `model-card.md` skill i created. its pretty minimal at the moment and just includes a brief intro and code examples. let me know if there is anything else we should add!",
+    "changed_files": 12,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44790",
-    "created_at": "2026-03-17T08:54:47Z",
-    "deletions": 19,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44837",
+    "created_at": "2026-03-18T21:45:31Z",
+    "deletions": 102,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44790/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44790",
+    "files_url": "https://github.com/huggingface/transformers/pull/44837/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44837",
     "labels": [],
     "merged": true,
-    "number": 44790,
-    "review_comments_count": 1,
+    "number": 44837,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "feat: added cache to the model linter",
-    "updated_at": "2026-03-24T15:28:29Z"
+    "title": "[docs] model cards",
+    "updated_at": "2026-03-20T22:40:41Z"
   },
   {
-    "additions": 21,
-    "author": "ydshieh",
+    "additions": 96,
+    "author": "tyler-romero",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Pass cu_seqlens derived from packed attention masks to FLA's ShortConvolution and chunk_gated_delta_rule kernels, preventing recurrent state from leaking across sequence boundaries during packed-sequence training. F\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44836",
+    "created_at": "2026-03-18T20:24:58Z",
+    "deletions": 20,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44836/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44836",
+    "labels": [],
+    "merged": false,
+    "number": 44836,
+    "review_comments_count": 3,
+    "state": "open",
+    "title": "Add cu_seqlens support to OlmoHybridGatedDeltaNet for packed sequences",
+    "updated_at": "2026-03-19T05:34:43Z"
+  },
+  {
+    "additions": 187,
+    "author": "remi-or",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Some configs from the hub have different types.",
-    "changed_files": 7,
+    "body_excerpt": "## Summary This PR adds the `return_logprobs` flag to the continuous batching, enabling the user to retrieve the log probabilites of the tokens generated. # Tests Added a test to compare with regular generate and it passes. All tests pass.\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44789",
-    "created_at": "2026-03-17T08:41:30Z",
-    "deletions": 21,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44835",
+    "created_at": "2026-03-18T17:48:15Z",
+    "deletions": 83,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44789/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44789",
+    "files_url": "https://github.com/huggingface/transformers/pull/44835/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44835",
     "labels": [],
     "merged": true,
-    "number": 44789,
-    "review_comments_count": 5,
+    "number": 44835,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix config loading issues (type issues)",
-    "updated_at": "2026-03-17T09:44:50Z"
+    "title": "[CB] Add an option to return logprobs",
+    "updated_at": "2026-03-23T18:35:31Z"
   },
   {
     "additions": 0,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "The pipeline() docstring included an example using the 'question-answering' task, but this task is not in SUPPORTED_TASKS and will raise an error when used. Remove this outdated example to avoid confusing users following the documentation.\u2026",
-    "changed_files": 1,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Oups this slipped through in https://github.com/huggingface/transformers/pull/44833",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44788",
-    "created_at": "2026-03-17T08:38:25Z",
-    "deletions": 5,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44834",
+    "created_at": "2026-03-18T17:07:11Z",
+    "deletions": 11,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44788/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44788",
+    "files_url": "https://github.com/huggingface/transformers/pull/44834/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44834",
     "labels": [],
-    "merged": false,
-    "number": 44788,
+    "merged": true,
+    "number": 44834,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "docs(pipelines): remove outdated question-answering example",
-    "updated_at": "2026-03-23T17:19:33Z"
+    "title": "Update more modular examples",
+    "updated_at": "2026-03-18T17:18:54Z"
   },
   {
-    "additions": 4,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "The question-answering pipeline was removed in v5.0.0 per MIGRATION_GUIDE_V5.md, but the non-English task guides still referenced it. This updates the Arabic, Chinese, Japanese, and Korean question answering task guides to remove usage of\u2026",
-    "changed_files": 4,
+    "additions": 299,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix the examples after the config change (https://github.com/huggingface/transformers/pull/41250), and re-run conversion as in general modelings changed quite a bit in the lib.",
+    "changed_files": 19,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44787",
-    "created_at": "2026-03-17T08:24:09Z",
-    "deletions": 66,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44833",
+    "created_at": "2026-03-18T16:35:34Z",
+    "deletions": 590,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44787/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44787",
+    "files_url": "https://github.com/huggingface/transformers/pull/44833/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44833",
     "labels": [],
     "merged": true,
-    "number": 44787,
+    "number": 44833,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "docs(tasks): remove references to removed question-answering pipeline",
-    "updated_at": "2026-03-17T16:23:50Z"
+    "title": "Fix and re-run modular converter on examples",
+    "updated_at": "2026-03-18T17:00:44Z"
   },
   {
-    "additions": 25,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "AMD Strix Halo APUs (gfx1151) experience OOM errors when loading large models via safetensors mmap due to unified memory architecture issues. This fix detects Strix Halo GPUs by checking the GPU architecture name (gfx1151) and forces a CPU\u2026",
-    "changed_files": 1,
+    "additions": 346,
+    "author": "IlyasMoutawwakil",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44786",
-    "created_at": "2026-03-17T08:17:32Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44832",
+    "created_at": "2026-03-18T15:33:15Z",
+    "deletions": 176,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44786/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44786",
+    "files_url": "https://github.com/huggingface/transformers/pull/44832/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44832",
     "labels": [],
-    "merged": false,
-    "number": 44786,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44832,
+    "review_comments_count": 37,
     "state": "closed",
-    "title": "fix(core_model_loading): disable mmap on Strix Halo to avoid OOM",
-    "updated_at": "2026-03-17T10:29:44Z"
+    "title": "DeepGEMM",
+    "updated_at": "2026-03-31T15:04:06Z"
   },
   {
-    "additions": 307,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "AMD Strix Halo APUs (e.g., Radeon 8060S) have issues with mmap-based tensor loading from safetensors, causing out-of-memory errors even when sufficient memory is available. This fix: - Adds `is_strix_halo()` helper to detect Strix Halo GPU\u2026",
-    "changed_files": 1,
+    "additions": 2,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix loading in Sam3 which currently doesn't match the state dict keys from checkpoint. Adding a correct base model prefix will add it to all state dict keys, making the ckpt load-able <!-- Congratulations! You've ma\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44785",
-    "created_at": "2026-03-17T06:55:31Z",
-    "deletions": 83,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44831",
+    "created_at": "2026-03-18T14:50:48Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44785/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44785",
+    "files_url": "https://github.com/huggingface/transformers/pull/44831/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44831",
     "labels": [],
-    "merged": false,
-    "number": 44785,
+    "merged": true,
+    "number": 44831,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(model_loading): Disable mmap on Strix Halo to avoid OOM",
-    "updated_at": "2026-03-17T10:28:06Z"
+    "title": "Fix loading issue in Sam3",
+    "updated_at": "2026-03-18T15:44:01Z"
   },
   {
-    "additions": 2,
-    "author": "BillionClaw",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "This PR fixes the DeepSeek tokenizer issue where spaces were lost during decoding in Transformers v5. ## Problem DeepSeek V2 and V3 models use SentencePiece tokenization (like Llama) but were falling back to the generic TokenizersBackend i\u2026",
-    "changed_files": 1,
+    "additions": 2103,
+    "author": "lashahub",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "This PR adds `AudioFlamingoNext` as a separate model name that inherits directly from `MusicFlamingo` #43538 and keeps the same architecture and behavior. Changes: - add `audioflamingonext` model files - register it in the auto mappings -\u2026",
+    "changed_files": 27,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44783",
-    "created_at": "2026-03-17T05:58:54Z",
-    "deletions": 0,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44830",
+    "created_at": "2026-03-18T14:31:45Z",
+    "deletions": 48,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44783/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44783",
+    "files_url": "https://github.com/huggingface/transformers/pull/44830/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44830",
     "labels": [],
     "merged": false,
-    "number": 44783,
-    "review_comments_count": 0,
+    "number": 44830,
+    "review_comments_count": 2,
     "state": "open",
-    "title": "fix(auto): Map deepseek_v2 and deepseek_v3 to LlamaTokenizer",
-    "updated_at": "2026-04-15T14:48:02Z"
+    "title": "Add AudioFlamingoNext model",
+    "updated_at": "2026-04-15T21:31:08Z"
   },
   {
-    "additions": 6,
-    "author": "JiwaniZakir",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes #44737 `XLNetModel.relative_positional_encoding` was creating all `torch.arange` tensors on CPU by default, then calling `.to(output_h.device)` at the call site to move them. Adds a `device` parameter to `relative_positional_encoding\u2026",
-    "changed_files": 1,
+    "additions": 101,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Last batch finally! Follow up of https://github.com/huggingface/transformers/pull/44759 and many other",
+    "changed_files": 33,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44782",
-    "created_at": "2026-03-17T05:11:36Z",
-    "deletions": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44828",
+    "created_at": "2026-03-18T13:52:32Z",
+    "deletions": 512,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44782/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44782",
+    "files_url": "https://github.com/huggingface/transformers/pull/44828/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44828",
     "labels": [],
     "merged": true,
-    "number": 44782,
+    "number": 44828,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: XLNet: relative_positional_encoding computes on CPU every forward",
-    "updated_at": "2026-03-19T13:30:48Z"
-  },
-  {
-    "additions": 5,
-    "author": "bensons",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Some model repos provide `extra_special_tokens` as a list in their tokenizer_config.json, which caused an `AttributeError: 'list' object has no attribute 'keys'`. This converts list inputs to a dict mapping each tok\u2026",
-    "changed_files": 2,
+    "title": "Remove cache_position in more models (4 and last one)",
+    "updated_at": "2026-03-18T16:02:46Z"
+  },
+  {
+    "additions": 80,
+    "author": "3outeille",
+    "author_association": "MEMBER",
+    "body_excerpt": "https://github.com/huggingface/transformers/pull/44825",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44781",
-    "created_at": "2026-03-17T04:59:02Z",
-    "deletions": 2849,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44827",
+    "created_at": "2026-03-18T13:36:53Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44781/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44781",
+    "files_url": "https://github.com/huggingface/transformers/pull/44827/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44827",
     "labels": [],
     "merged": false,
-    "number": 44781,
-    "review_comments_count": 0,
+    "number": 44827,
+    "review_comments_count": 6,
     "state": "open",
-    "title": "Fix `_set_model_specific_special_tokens` to accept list-format `extra_special_tokens`",
-    "updated_at": "2026-03-27T23:19:21Z"
+    "title": "Fix Mistral4 tests",
+    "updated_at": "2026-03-26T16:04:20Z"
   },
   {
-    "additions": 145,
-    "author": "LincolnBurrows2017",
+    "additions": 55,
+    "author": "BillionClaw",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixed logic error in is_tiktoken_available function. The original code `return with_blobfile and _is_package_available(\"blobfile\")[0] or True` would always return True due to operator precedence.",
-    "changed_files": 8,
+    "body_excerpt": "Fixes #44821 This PR fixes the issue where `AutoImageProcessor.from_pretrained()` was unable to load from a URL (e.g., `https://huggingface.co/.../raw/main/config.json`). The bug was introduced in transformers>=5.3.0. Prior versions (e.g.,\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44778",
-    "created_at": "2026-03-16T23:41:29Z",
-    "deletions": 28,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44826",
+    "created_at": "2026-03-18T12:08:35Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44778/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44778",
+    "files_url": "https://github.com/huggingface/transformers/pull/44826/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44826",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44778,
+    "number": 44826,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: correct logic error in is_tiktoken_available function",
-    "updated_at": "2026-03-18T13:15:37Z"
+    "title": "fix: allow AutoImageProcessor to load from URL",
+    "updated_at": "2026-03-19T13:28:04Z"
   },
   {
-    "additions": 35,
-    "author": "stevhliu",
+    "additions": 1,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "adds docs for #43705 (enable bidirectional attention for decoder-only models)",
+    "body_excerpt": "# What does this PR do? As per the title. cc @3outeille as I know you're looking into it",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44777",
-    "created_at": "2026-03-16T21:58:40Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44825",
+    "created_at": "2026-03-18T12:05:35Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44777/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44777",
+    "files_url": "https://github.com/huggingface/transformers/pull/44825/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44825",
     "labels": [],
     "merged": true,
-    "number": 44777,
-    "review_comments_count": 1,
+    "number": 44825,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] is_causal feature",
-    "updated_at": "2026-03-17T19:50:43Z"
+    "title": "[CI] Temporarily skip Mistral4 tests as they almost all fail",
+    "updated_at": "2026-03-18T12:15:34Z"
   },
   {
-    "additions": 0,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "the doc-builder is breaking because it can't find `Mistral4ForQuestionAnswering`, which looks like it doesn't exist",
+    "additions": 5,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Description Fixes #44737 The `relative_positional_encoding` function in XLNet was computing all positional encodings on CPU every forward pass because the `torch.arange` calls were missing the `device` parameter. ## Changes - Added devi\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44776",
-    "created_at": "2026-03-16T20:43:33Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44824",
+    "created_at": "2026-03-18T11:55:01Z",
     "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44776/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44776",
-    "labels": [],
-    "merged": true,
-    "number": 44776,
+    "files_url": "https://github.com/huggingface/transformers/pull/44824/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44824",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44824,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[fix] mistral 4 docs",
-    "updated_at": "2026-03-16T21:11:29Z"
+    "title": "fix(xlnet): add device parameter to relative_positional_encoding",
+    "updated_at": "2026-03-18T13:17:38Z"
   },
   {
-    "additions": 177,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "refactors the current [Parallelism methods](https://huggingface.co/docs/transformers/main/en/perf_train_gpu_many#zero-data-parallelism-pipeline-parallelism-and-model-parallelism-3d-parallelism) doc to: - focus on practical examples of comb\u2026",
+    "additions": 41,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "This PR fixes #44821 where `AutoImageProcessor.from_pretrained()` couldn't load from a direct URL to a config file. ## Problem When passing a URL like `https://huggingface.co/jinfengxie/BFMS_1014/raw/main/config.json` to `AutoImageProcesso\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44775",
-    "created_at": "2026-03-16T20:23:29Z",
-    "deletions": 109,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44775/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44775",
-    "labels": [],
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44823",
+    "created_at": "2026-03-18T11:54:24Z",
+    "deletions": 5,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44823/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44823",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44775,
+    "number": 44823,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[docs] n-d parallelism",
-    "updated_at": "2026-03-16T20:28:48Z"
+    "state": "closed",
+    "title": "fix: AutoImageProcessor from URL loading",
+    "updated_at": "2026-03-18T13:17:48Z"
   },
   {
-    "additions": 0,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Remove `is_causal` from `EuroBertConfig`",
-    "changed_files": 2,
+    "additions": 4,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #44737 The relative_positional_encoding method in XLNetModel was creating tensors using torch.arange() without specifying device=, causing the entire sinusoidal positional encoding computation to run on CPU every forward pass. Only t\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44774",
-    "created_at": "2026-03-16T18:56:19Z",
-    "deletions": 6,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44822",
+    "created_at": "2026-03-18T11:48:28Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44774/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44774",
-    "labels": [],
-    "merged": true,
-    "number": 44774,
+    "files_url": "https://github.com/huggingface/transformers/pull/44822/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44822",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44822,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Remove `is_causal` from `EuroBertConfig`",
-    "updated_at": "2026-03-17T09:33:21Z"
+    "title": "fix: XLNet relative_positional_encoding device placement",
+    "updated_at": "2026-03-18T13:17:30Z"
   },
   {
-    "additions": 3,
-    "author": "githubnemo",
-    "author_association": "MEMBER",
-    "body_excerpt": "The links to the quantization offloading were outdated and 4-bit quantization also supports offloading which should be mentioned. cc @SunMarc",
-    "changed_files": 3,
+    "additions": 14,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #44811 ## Problem When calling `processor.batch_decode(predicted_ids, skip_special_tokens=False)` with the output from `model.generate()` (without `return_dict_in_generate=True`), the `skip_special_tokens` parameter was being ignored\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44772",
-    "created_at": "2026-03-16T18:46:13Z",
-    "deletions": 3,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44820",
+    "created_at": "2026-03-18T10:57:12Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44772/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44772",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44820/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44820",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44772,
+    "number": 44820,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "bitsandbytes: Update links and docs",
-    "updated_at": "2026-03-17T15:57:56Z"
+    "state": "closed",
+    "title": "fix(whisper): respect skip_special_tokens in batch_decode",
+    "updated_at": "2026-03-18T13:17:20Z"
   },
   {
     "additions": 2,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? wtf",
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Description This PR fixes the DeepSeek tokenizer issue (#44779) where tokenization produces incorrect results in v5. ### Problem In transformers v5, the DeepSeek tokenizer (DeepSeek-R1) was producing incorrect results: - Input: \"How are\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44771",
-    "created_at": "2026-03-16T18:45:11Z",
-    "deletions": 1,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44771/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44771",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44819",
+    "created_at": "2026-03-18T10:55:22Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44819/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44819",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44819,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix(tokenizer): add deepseek_v2 and deepseek_v3 to incorrect hub tokenizer class list",
+    "updated_at": "2026-03-18T14:11:16Z"
+  },
+  {
+    "additions": 64,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Description Fixes #44805 When training multimodal models (Qwen3-VL, GLM-4.6V, Qwen3-VL-MoE) with LoRA adapters, the `attention_mask` and `mm_token_type_ids` tensors can have different shapes. This causes an IndexError when the `get_rope\u2026",
+    "changed_files": 4,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44818",
+    "created_at": "2026-03-18T10:46:22Z",
+    "deletions": 7,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44818/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44818",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44818,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix: resolve mask shape mismatch IndexError in multimodal VL models",
+    "updated_at": "2026-03-18T10:51:43Z"
+  },
+  {
+    "additions": 28,
+    "author": "JJJYmmm",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fix https://github.com/QwenLM/Qwen3.5/issues/97. This PR adds `enable_thinking` to the chat-template kwargs. With this change, `enable_thinking` is treated as a template-level argument in the tokenize=True path, so\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44817",
+    "created_at": "2026-03-18T10:44:11Z",
+    "deletions": 6,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44817/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44817",
     "labels": [],
     "merged": false,
-    "number": 44771,
+    "number": 44817,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "wtf",
-    "updated_at": "2026-03-16T18:56:00Z"
+    "state": "closed",
+    "title": "[Misc] add enable_thinking to template kwargs",
+    "updated_at": "2026-03-20T14:56:04Z"
   },
   {
-    "additions": 203,
-    "author": "zucchini-nlp",
+    "additions": 98,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix tests failing because of `strict` type validation and decorate two missing configs, Nemotron and VibeVoice",
-    "changed_files": 12,
+    "body_excerpt": "# What does this PR do? As per the title. This is the last of many PR to remove the `cache_position`. At this point, all the models were already updated to not use them, and they are fully ignored in all the modelings. So this removes thei\u2026",
+    "changed_files": 57,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44770",
-    "created_at": "2026-03-16T18:44:03Z",
-    "deletions": 268,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44816",
+    "created_at": "2026-03-18T10:32:04Z",
+    "deletions": 375,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44770/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44770",
+    "files_url": "https://github.com/huggingface/transformers/pull/44816/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44816",
     "labels": [],
     "merged": true,
-    "number": 44770,
-    "review_comments_count": 1,
+    "number": 44816,
+    "review_comments_count": 14,
     "state": "closed",
-    "title": "Fix configs with `@strict`",
-    "updated_at": "2026-03-17T15:39:43Z"
+    "title": "[generate] Never use `cache_position` anymore in generation",
+    "updated_at": "2026-03-19T14:18:28Z"
   },
   {
-    "additions": 145,
-    "author": "LincolnBurrows2017",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary The `is_batched_video()` and `convert_pil_frames_to_video()` functions in `src/transformers/video_utils.py` were accessing `videos[0]` without first checking if the list is empty, causing `IndexError` when empty lists are passed\u2026",
-    "changed_files": 8,
+    "additions": 135,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44769",
-    "created_at": "2026-03-16T18:40:07Z",
-    "deletions": 28,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44769/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44769",
-    "labels": [
-      "Code agent slop"
-    ],
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44815",
+    "created_at": "2026-03-18T09:54:18Z",
+    "deletions": 23,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44815/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44815",
+    "labels": [],
     "merged": false,
-    "number": 44769,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix: Handle empty lists in video_utils functions",
-    "updated_at": "2026-03-18T13:15:55Z"
+    "number": 44815,
+    "review_comments_count": 2,
+    "state": "open",
+    "title": "Dequant fix",
+    "updated_at": "2026-03-24T14:39:52Z"
   },
   {
-    "additions": 20,
-    "author": "michaelbenayoun",
-    "author_association": "MEMBER",
-    "body_excerpt": "The function `add_tensor_parallel_hooks_to_module` has unused parameters, in this PR we: - Remove `tp_plan`, which is not used. - Remove `parameter_name` which is not used - Remove `layer_name`. This parameter is only used for logging purp\u2026",
+    "additions": 2,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #44792 This PR fixes the failing test `test_model_generate_images` for the Janus model. ## Problem When generating images with the Janus model, `generation_config.num_return_sequences` and `generation_config.max_length` can be `None`\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44768",
-    "created_at": "2026-03-16T18:29:52Z",
-    "deletions": 9,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44814",
+    "created_at": "2026-03-18T09:51:34Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44768/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44768",
+    "files_url": "https://github.com/huggingface/transformers/pull/44814/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44814",
     "labels": [],
-    "merged": true,
-    "number": 44768,
-    "review_comments_count": 3,
+    "merged": false,
+    "number": 44814,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Remove unused parameters and improve add_tensor_parallel_hooks_t\u2026",
-    "updated_at": "2026-04-09T17:11:55Z"
+    "title": "fix(janus): handle None values in image generation mode",
+    "updated_at": "2026-03-18T10:42:50Z"
   },
   {
-    "additions": 11,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? EuroBertConfig was missing `@strict(accept_kwargs=True)` unlike its parent LlamaConfig, causing failures when reloading saved configs that include extra keys like `architectures`. Also fixed the test helper passing\u2026",
-    "changed_files": 3,
+    "additions": 20,
+    "author": "hf-security-analysis[bot]",
+    "author_association": "NONE",
+    "body_excerpt": "Update `.github/workflows/pr-repo-consistency-bot.yml` workflow configuration. cc @ydshieh Closes huggingface/tracking-issues#26",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44767",
-    "created_at": "2026-03-16T17:31:26Z",
-    "deletions": 5,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44813",
+    "created_at": "2026-03-18T09:49:16Z",
+    "deletions": 46,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44767/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44767",
+    "files_url": "https://github.com/huggingface/transformers/pull/44813/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44813",
     "labels": [],
-    "merged": true,
-    "number": 44767,
-    "review_comments_count": 6,
+    "merged": false,
+    "number": 44813,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix: Eurobert model was missing @strict decorator and invalid test kwargs",
-    "updated_at": "2026-03-16T19:02:31Z"
+    "title": "chore: update pr-repo-consistency-bot.yml",
+    "updated_at": "2026-03-18T10:08:04Z"
   },
   {
-    "additions": 26,
-    "author": "itazap",
+    "additions": 7,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "for when remote code tries to import from `tokenization_xxx_fast`",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Some checks (for example, modular checks) really require the installation from PR branch.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44766",
-    "created_at": "2026-03-16T17:30:23Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44812",
+    "created_at": "2026-03-18T09:40:36Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44766/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44766",
+    "files_url": "https://github.com/huggingface/transformers/pull/44812/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44812",
     "labels": [],
     "merged": true,
-    "number": 44766,
+    "number": 44812,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "support xxxFast alias in v5 tokenizers",
-    "updated_at": "2026-03-18T13:40:05Z"
+    "title": "Fix repo-check bot",
+    "updated_at": "2026-03-18T09:49:50Z"
   },
   {
-    "additions": 19,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following failing tests were identified and fixed in this PR: \u2192 **PaliGemma 2:** The [PaliGemma 1 test class](https://github.com/huggingface/transformers/blob/main/tests/models/paligemma/test_modeling_paligemm\u2026",
+    "additions": 7,
+    "author": "LincolnBurrows2017",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixed bare except clause in _safe_convert_tensor function to catch only Exception type.",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44765",
-    "created_at": "2026-03-16T17:26:22Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44809",
+    "created_at": "2026-03-18T05:47:03Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44765/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44765",
-    "labels": [],
-    "merged": true,
-    "number": 44765,
+    "files_url": "https://github.com/huggingface/transformers/pull/44809/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44809",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44809,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(testing): Fix PaliGemma 2 and PaddleOCR-VL test failures on main",
-    "updated_at": "2026-03-20T13:55:55Z"
+    "title": "fix: replace bare except with Exception in Fuyu image processing",
+    "updated_at": "2026-03-18T13:17:11Z"
   },
   {
-    "additions": 12,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes the siglip import. that was also crashing the test fetcher",
-    "changed_files": 3,
+    "additions": 2759,
+    "author": "zhang-prog",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": null,
+    "changed_files": 31,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44764",
-    "created_at": "2026-03-16T17:15:40Z",
-    "deletions": 4,
+    "comments_count": 12,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44808",
+    "created_at": "2026-03-18T04:29:07Z",
+    "deletions": 21,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44764/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44764",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44808/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44808",
+    "labels": [
+      "New model"
+    ],
     "merged": true,
-    "number": 44764,
-    "review_comments_count": 2,
+    "number": 44808,
+    "review_comments_count": 53,
     "state": "closed",
-    "title": "fix: sig lip import",
-    "updated_at": "2026-03-16T17:38:41Z"
+    "title": "[Model] Add PP-OCRv5_server_rec and  PP-OCRv5_mobile_rec models Support",
+    "updated_at": "2026-03-18T20:24:50Z"
   },
   {
-    "additions": 17,
-    "author": "xenova",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Adds support for MLP mixers, used by [nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16). Previously, it would crash because it would not recognize the `-` char in t\u2026",
+    "additions": 0,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "The question-answering pipeline was removed in v5.3 per the migration guide, but the Chinese, Korean, and French quicktour docs still listed it as an available pipeline task. This removes those outdated references to avoid confusing users\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44763",
-    "created_at": "2026-03-16T17:04:36Z",
-    "deletions": 5,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44807",
+    "created_at": "2026-03-18T03:41:52Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44763/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44763",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44807/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44807",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44763,
-    "review_comments_count": 1,
+    "number": 44807,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[nemotron_h] Add support for MLP mixers",
-    "updated_at": "2026-04-14T13:46:14Z"
+    "title": "docs(quicktour): remove question-answering pipeline from quicktour tables",
+    "updated_at": "2026-03-18T15:38:09Z"
   },
   {
-    "additions": 4,
+    "additions": 23,
     "author": "BillionClaw",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "XLNet.relative_positional_encoding creates intermediate tensors on CPU every forward pass because torch.arange was missing the device parameter. This causes unnecessary CPU-GPU transfers when running on CUDA. Added device=self.device to al\u2026",
+    "body_excerpt": "This PR fixes OOM errors when loading models on AMD Strix Halo APUs. ## Problem AMD Strix Halo (Radeon 8060S/8050S) uses unified memory architecture where memory-mapped file loading doesn't work well with the current amdgpu driver. This ca\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44762",
-    "created_at": "2026-03-16T16:17:54Z",
-    "deletions": 4,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44806",
+    "created_at": "2026-03-18T03:33:59Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44762/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44762",
+    "files_url": "https://github.com/huggingface/transformers/pull/44806/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44806",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44762,
+    "number": 44806,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: Cache XLNet relative_positional_encoding to avoid CPU computation",
-    "updated_at": "2026-03-18T15:16:14Z"
+    "title": "fix(core_model_loading): disable mmap on Strix Halo to avoid OOM",
+    "updated_at": "2026-03-18T12:30:21Z"
   },
   {
-    "additions": 152,
-    "author": "tarekziade",
+    "additions": 137,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This adds rule 10: ``` Direct config definitions must use @strict(accept_kwargs=True). ```",
+    "body_excerpt": "updates the peft docs: - a more complete training section with a full code snippet, describe saving behavior, resuming from a checkpoint, and distributed training - adds some undocumented API methods (`delete_adapter`, `active_adapters`) -\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44761",
-    "created_at": "2026-03-16T16:05:03Z",
-    "deletions": 7,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44804",
+    "created_at": "2026-03-18T00:08:54Z",
+    "deletions": 89,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44761/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44761",
+    "files_url": "https://github.com/huggingface/transformers/pull/44804/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44804",
     "labels": [],
     "merged": true,
-    "number": 44761,
-    "review_comments_count": 7,
+    "number": 44804,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "model-linter: Added rule 10",
-    "updated_at": "2026-03-17T08:52:19Z"
+    "title": "[docs] peft",
+    "updated_at": "2026-03-23T17:14:58Z"
   },
   {
-    "additions": 2090,
-    "author": "juliendenize",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 15,
+    "additions": 1341,
+    "author": "yonigozlan",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? (Finally) add support for checking+fixing both generated files and modular files in `check_auto_docstrings`. Also `auto_docstring` was recently added to configs, and this PR updates `check_auto_docstrings` to suppor\u2026",
+    "changed_files": 244,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 12,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44760",
-    "created_at": "2026-03-16T15:54:11Z",
-    "deletions": 4,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44803",
+    "created_at": "2026-03-17T22:40:45Z",
+    "deletions": 1105,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44760/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44760",
-    "labels": [
-      "New model"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44803/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44803",
+    "labels": [],
     "merged": true,
-    "number": 44760,
-    "review_comments_count": 8,
+    "number": 44803,
+    "review_comments_count": 24,
     "state": "closed",
-    "title": "Add Mistral 4",
-    "updated_at": "2026-03-20T10:44:48Z"
+    "title": "Support Modular (!!) + Configs in `check_auto_docstrings`",
+    "updated_at": "2026-03-24T17:59:12Z"
   },
   {
-    "additions": 419,
-    "author": "Cyrilvallez",
+    "additions": 12,
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Follow-up of many related PR, last one in time being https://github.com/huggingface/transformers/pull/44602. This PR completes all the models that may need non-trivial treatment. Only about 30-40 models still have m\u2026",
-    "changed_files": 42,
+    "body_excerpt": "deepseek_v2, deepseek_v3, and modernbert fix for having incorrect tokenizer class on the hub fixes: https://github.com/huggingface/transformers/issues/44779, https://github.com/huggingface/transformers/pull/44783",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44759",
-    "created_at": "2026-03-16T15:38:13Z",
-    "deletions": 983,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44801",
+    "created_at": "2026-03-17T17:40:25Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44759/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44759",
+    "files_url": "https://github.com/huggingface/transformers/pull/44801/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44801",
     "labels": [],
     "merged": true,
-    "number": 44759,
-    "review_comments_count": 26,
+    "number": 44801,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Remove cache_position in more models (3)",
-    "updated_at": "2026-03-18T13:09:37Z"
+    "title": "deepseek_v2, deepseek_v3, and modernbert fix for having incorrect tokenizer class on the hub",
+    "updated_at": "2026-03-19T13:11:54Z"
   },
   {
-    "additions": 825,
-    "author": "LysandreJik",
+    "additions": 36,
+    "author": "aayushbaluni",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #44488 `CamembertTokenizer` raised `ValueError: too many values to unpack (expected 2)` when loading models like `cjvt/sleng-bert` that provide vocab as a dict `{token: id}` from `tokenizer.json` (BPE format). The tokenize\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44800",
+    "created_at": "2026-03-17T17:20:35Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44800/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44800",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44800,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix: handle dict vocab in CamembertTokenizer for tokenizer.json (#44488)",
+    "updated_at": "2026-03-18T15:37:54Z"
+  },
+  {
+    "additions": 327,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "Provides a nicer feedback when `transformers chat` loads a model, instead of hanging https://github.com/user-attachments/assets/8f68f914-b702-4430-b97f-e8cc25326b70 <p>Adds a <code>POST /load_model</code> endpoint to <code>transformers ser\u2026",
-    "changed_files": 7,
+    "body_excerpt": "updates the Hardware section of the docs for training: - combined CPU/Distributed CPU into a single doc - add more info to the Gaudi doc (mixed precision, torch.compile, distributed training) - add more info to the MPS doc (mixed precision\u2026",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44758",
-    "created_at": "2026-03-16T15:02:15Z",
-    "deletions": 63,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44799",
+    "created_at": "2026-03-17T17:19:51Z",
+    "deletions": 574,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44758/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44758",
+    "files_url": "https://github.com/huggingface/transformers/pull/44799/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44799",
     "labels": [],
     "merged": true,
-    "number": 44758,
-    "review_comments_count": 20,
+    "number": 44799,
+    "review_comments_count": 24,
     "state": "closed",
-    "title": "Propagate the model loading from transformers serve to chat",
-    "updated_at": "2026-03-19T17:20:03Z"
+    "title": "[docs] training on specific hardware",
+    "updated_at": "2026-04-13T07:12:08Z"
   },
   {
-    "additions": 1,
-    "author": "dacorvo",
-    "author_association": "MEMBER",
-    "body_excerpt": "## Summary - `_valid_auto_compile_criteria()` gates auto-compilation on `device.type in [\"cuda\", \"xpu\"]`, excluding Neuron devices. This means `torch.compile` never triggers automatically on Neuron even when `StaticCache` is used (which se\u2026",
+    "additions": 17,
+    "author": "divyanks",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44757",
-    "created_at": "2026-03-16T14:54:38Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44798",
+    "created_at": "2026-03-17T16:51:46Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44757/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44757",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44798/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44798",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44757,
-    "review_comments_count": 6,
-    "state": "open",
-    "title": "Add Neuron to auto-compile hardware list",
-    "updated_at": "2026-04-14T20:04:30Z"
+    "number": 44798,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Add MPS (Apple Silicon) example and documentation",
+    "updated_at": "2026-03-18T15:37:09Z"
   },
   {
-    "additions": 4,
-    "author": "zucchini-nlp",
+    "additions": 1,
+    "author": "vasqu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 6,
+    "body_excerpt": "Depends on #44887 and kernels being version `12.3` Works OOB with little changes! Example script for demonstration: ```python from transformers import AutoModelForCausalLM, AutoTokenizer fa_version = 4 #model_id = \"openai/gpt-oss-20b\" mode\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44755",
-    "created_at": "2026-03-16T14:08:34Z",
-    "deletions": 148,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44797",
+    "created_at": "2026-03-17T15:35:59Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44755/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44755",
+    "files_url": "https://github.com/huggingface/transformers/pull/44797/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44797",
     "labels": [],
-    "merged": false,
-    "number": 44755,
+    "merged": true,
+    "number": 44797,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Dont merge, testing smth",
-    "updated_at": "2026-03-18T10:09:15Z"
+    "title": "[`FA4`] Add kernels fallback",
+    "updated_at": "2026-03-20T19:03:24Z"
   },
-  {
-    "additions": 20,
-    "author": "gambletan",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44737 - Added `device=self.device` to all four `torch.arange()` calls in `XLNetModel.relative_positional_encoding()` so that intermediate tensors are created directly on the model's device instead of always on CPU. - With\u2026",
-    "changed_files": 2,
+  {
+    "additions": 4522,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR refactors transformers serve so that it is not in a single file. We split it into multiple files with clear responsabilities. There were 2,293 lines initially in the serve.py file. ``` \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2026",
+    "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44753",
-    "created_at": "2026-03-16T14:01:08Z",
-    "deletions": 4,
+    "comments_count": 15,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44796",
+    "created_at": "2026-03-17T13:04:06Z",
+    "deletions": 3100,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44753/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44753",
+    "files_url": "https://github.com/huggingface/transformers/pull/44796/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44796",
     "labels": [],
-    "merged": false,
-    "number": 44753,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44796,
+    "review_comments_count": 40,
     "state": "closed",
-    "title": "fix: add device= to torch.arange in XLNet relative_positional_encoding",
-    "updated_at": "2026-03-18T13:28:40Z"
+    "title": "[refactor] Serving into proper modules",
+    "updated_at": "2026-04-01T16:03:37Z"
   },
   {
-    "additions": 100,
-    "author": "LincolnBurrows2017",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR fix? The `flatten()` function in `tokenization_utils_base.py` had a bug where it was checking `arr[0]` instead of `sub_arr` when determining if an element should be recursively flattened. ### Bug Details - **File**: `s\u2026",
-    "changed_files": 6,
+    "additions": 105726,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Adds support for a more generic path, aligned with the rest of the loading! model | PR | main ----|-----|---------- \"gdax/Qwen1.5-MoE-A2.7B_gguf\"| 1min 5s |1min 18s",
+    "changed_files": 1980,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44751",
-    "created_at": "2026-03-16T13:40:44Z",
-    "deletions": 29,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44751/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44751",
-    "labels": [
-      "Code agent slop"
-    ],
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44794",
+    "created_at": "2026-03-17T11:35:33Z",
+    "deletions": 82149,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44794/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44794",
+    "labels": [],
     "merged": false,
-    "number": 44751,
+    "number": 44794,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix: Correct variable reference in flatten() function",
-    "updated_at": "2026-03-18T13:16:12Z"
+    "state": "open",
+    "title": "Refacto GGUF weight conversion",
+    "updated_at": "2026-04-15T09:58:59Z"
   },
   {
-    "additions": 1,
-    "author": "juliendenize",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds `apply_yarn_scaling` as an optional key for yarn repo. This was requested as part of a vLLM PR https://github.com/vllm-project/vllm/pull/37104 that seeks to silence some rope issues when converting Mist\u2026",
-    "changed_files": 1,
+    "additions": 40,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #44792 - Handles None values in Janus model's image generation mode. The `generate()` method for image generation had several places where it assumed certain config values would always be set, causing failure\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44747",
-    "created_at": "2026-03-16T10:32:54Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44793",
+    "created_at": "2026-03-17T11:29:47Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44747/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44747",
+    "files_url": "https://github.com/huggingface/transformers/pull/44793/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44793",
     "labels": [],
     "merged": false,
-    "number": 44747,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add apply_yarn_scaling as optional key to yarn",
-    "updated_at": "2026-03-16T12:48:08Z"
+    "number": 44793,
+    "review_comments_count": 2,
+    "state": "open",
+    "title": "fix(janus): Handle None values in image generation mode",
+    "updated_at": "2026-03-18T10:43:24Z"
   },
   {
-    "additions": 202,
-    "author": "LincolnBurrows2017",
+    "additions": 20,
+    "author": "BillionClaw",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixed Issue #44737: XLNet relative_positional_encoding function missing device parameter in torch.arange calls.",
-    "changed_files": 11,
+    "body_excerpt": "# What does this PR do? AMD Strix Halo APUs (gfx1151) have a driver bug where safetensors mmap doesn't release memory properly with the unified memory architecture. This causes OOM errors when loading models that should fit in memory (e.g.\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44745",
-    "created_at": "2026-03-16T09:39:30Z",
-    "deletions": 33,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44791",
+    "created_at": "2026-03-17T10:33:22Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44745/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44745",
+    "files_url": "https://github.com/huggingface/transformers/pull/44791/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44791",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44745,
+    "number": 44791,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: add device parameter to torch.arange calls in XLNet",
-    "updated_at": "2026-03-18T13:16:43Z"
+    "title": "fix: disable mmap on Strix Halo APUs to avoid OOM",
+    "updated_at": "2026-03-18T12:33:05Z"
   },
   {
-    "additions": 35,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": null,
-    "changed_files": 2,
+    "additions": 72,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This patch - adds a simple cache to the model linter so we skip files that did not change and were valid - reworks `Makefile` targets",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44739",
-    "created_at": "2026-03-16T07:15:33Z",
-    "deletions": 7,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44790",
+    "created_at": "2026-03-17T08:54:47Z",
+    "deletions": 19,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44739/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44739",
+    "files_url": "https://github.com/huggingface/transformers/pull/44790/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44790",
     "labels": [],
     "merged": true,
-    "number": 44739,
-    "review_comments_count": 12,
+    "number": 44790,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "fix series of failed test case for janus model",
-    "updated_at": "2026-04-01T08:24:26Z"
+    "title": "feat: added cache to the model linter",
+    "updated_at": "2026-03-24T15:28:29Z"
   },
   {
-    "additions": 6,
-    "author": "yunhaoli24",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes #44031 ## The Problem The condition for calling `_patch_mistral_regex` was too broad (`vocab_size > 100000`), causing non-Mistral models like Qwen, LLaMA, BGE-Reranker to show incorrect regex pattern warnings\u2026",
-    "changed_files": 1,
+    "additions": 21,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Some configs from the hub have different types.",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44736",
-    "created_at": "2026-03-16T06:00:47Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44789",
+    "created_at": "2026-03-17T08:41:30Z",
+    "deletions": 21,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44736/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44736",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44736,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44789/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44789",
+    "labels": [],
+    "merged": true,
+    "number": 44789,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "fix: resolve false-positive regex warning for non-mistral models",
-    "updated_at": "2026-03-18T15:08:59Z"
+    "title": "Fix config loading issues (type issues)",
+    "updated_at": "2026-03-17T09:44:50Z"
   },
   {
-    "additions": 1,
-    "author": "mango766",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes a crash in `transformers serve` when the `/v1/responses` streaming endpoint attempts to reuse a KV cache from a previous request in the same conversation session. ### The bug In `generate_response`, `inputs`\u2026",
+    "additions": 0,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "The pipeline() docstring included an example using the 'question-answering' task, but this task is not in SUPPORTED_TASKS and will raise an error when used. Remove this outdated example to avoid confusing users following the documentation.\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44735",
-    "created_at": "2026-03-16T04:09:32Z",
-    "deletions": 1,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44788",
+    "created_at": "2026-03-17T08:38:25Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44735/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44735",
+    "files_url": "https://github.com/huggingface/transformers/pull/44788/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44788",
     "labels": [],
-    "merged": true,
-    "number": 44735,
+    "merged": false,
+    "number": 44788,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix tensor indexing crash in serve generate_response KV cache continuation",
-    "updated_at": "2026-03-16T15:27:59Z"
+    "title": "docs(pipelines): remove outdated question-answering example",
+    "updated_at": "2026-03-23T17:19:33Z"
   },
   {
-    "additions": 28,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@IlyasMoutawwakil, pls help review, thx!",
-    "changed_files": 3,
+    "additions": 4,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "The question-answering pipeline was removed in v5.0.0 per MIGRATION_GUIDE_V5.md, but the non-English task guides still referenced it. This updates the Arabic, Chinese, Japanese, and Korean question answering task guides to remove usage of\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44733",
-    "created_at": "2026-03-16T02:55:54Z",
-    "deletions": 10,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44787",
+    "created_at": "2026-03-17T08:24:09Z",
+    "deletions": 66,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44733/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44733",
+    "files_url": "https://github.com/huggingface/transformers/pull/44787/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44787",
     "labels": [],
     "merged": true,
-    "number": 44733,
+    "number": 44787,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix bug and add XPU Expectations for qwen2 and jamba tests",
-    "updated_at": "2026-04-01T08:24:40Z"
+    "title": "docs(tasks): remove references to removed question-answering pipeline",
+    "updated_at": "2026-03-17T16:23:50Z"
   },
   {
-    "additions": 1,
-    "author": "Defalt-Meh",
+    "additions": 25,
+    "author": "BillionClaw",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? While running SmolVLM tests I noticed this warning in the output: ``` tests/test_video_processing_common.py:57: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider convert\u2026",
+    "body_excerpt": "AMD Strix Halo APUs (gfx1151) experience OOM errors when loading large models via safetensors mmap due to unified memory architecture issues. This fix detects Strix Halo GPUs by checking the GPU architecture name (gfx1151) and forces a CPU\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44731",
-    "created_at": "2026-03-15T23:26:31Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44786",
+    "created_at": "2026-03-17T08:17:32Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44731/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44731",
+    "files_url": "https://github.com/huggingface/transformers/pull/44786/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44786",
     "labels": [],
     "merged": false,
-    "number": 44731,
+    "number": 44786,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[Tests] Fix slow video tensor creation from list of numpy arrays in SmolVLM",
-    "updated_at": "2026-03-15T23:26:31Z"
+    "state": "closed",
+    "title": "fix(core_model_loading): disable mmap on Strix Halo to avoid OOM",
+    "updated_at": "2026-03-17T10:29:44Z"
   },
   {
-    "additions": 9,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? It's unclear why the config class has `model_type = \"mlcd_vision_model\"` but the model on the hub has \"model_type\": \"mlcd\". This leads to the following failures (load from hub --> save locally --> local locally) ```\u2026",
-    "changed_files": 2,
+    "additions": 307,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "AMD Strix Halo APUs (e.g., Radeon 8060S) have issues with mmap-based tensor loading from safetensors, causing out-of-memory errors even when sufficient memory is available. This fix: - Adds `is_strix_halo()` helper to detect Strix Halo GPU\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44730",
-    "created_at": "2026-03-15T20:44:32Z",
-    "deletions": 4,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44785",
+    "created_at": "2026-03-17T06:55:31Z",
+    "deletions": 83,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44730/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44730",
+    "files_url": "https://github.com/huggingface/transformers/pull/44785/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44785",
     "labels": [],
-    "merged": true,
-    "number": 44730,
-    "review_comments_count": 6,
+    "merged": false,
+    "number": 44785,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix `mlcd` auto config/model/mapping issues",
-    "updated_at": "2026-03-16T12:12:30Z"
+    "title": "fix(model_loading): Disable mmap on Strix Halo to avoid OOM",
+    "updated_at": "2026-03-17T10:28:06Z"
   },
   {
-    "additions": 214,
-    "author": "xenova",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR introduces a helper utility function, `int_div_ceil`, which performs `math.ceil(a / b)` for non-negative integer operands. This is necessary as the current approach is both error-prone and imprecise (especia\u2026",
-    "changed_files": 58,
+    "additions": 2,
+    "author": "BillionClaw",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "This PR fixes the DeepSeek tokenizer issue where spaces were lost during decoding in Transformers v5. ## Problem DeepSeek V2 and V3 models use SentencePiece tokenization (like Llama) but were falling back to the generic TokenizersBackend i\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44729",
-    "created_at": "2026-03-15T20:29:38Z",
-    "deletions": 225,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44783",
+    "created_at": "2026-03-17T05:58:54Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44729/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44729",
+    "files_url": "https://github.com/huggingface/transformers/pull/44783/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44783",
     "labels": [],
     "merged": false,
-    "number": 44729,
+    "number": 44783,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Avoid floating point math for ceil operations",
-    "updated_at": "2026-03-15T20:49:34Z"
+    "state": "closed",
+    "title": "fix(auto): Map deepseek_v2 and deepseek_v3 to LlamaTokenizer",
+    "updated_at": "2026-04-16T10:47:47Z"
   },
   {
-    "additions": 88,
-    "author": "ajmeese7",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Fixes a GPU memory leak in `Bnb4bitQuantize.convert()` where float16 source tensors are never freed during 4-bit quantized model loading via `from_pretrained`, causing OOM on models whose float16 size exceeds GPU VR\u2026",
-    "changed_files": 2,
+    "additions": 6,
+    "author": "JiwaniZakir",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fixes #44737 `XLNetModel.relative_positional_encoding` was creating all `torch.arange` tensors on CPU by default, then calling `.to(output_h.device)` at the call site to move them. Adds a `device` parameter to `relative_positional_encoding\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44728",
-    "created_at": "2026-03-15T19:56:44Z",
-    "deletions": 1,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44782",
+    "created_at": "2026-03-17T05:11:36Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44728/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44728",
+    "files_url": "https://github.com/huggingface/transformers/pull/44782/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44782",
     "labels": [],
-    "merged": false,
-    "number": 44728,
+    "merged": true,
+    "number": 44782,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix float16 memory leak during 4-bit quantized model loading",
-    "updated_at": "2026-03-16T20:53:54Z"
+    "title": "fix: XLNet: relative_positional_encoding computes on CPU every forward",
+    "updated_at": "2026-03-19T13:30:48Z"
   },
   {
-    "additions": 202,
-    "author": "LincolnBurrows2017",
+    "additions": 5,
+    "author": "bensons",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixed issue where kwargs like force_download, proxies, token were not being passed to cached_file function.",
-    "changed_files": 11,
+    "body_excerpt": "# What does this PR do? Some model repos provide `extra_special_tokens` as a list in their tokenizer_config.json, which caused an `AttributeError: 'list' object has no attribute 'keys'`. This converts list inputs to a dict mapping each tok\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44727",
-    "created_at": "2026-03-15T19:41:24Z",
-    "deletions": 33,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44781",
+    "created_at": "2026-03-17T04:59:02Z",
+    "deletions": 2849,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44727/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44727",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44781/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44781",
+    "labels": [],
     "merged": false,
-    "number": 44727,
+    "number": 44781,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: AutoProcessor.from_pretrained not passing kwargs to cached_file",
-    "updated_at": "2026-03-18T13:15:46Z"
+    "state": "open",
+    "title": "Fix `_set_model_specific_special_tokens` to accept list-format `extra_special_tokens`",
+    "updated_at": "2026-03-27T23:19:21Z"
   },
   {
-    "additions": 198,
+    "additions": 145,
     "author": "LincolnBurrows2017",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Replaced bare except clause with except Exception in _safe_convert_tensor function to follow Python best practices (PEP 8).",
-    "changed_files": 10,
+    "body_excerpt": "Fixed logic error in is_tiktoken_available function. The original code `return with_blobfile and _is_package_available(\"blobfile\")[0] or True` would always return True due to operator precedence.",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44725",
-    "created_at": "2026-03-15T17:41:18Z",
-    "deletions": 29,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44778",
+    "created_at": "2026-03-16T23:41:29Z",
+    "deletions": 28,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44725/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44725",
+    "files_url": "https://github.com/huggingface/transformers/pull/44778/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44778",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44725,
+    "number": 44778,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: replace bare except with Exception in Fuyu image processing",
-    "updated_at": "2026-03-18T13:16:22Z"
+    "title": "fix: correct logic error in is_tiktoken_available function",
+    "updated_at": "2026-03-18T13:15:37Z"
   },
   {
-    "additions": 6,
-    "author": "ydshieh",
+    "additions": 35,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? TO be explained.",
-    "changed_files": 5,
+    "body_excerpt": "adds docs for #43705 (enable bidirectional attention for decoder-only models)",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44724",
-    "created_at": "2026-03-15T17:14:12Z",
-    "deletions": 5,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44724/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44724",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44777",
+    "created_at": "2026-03-16T21:58:40Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44777/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44777",
     "labels": [],
-    "merged": false,
-    "number": 44724,
+    "merged": true,
+    "number": 44777,
     "review_comments_count": 1,
-    "state": "open",
-    "title": "Fix some missing / incorrect entries in auto files",
-    "updated_at": "2026-03-16T09:59:56Z"
+    "state": "closed",
+    "title": "[docs] is_causal feature",
+    "updated_at": "2026-03-17T19:50:43Z"
   },
   {
-    "additions": 12,
-    "author": "aashirpersonal",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary This PR fixes #44716 by exposing and forwarding `interpolate_pos_encoding` through the Pixio embedding/model call chain so the option is actually usable from `PixioModel.forward()`. ### Changes - Added `interpolate_pos_encoding:\u2026",
-    "changed_files": 2,
+    "additions": 0,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "the doc-builder is breaking because it can't find `Mistral4ForQuestionAnswering`, which looks like it doesn't exist",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44723",
-    "created_at": "2026-03-15T16:52:03Z",
-    "deletions": 6,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44723/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44723",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44723,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix: propagate interpolate_pos_encoding through PixioEmbeddings and PixioModel",
-    "updated_at": "2026-03-18T15:05:52Z"
-  },
-  {
-    "additions": 38,
-    "author": "chandan11248",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Migrates the GPT-J model to use the new `@capture_outputs` and `@can_return_tuple` decorators for standardized output collection, as described in #43979. ### Changes - Added `_can_record_outputs` to `GPTJPreTrained\u2026",
-    "changed_files": 2,
-    "cluster_id": "cluster-43979-11",
-    "cluster_ids": [
-      "cluster-43979-11"
-    ],
-    "cluster_role": "member",
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44722",
-    "created_at": "2026-03-15T15:33:25Z",
-    "deletions": 110,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44776",
+    "created_at": "2026-03-16T20:43:33Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44722/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44722",
+    "files_url": "https://github.com/huggingface/transformers/pull/44776/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44776",
     "labels": [],
-    "merged": false,
-    "number": 44722,
+    "merged": true,
+    "number": 44776,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Refactor gptj output tracing to use standardized decorators",
-    "updated_at": "2026-03-19T18:12:59Z"
-  },
-  {
-    "additions": 4,
-    "author": "rsmed31",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44716 `PixioPatchEmbeddings.forward` already accepted `interpolate_pos_encoding` but it was silently dropped \u2014 never passed from `PixioEmbeddings.forward` or `PixioModel.forward`, making the parameter effectively unusable\u2026",
-    "changed_files": 1,
+    "state": "closed",
+    "title": "[fix] mistral 4 docs",
+    "updated_at": "2026-03-16T21:11:29Z"
+  },
+  {
+    "additions": 177,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "refactors the current [Parallelism methods](https://huggingface.co/docs/transformers/main/en/perf_train_gpu_many#zero-data-parallelism-pipeline-parallelism-and-model-parallelism-3d-parallelism) doc to: - focus on practical examples of comb\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44718",
-    "created_at": "2026-03-14T23:57:14Z",
-    "deletions": 3,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44718/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44718",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44775",
+    "created_at": "2026-03-16T20:23:29Z",
+    "deletions": 109,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44775/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44775",
     "labels": [],
     "merged": false,
-    "number": 44718,
+    "number": 44775,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix: propagate interpolate_pos_encoding through PixioEmbeddings and PixioModel",
-    "updated_at": "2026-03-15T17:58:58Z"
+    "state": "open",
+    "title": "[docs] n-d parallelism",
+    "updated_at": "2026-03-16T20:28:48Z"
   },
   {
-    "additions": 15,
+    "additions": 0,
     "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As discussed internally, some component model classes didn't specify the correct config classes. This PR fixes them (those I could found - because the tiny model creation script fails due to those mistakes).",
-    "changed_files": 7,
+    "body_excerpt": "# What does this PR do? Remove `is_causal` from `EuroBertConfig`",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44715",
-    "created_at": "2026-03-14T21:11:52Z",
-    "deletions": 2,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44774",
+    "created_at": "2026-03-16T18:56:19Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44715/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44715",
+    "files_url": "https://github.com/huggingface/transformers/pull/44774/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44774",
     "labels": [],
     "merged": true,
-    "number": 44715,
+    "number": 44774,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix missing / incorrect `config` class in some model class definitions",
-    "updated_at": "2026-03-15T11:19:51Z"
+    "title": "Remove `is_causal` from `EuroBertConfig`",
+    "updated_at": "2026-03-17T09:33:21Z"
   },
   {
-    "additions": 181,
-    "author": "LincolnBurrows2017",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes issue #44625: Qwen3.5 num_labels not propagating from core config to text_config. When calling `AutoConfig.from_pretrained(\"Qwen3.5\", num_labels=1)`, the main config gets `num_labels=1` but `text_config` still has default\u2026",
-    "changed_files": 8,
+    "additions": 3,
+    "author": "githubnemo",
+    "author_association": "MEMBER",
+    "body_excerpt": "The links to the quantization offloading were outdated and 4-bit quantization also supports offloading which should be mentioned. cc @SunMarc",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44714",
-    "created_at": "2026-03-14T20:42:46Z",
-    "deletions": 26,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44772",
+    "created_at": "2026-03-16T18:46:13Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44714/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44714",
+    "files_url": "https://github.com/huggingface/transformers/pull/44772/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44772",
     "labels": [],
     "merged": false,
-    "number": 44714,
+    "number": 44772,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: propagate num_labels to text_config for Qwen models",
-    "updated_at": "2026-03-18T12:56:27Z"
+    "state": "open",
+    "title": "bitsandbytes: Update links and docs",
+    "updated_at": "2026-03-17T15:57:56Z"
   },
   {
-    "additions": 15,
-    "author": "kulkarni-rohan",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Applies the output tracing refactor to ColQwen2ForRetrieval as part of the broader effort tracked in issue #43979 to modernize output handling across all models in the library. Changes in both modular_colqwen2.py and modeling_colqwen2.py:\u2026",
-    "changed_files": 2,
+    "additions": 2,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? wtf",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44713",
-    "created_at": "2026-03-14T20:20:14Z",
-    "deletions": 28,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44713/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44713",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44771",
+    "created_at": "2026-03-16T18:45:11Z",
+    "deletions": 1,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44771/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44771",
     "labels": [],
     "merged": false,
-    "number": 44713,
+    "number": 44771,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[ColQwen2] Refactor output tracing (issue #43979)",
-    "updated_at": "2026-03-14T20:21:24Z"
+    "title": "wtf",
+    "updated_at": "2026-03-16T18:56:00Z"
   },
   {
-    "additions": 2,
-    "author": "ydshieh",
+    "additions": 203,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? torch 2.11 is going to be released soon, but we still use 2.9. Let's update it to 2.10 so at least a run with torch 2.10, before we update to torch 2.11 later.",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Fix tests failing because of `strict` type validation and decorate two missing configs, Nemotron and VibeVoice",
+    "changed_files": 12,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44712",
-    "created_at": "2026-03-14T20:18:01Z",
-    "deletions": 2,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44770",
+    "created_at": "2026-03-16T18:44:03Z",
+    "deletions": 268,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44712/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44712",
+    "files_url": "https://github.com/huggingface/transformers/pull/44770/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44770",
     "labels": [],
     "merged": true,
-    "number": 44712,
-    "review_comments_count": 0,
+    "number": 44770,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Update Nvidia CI docker file to use torch 2.10",
-    "updated_at": "2026-03-14T20:29:04Z"
+    "title": "Fix configs with `@strict`",
+    "updated_at": "2026-03-17T15:39:43Z"
   },
   {
-    "additions": 339,
-    "author": "anuq",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes #35141. When `tie_word_embeddings=False`, calling `resize_token_embeddings()` creates a new `nn.Linear` for the LM head via `_get_resized_lm_head()`. The new module's weight and bias tensors do **not** carry\u2026",
-    "changed_files": 4,
+    "additions": 145,
+    "author": "LincolnBurrows2017",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary The `is_batched_video()` and `convert_pil_frames_to_video()` functions in `src/transformers/video_utils.py` were accessing `videos[0]` without first checking if the list is empty, causing `IndexError` when empty lists are passed\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44711",
-    "created_at": "2026-03-14T19:21:21Z",
-    "deletions": 205,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44769",
+    "created_at": "2026-03-16T18:40:07Z",
+    "deletions": 28,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44711/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44711",
+    "files_url": "https://github.com/huggingface/transformers/pull/44769/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44769",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44711,
+    "number": 44769,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: mark new lm_head params as `_is_hf_initialized` after `resize_token_embeddings`",
-    "updated_at": "2026-03-20T13:36:58Z"
+    "title": "Fix: Handle empty lists in video_utils functions",
+    "updated_at": "2026-03-18T13:15:55Z"
   },
   {
-    "additions": 12,
-    "author": "he-yufeng",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes `AutoProcessor.from_pretrained` silently dropping hub kwargs like `force_download`, `cache_dir`, `token`, `revision`, etc. ### The bug The existing code on line ~300 filters kwargs using `inspect.signature(ca\u2026",
+    "additions": 20,
+    "author": "michaelbenayoun",
+    "author_association": "MEMBER",
+    "body_excerpt": "The function `add_tensor_parallel_hooks_to_module` has unused parameters, in this PR we: - Remove `tp_plan`, which is not used. - Remove `parameter_name` which is not used - Remove `layer_name`. This parameter is only used for logging purp\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44710",
-    "created_at": "2026-03-14T18:33:53Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44768",
+    "created_at": "2026-03-16T18:29:52Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44710/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44710",
+    "files_url": "https://github.com/huggingface/transformers/pull/44768/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44768",
     "labels": [],
     "merged": true,
-    "number": 44710,
-    "review_comments_count": 0,
+    "number": 44768,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Fix AutoProcessor.from_pretrained silently dropping hub kwargs",
-    "updated_at": "2026-03-25T18:13:14Z"
+    "title": "Remove unused parameters and improve add_tensor_parallel_hooks_t\u2026",
+    "updated_at": "2026-04-09T17:11:55Z"
   },
   {
-    "additions": 6778,
-    "author": "LucasMa2025",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "# \ud83c\udf9b\ufe0f Add Configurable Generation Scheduler and State Machine for `generate()` ## Summary This PR introduces a **fully optional, zero-intrusion** Generation Scheduler (`GenerationScheduler`) and explicit state machine (`GenerationStateMachi\u2026",
-    "changed_files": 15,
+    "additions": 11,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? EuroBertConfig was missing `@strict(accept_kwargs=True)` unlike its parent LlamaConfig, causing failures when reloading saved configs that include extra keys like `architectures`. Also fixed the test helper passing\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44708",
-    "created_at": "2026-03-14T17:13:34Z",
-    "deletions": 7,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44708/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44708",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44767",
+    "created_at": "2026-03-16T17:31:26Z",
+    "deletions": 5,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44767/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44767",
     "labels": [],
-    "merged": false,
-    "number": 44708,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44767,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "Add Configurable Generation Scheduler and State Machine for `generate()`",
-    "updated_at": "2026-03-14T19:19:11Z"
+    "title": "Fix: Eurobert model was missing @strict decorator and invalid test kwargs",
+    "updated_at": "2026-03-16T19:02:31Z"
   },
   {
-    "additions": 3,
-    "author": "saivedant169",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes part of #32937 ## What does this PR do? Adds `position_ids` as an explicit parameter to `MptForCausalLM.forward()` and `MptModel.forward()`, bringing MPT in line with other CausalLM models. Same rationale as the Bloom PR (#44706) \u2014 M\u2026",
-    "changed_files": 1,
+    "additions": 26,
+    "author": "itazap",
+    "author_association": "MEMBER",
+    "body_excerpt": "for when remote code tries to import from `tokenization_xxx_fast`",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44707",
-    "created_at": "2026-03-14T17:12:16Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44707/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44707",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44707,
+    "cluster_role": null,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44766",
+    "created_at": "2026-03-16T17:30:23Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44766/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44766",
+    "labels": [],
+    "merged": true,
+    "number": 44766,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add position_ids to MptForCausalLM forward pass",
-    "updated_at": "2026-03-18T13:39:36Z"
+    "title": "support xxxFast alias in v5 tokenizers",
+    "updated_at": "2026-03-18T13:40:05Z"
   },
   {
-    "additions": 3,
-    "author": "saivedant169",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes part of #32937 ## What does this PR do? Adds `position_ids` as an explicit parameter to `BloomForCausalLM.forward()` and `BloomModel.forward()`, bringing Bloom in line with other CausalLM models like Llama, Falcon, Gemma, and Mistral\u2026",
-    "changed_files": 1,
+    "additions": 19,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? The following failing tests were identified and fixed in this PR: \u2192 **PaliGemma 2:** The [PaliGemma 1 test class](https://github.com/huggingface/transformers/blob/main/tests/models/paligemma/test_modeling_paligemm\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44706",
-    "created_at": "2026-03-14T17:09:11Z",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44765",
+    "created_at": "2026-03-16T17:26:22Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44706/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44706",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44706,
+    "files_url": "https://github.com/huggingface/transformers/pull/44765/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44765",
+    "labels": [],
+    "merged": true,
+    "number": 44765,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add position_ids to BloomForCausalLM forward pass",
-    "updated_at": "2026-03-18T13:39:51Z"
+    "title": "fix(testing): Fix PaliGemma 2 and PaddleOCR-VL test failures on main",
+    "updated_at": "2026-03-20T13:55:55Z"
   },
   {
-    "additions": 14,
-    "author": "saivedant169",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes part of #32937 ## What does this PR do? RoFormer introduced rotary position embeddings, but its `ForCausalLM` forward method doesn't accept `position_ids` \u2014 which means callers can't specify custom positions for packed sequences or f\u2026",
-    "changed_files": 1,
+    "additions": 12,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes the siglip import. that was also crashing the test fetcher",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44705",
-    "created_at": "2026-03-14T16:48:06Z",
-    "deletions": 1,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44764",
+    "created_at": "2026-03-16T17:15:40Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44705/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44705",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44705,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44764/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44764",
+    "labels": [],
+    "merged": true,
+    "number": 44764,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Add position_ids to RoFormerForCausalLM forward pass",
-    "updated_at": "2026-03-18T13:40:05Z"
+    "title": "fix: sig lip import",
+    "updated_at": "2026-03-16T17:38:41Z"
   },
   {
-    "additions": 26,
-    "author": "vasqu",
+    "additions": 17,
+    "author": "xenova",
     "author_association": "MEMBER",
-    "body_excerpt": "As per title, it seems that the `cute` subfolder can be even distributed if you only install FA2 which implies something wrong. Now we check under the (normalized) distribution names",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Adds support for MLP mixers, used by [nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16). Previously, it would crash because it would not recognize the `-` char in t\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44703",
-    "created_at": "2026-03-14T14:46:02Z",
-    "deletions": 10,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44763",
+    "created_at": "2026-03-16T17:04:36Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44703/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44703",
+    "files_url": "https://github.com/huggingface/transformers/pull/44763/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44763",
     "labels": [],
-    "merged": true,
-    "number": 44703,
+    "merged": false,
+    "number": 44763,
     "review_comments_count": 1,
     "state": "closed",
-    "title": "[`FA`] Fix fa detection",
-    "updated_at": "2026-03-14T17:19:07Z"
+    "title": "[nemotron_h] Add support for MLP mixers",
+    "updated_at": "2026-04-14T13:46:14Z"
   },
   {
-    "additions": 148,
-    "author": "LincolnBurrows2017",
+    "additions": 4,
+    "author": "BillionClaw",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR fix? The `rms_norm_eps` parameter in `MistralConfig` was incorrectly typed as `int | None` but defaults to `1e-6` which is a float. This parameter is passed to `MistralRMSNorm` which expects `eps: float`. ### Bug Detai\u2026",
-    "changed_files": 8,
+    "body_excerpt": "XLNet.relative_positional_encoding creates intermediate tensors on CPU every forward pass because torch.arange was missing the device parameter. This causes unnecessary CPU-GPU transfers when running on CUDA. Added device=self.device to al\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44702",
-    "created_at": "2026-03-14T14:41:15Z",
-    "deletions": 25,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44762",
+    "created_at": "2026-03-16T16:17:54Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44702/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44702",
+    "files_url": "https://github.com/huggingface/transformers/pull/44762/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44762",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44702,
+    "number": 44762,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: Correct rms_norm_eps type hint from int to float in MistralConfig",
-    "updated_at": "2026-03-18T13:00:12Z"
+    "title": "fix: Cache XLNet relative_positional_encoding to avoid CPU computation",
+    "updated_at": "2026-03-18T15:16:14Z"
   },
   {
-    "additions": 219,
-    "author": "hmellor",
+    "additions": 152,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "These models have `base_model_pp_plan`s but currently do not work because the base model's forward pass depends on all the `layers` being `Qwen2VLDecoderLayer`. i.e. if one of the layers is removed/replaced with `Identity`, `decoder_layer.\u2026",
-    "changed_files": 52,
+    "body_excerpt": "# What does this PR do? This adds rule 10: ``` Direct config definitions must use @strict(accept_kwargs=True). ```",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44699",
-    "created_at": "2026-03-14T11:44:24Z",
-    "deletions": 148,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44761",
+    "created_at": "2026-03-16T16:05:03Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44699/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44699",
+    "files_url": "https://github.com/huggingface/transformers/pull/44761/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44761",
     "labels": [],
     "merged": true,
-    "number": 44699,
-    "review_comments_count": 0,
+    "number": 44761,
+    "review_comments_count": 7,
     "state": "closed",
-    "title": "Fix several based models' pipeline parallel support",
-    "updated_at": "2026-03-20T13:53:27Z"
+    "title": "model-linter: Added rule 10",
+    "updated_at": "2026-03-17T08:52:19Z"
   },
   {
-    "additions": 1,
-    "author": "hmellor",
-    "author_association": "MEMBER",
-    "body_excerpt": "The typo in the `elif` chain meant that `image` and `video` modalidty encoders could not be set using this method. This PR fixes the typo so that they can.",
-    "changed_files": 1,
+    "additions": 2090,
+    "author": "juliendenize",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44698",
-    "created_at": "2026-03-14T11:18:54Z",
-    "deletions": 1,
+    "comments_count": 12,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44760",
+    "created_at": "2026-03-16T15:54:11Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44698/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44698",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44760/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44760",
+    "labels": [
+      "New model"
+    ],
     "merged": true,
-    "number": 44698,
-    "review_comments_count": 0,
+    "number": 44760,
+    "review_comments_count": 8,
     "state": "closed",
-    "title": "Fix `set_encoder`",
-    "updated_at": "2026-03-14T13:42:00Z"
+    "title": "Add Mistral 4",
+    "updated_at": "2026-03-20T10:44:48Z"
   },
   {
-    "additions": 75,
-    "author": "LincolnBurrows2017",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Description The `torch_float` function in `src/transformers/utils/generic.py` was incorrectly returning `int(x)` in two places where it should return `float(x)`: 1. When torch is not available (fallback case) 2. When not in a tracing co\u2026",
-    "changed_files": 4,
+    "additions": 419,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Follow-up of many related PR, last one in time being https://github.com/huggingface/transformers/pull/44602. This PR completes all the models that may need non-trivial treatment. Only about 30-40 models still have m\u2026",
+    "changed_files": 42,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44697",
-    "created_at": "2026-03-14T10:44:12Z",
-    "deletions": 25,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44759",
+    "created_at": "2026-03-16T15:38:13Z",
+    "deletions": 983,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44697/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44697",
+    "files_url": "https://github.com/huggingface/transformers/pull/44759/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44759",
     "labels": [],
-    "merged": false,
-    "number": 44697,
-    "review_comments_count": 1,
-    "state": "open",
-    "title": "fix: torch_float should return float, not int",
-    "updated_at": "2026-03-17T19:29:02Z"
+    "merged": true,
+    "number": 44759,
+    "review_comments_count": 26,
+    "state": "closed",
+    "title": "Remove cache_position in more models (3)",
+    "updated_at": "2026-03-18T13:09:37Z"
   },
   {
-    "additions": 19,
-    "author": "hmellor",
+    "additions": 825,
+    "author": "LysandreJik",
     "author_association": "MEMBER",
-    "body_excerpt": "In configs, `base_model_pp_plan` and `base_model_tp_plan` default to `None` In models, `_pp_plan` and `_tp_plan` _look like_ they default to `None` based on the class variables, but will actually always be a dict because of `post_init`. Th\u2026",
-    "changed_files": 1,
+    "body_excerpt": "Provides a nicer feedback when `transformers chat` loads a model, instead of hanging https://github.com/user-attachments/assets/8f68f914-b702-4430-b97f-e8cc25326b70 <p>Adds a <code>POST /load_model</code> endpoint to <code>transformers ser\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44696",
-    "created_at": "2026-03-14T09:41:07Z",
-    "deletions": 13,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44758",
+    "created_at": "2026-03-16T15:02:15Z",
+    "deletions": 63,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44696/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44696",
+    "files_url": "https://github.com/huggingface/transformers/pull/44758/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44758",
     "labels": [],
     "merged": true,
-    "number": 44696,
-    "review_comments_count": 5,
+    "number": 44758,
+    "review_comments_count": 20,
     "state": "closed",
-    "title": "Fix `supports_{tp/pp}_plan`",
-    "updated_at": "2026-03-31T13:12:56Z"
+    "title": "Propagate the model loading from transformers serve to chat",
+    "updated_at": "2026-03-19T17:20:03Z"
   },
   {
-    "additions": 4,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following failing tests were identified and fixed in this PR: \u2192 **Kyutai Speech-To-Text**: [The PR [processors] Unbloating simple processors](https://github.com/huggingface/transformers/pull/40377), [refactore\u2026",
-    "changed_files": 2,
+    "additions": 1,
+    "author": "dacorvo",
+    "author_association": "MEMBER",
+    "body_excerpt": "## Summary - `_valid_auto_compile_criteria()` gates auto-compilation on `device.type in [\"cuda\", \"xpu\"]`, excluding Neuron devices. This means `torch.compile` never triggers automatically on Neuron even when `StaticCache` is used (which se\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44695",
-    "created_at": "2026-03-14T09:05:35Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44757",
+    "created_at": "2026-03-16T14:54:38Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44695/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44695",
+    "files_url": "https://github.com/huggingface/transformers/pull/44757/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44757",
     "labels": [],
     "merged": true,
-    "number": 44695,
-    "review_comments_count": 3,
+    "number": 44757,
+    "review_comments_count": 10,
     "state": "closed",
-    "title": "fix(testing): Fix Kyutai Speech-To-Text and LongCatFlash test failures on main CI",
-    "updated_at": "2026-04-09T15:41:05Z"
+    "title": "Add Neuron to auto-compile hardware list",
+    "updated_at": "2026-04-16T13:47:59Z"
   },
   {
-    "additions": 143,
-    "author": "LincolnBurrows2017",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes issue #44625: Qwen3.5 num_labels not propagated from core config to text config. When loading `AutoConfig.from_pretrained(\"Qwen3.5\", num_labels=1)`, the outer config gets `num_labels=1` but the inner `text_config` still ha\u2026",
-    "changed_files": 7,
+    "additions": 4,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44693",
-    "created_at": "2026-03-14T05:43:00Z",
-    "deletions": 30,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44755",
+    "created_at": "2026-03-16T14:08:34Z",
+    "deletions": 148,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44693/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44693",
+    "files_url": "https://github.com/huggingface/transformers/pull/44755/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44755",
     "labels": [],
     "merged": false,
-    "number": 44693,
+    "number": 44755,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: Propagate num_labels to text_config in Qwen3.5",
-    "updated_at": "2026-03-18T12:56:25Z"
+    "title": "Dont merge, testing smth",
+    "updated_at": "2026-03-18T10:09:15Z"
   },
   {
-    "additions": 18,
+    "additions": 20,
     "author": "gambletan",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44514. `Qwen2_5_VLProcessor.apply_chat_template` crashes with `ValueError` when called with batched input and `padding=False` (the default). The root cause is `np.array(text_inputs[\"input_ids\"])` which fails when sequence\u2026",
+    "body_excerpt": "## Summary Fixes #44737 - Added `device=self.device` to all four `torch.arange()` calls in `XLNetModel.relative_positional_encoding()` so that intermediate tensors are created directly on the model's device instead of always on CPU. - With\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44692",
-    "created_at": "2026-03-14T04:14:38Z",
-    "deletions": 10,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44753",
+    "created_at": "2026-03-16T14:01:08Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44692/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44692",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44753/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44753",
+    "labels": [],
     "merged": false,
-    "number": 44692,
+    "number": 44753,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: handle ragged input_ids in Qwen2_5_VLProcessor.apply_chat_template",
-    "updated_at": "2026-03-18T12:44:18Z"
+    "title": "fix: add device= to torch.arange in XLNet relative_positional_encoding",
+    "updated_at": "2026-03-18T13:28:40Z"
   },
   {
-    "additions": 23,
-    "author": "gambletan",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary - Fixes `num_labels` (and `id2label`/`label2id`) not being propagated from the outer `Qwen3_5Config` to its inner `text_config` when passed via `AutoConfig.from_pretrained(..., num_labels=1)`. - When `text_config` is `None` or a\u2026",
-    "changed_files": 2,
+    "additions": 100,
+    "author": "LincolnBurrows2017",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR fix? The `flatten()` function in `tokenization_utils_base.py` had a bug where it was checking `arr[0]` instead of `sub_arr` when determining if an element should be recursively flattened. ### Bug Details - **File**: `s\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44691",
-    "created_at": "2026-03-14T04:10:54Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44751",
+    "created_at": "2026-03-16T13:40:44Z",
+    "deletions": 29,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44691/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44691",
+    "files_url": "https://github.com/huggingface/transformers/pull/44751/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44751",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44691,
+    "number": 44751,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Qwen3.5 num_labels not propagated to text_config",
-    "updated_at": "2026-03-18T12:57:19Z"
+    "title": "Fix: Correct variable reference in flatten() function",
+    "updated_at": "2026-03-18T13:16:12Z"
   },
   {
-    "additions": 6,
-    "author": "gambletan",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44360 The `GlmMoeDsaIndexer` is missing a ReLU activation on the per-head dot-product scores before the weighted sum across heads. The reference DeepSeek V3.2 implementation applies ReLU inside the `fp8_index` kernel: ```\u2026",
-    "changed_files": 2,
+    "additions": 1,
+    "author": "juliendenize",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds `apply_yarn_scaling` as an optional key for yarn repo. This was requested as part of a vLLM PR https://github.com/vllm-project/vllm/pull/37104 that seeks to silence some rope issues when converting Mist\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44690",
-    "created_at": "2026-03-14T03:44:37Z",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44747",
+    "created_at": "2026-03-16T10:32:54Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44690/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44690",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44747/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44747",
+    "labels": [],
     "merged": false,
-    "number": 44690,
+    "number": 44747,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix missing ReLU in GLM-MOE-DSA indexer scoring",
-    "updated_at": "2026-03-18T12:40:23Z"
+    "title": "Add apply_yarn_scaling as optional key to yarn",
+    "updated_at": "2026-03-16T12:48:08Z"
   },
   {
-    "additions": 141,
+    "additions": 202,
     "author": "LincolnBurrows2017",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes issue #44625: Qwen3.5 num_labels not propagating to text_config. When calling `AutoConfig.from_pretrained(\"Qwen3.5\", num_labels=1)`, the main config gets `num_labels=1` but text_config still has default `num_labels=2`. Thi\u2026",
-    "changed_files": 6,
+    "body_excerpt": "Fixed Issue #44737: XLNet relative_positional_encoding function missing device parameter in torch.arange calls.",
+    "changed_files": 11,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44688",
-    "created_at": "2026-03-14T00:40:50Z",
-    "deletions": 23,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44745",
+    "created_at": "2026-03-16T09:39:30Z",
+    "deletions": 33,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44688/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44688",
+    "files_url": "https://github.com/huggingface/transformers/pull/44745/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44745",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44688,
+    "number": 44745,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: Propagate num_labels to text_config in Qwen models",
-    "updated_at": "2026-03-18T12:56:41Z"
+    "title": "fix: add device parameter to torch.arange calls in XLNet",
+    "updated_at": "2026-03-18T13:16:43Z"
   },
   {
-    "additions": 8,
-    "author": "vxa8502",
+    "additions": 35,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": null,
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44739",
+    "created_at": "2026-03-16T07:15:33Z",
+    "deletions": 7,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44739/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44739",
+    "labels": [],
+    "merged": true,
+    "number": 44739,
+    "review_comments_count": 12,
+    "state": "closed",
+    "title": "fix series of failed test case for janus model",
+    "updated_at": "2026-04-01T08:24:26Z"
+  },
+  {
+    "additions": 6,
+    "author": "yunhaoli24",
     "author_association": "NONE",
-    "body_excerpt": "Fixes partial #32937 Adds explicit `position_ids` threading through GPT-Neo's attention layers to enable flash attention's packed sequence optimization. ## Context GPT-Neo uses learned absolute position embeddings (`wpe`) applied at the mo\u2026",
+    "body_excerpt": "## What does this PR do? Fixes #44031 ## The Problem The condition for calling `_patch_mistral_regex` was too broad (`vocab_size > 100000`), causing non-Mistral models like Qwen, LLaMA, BGE-Reranker to show incorrect regex pattern warnings\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44687",
-    "created_at": "2026-03-13T23:28:55Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44736",
+    "created_at": "2026-03-16T06:00:47Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44687/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44687",
+    "files_url": "https://github.com/huggingface/transformers/pull/44736/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44736",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44687,
+    "number": 44736,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add explicit position_ids to GPT-Neo attention layers",
-    "updated_at": "2026-03-18T13:06:49Z"
+    "title": "fix: resolve false-positive regex warning for non-mistral models",
+    "updated_at": "2026-03-18T15:08:59Z"
   },
   {
-    "additions": 615,
-    "author": "tejasae-afk",
-    "author_association": "NONE",
-    "body_excerpt": "During an automated code review of src/transformers/models/marian/convert_marian_to_pytorch.py, the following issue was identified. Use safe_load in convert marian to pytorch. yaml.load on untrusted input can construct arbitrary Python obj\u2026",
-    "changed_files": 80,
+    "additions": 1,
+    "author": "mango766",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes a crash in `transformers serve` when the `/v1/responses` streaming endpoint attempts to reuse a KV cache from a previous request in the same conversation session. ### The bug In `generate_response`, `inputs`\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44735",
+    "created_at": "2026-03-16T04:09:32Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44735/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44735",
+    "labels": [],
+    "merged": true,
+    "number": 44735,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix tensor indexing crash in serve generate_response KV cache continuation",
+    "updated_at": "2026-03-16T15:27:59Z"
+  },
+  {
+    "additions": 28,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@IlyasMoutawwakil, pls help review, thx!",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44686",
-    "created_at": "2026-03-13T21:22:07Z",
-    "deletions": 259,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44733",
+    "created_at": "2026-03-16T02:55:54Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44686/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44686",
+    "files_url": "https://github.com/huggingface/transformers/pull/44733/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44733",
     "labels": [],
-    "merged": false,
-    "number": 44686,
+    "merged": true,
+    "number": 44733,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Use safe_load in convert marian to pytorch",
-    "updated_at": "2026-03-14T03:54:31Z"
+    "title": "Fix bug and add XPU Expectations for qwen2 and jamba tests",
+    "updated_at": "2026-04-01T08:24:40Z"
   },
   {
-    "additions": 10,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? For tiny model creation script - new added model test files still miss this argument ...",
-    "changed_files": 3,
+    "additions": 1,
+    "author": "Defalt-Meh",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? While running SmolVLM tests I noticed this warning in the output: ``` tests/test_video_processing_common.py:57: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider convert\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44685",
-    "created_at": "2026-03-13T20:53:41Z",
-    "deletions": 3,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44731",
+    "created_at": "2026-03-15T23:26:31Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44685/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44685",
+    "files_url": "https://github.com/huggingface/transformers/pull/44731/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44731",
     "labels": [],
-    "merged": true,
-    "number": 44685,
+    "merged": false,
+    "number": 44731,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix more model tester missing `parent` issue",
-    "updated_at": "2026-03-13T21:03:46Z"
+    "state": "open",
+    "title": "[Tests] Fix slow video tensor creation from list of numpy arrays in SmolVLM",
+    "updated_at": "2026-03-15T23:26:31Z"
   },
   {
-    "additions": 41,
-    "author": "ntenenz",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "\u2026 # What does this PR do? In torch versions >= 2.9.0, it requests the lse from flex_attenetion using `AuxRequest` instead of the deprecated `return_lse`, which triggers a warning and can break tracing. Fixes #44683 ## Before submitting - [\u2026",
-    "changed_files": 1,
+    "additions": 9,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? It's unclear why the config class has `model_type = \"mlcd_vision_model\"` but the model on the hub has \"model_type\": \"mlcd\". This leads to the following failures (load from hub --> save locally --> local locally) ```\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44684",
-    "created_at": "2026-03-13T20:16:35Z",
-    "deletions": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44730",
+    "created_at": "2026-03-15T20:44:32Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44684/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44684",
+    "files_url": "https://github.com/huggingface/transformers/pull/44730/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44730",
     "labels": [],
     "merged": true,
-    "number": 44684,
-    "review_comments_count": 8,
+    "number": 44730,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "update flex attention to use `return_aux` instead of `return_lse` when torch verison >= 2.9",
-    "updated_at": "2026-03-18T11:44:18Z"
+    "title": "Fix `mlcd` auto config/model/mapping issues",
+    "updated_at": "2026-03-16T12:12:30Z"
   },
   {
-    "additions": 301,
-    "author": "SunMarc",
+    "additions": 214,
+    "author": "xenova",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Llama cpp integration in transformers serve. Minor changes to add llama.cpp integration Mostly changes on serve to fix latency for streaming and non streaming",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? This PR introduces a helper utility function, `int_div_ceil`, which performs `math.ceil(a / b)` for non-negative integer operands. This is necessary as the current approach is both error-prone and imprecise (especia\u2026",
+    "changed_files": 58,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44682",
-    "created_at": "2026-03-13T18:52:41Z",
-    "deletions": 73,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44729",
+    "created_at": "2026-03-15T20:29:38Z",
+    "deletions": 225,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44682/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44682",
+    "files_url": "https://github.com/huggingface/transformers/pull/44729/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44729",
     "labels": [],
     "merged": false,
-    "number": 44682,
+    "number": 44729,
     "review_comments_count": 0,
     "state": "open",
-    "title": "transformers serve + llamacpp",
-    "updated_at": "2026-03-14T07:05:29Z"
+    "title": "Avoid floating point math for ceil operations",
+    "updated_at": "2026-03-15T20:49:34Z"
   },
   {
-    "additions": 47,
-    "author": "dacorvo",
-    "author_association": "MEMBER",
-    "body_excerpt": "Fixes #44679 ## Summary - Custom attention kernels registered via `load_and_register_attn_kernel` currently get hardcoded `flash_attention_2` mask dispatch, which produces 2D or `None` masks - Kernels that need SDPA-style 4D boolean masks\u2026",
+    "additions": 88,
+    "author": "ajmeese7",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Fixes a GPU memory leak in `Bnb4bitQuantize.convert()` where float16 source tensors are never freed during 4-bit quantized model loading via `from_pretrained`, causing OOM on models whose float16 size exceeds GPU VR\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44680",
-    "created_at": "2026-03-13T17:55:54Z",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44728",
+    "created_at": "2026-03-15T19:56:44Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44680/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44680",
+    "files_url": "https://github.com/huggingface/transformers/pull/44728/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44728",
     "labels": [],
     "merged": false,
-    "number": 44680,
-    "review_comments_count": 12,
-    "state": "open",
-    "title": "Allow kernel modules to declare their preferred mask function",
-    "updated_at": "2026-04-14T19:29:06Z"
+    "number": 44728,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix float16 memory leak during 4-bit quantized model loading",
+    "updated_at": "2026-03-16T20:53:54Z"
   },
   {
-    "additions": 9,
-    "author": "JokeYoonic",
+    "additions": 202,
+    "author": "LincolnBurrows2017",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Problem: - On macOS ARM64 + Python 3.13 + transformers 5.x, GPT-2 model's lm_head forward pass produces NaN/Inf values during inference - Root cause: lm_head.weight is tied to transformer.wte.weight, and the shared memory reference causes\u2026",
-    "changed_files": 1,
+    "body_excerpt": "Fixed issue where kwargs like force_download, proxies, token were not being passed to cached_file function.",
+    "changed_files": 11,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44676",
-    "created_at": "2026-03-13T16:28:01Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44727",
+    "created_at": "2026-03-15T19:41:24Z",
+    "deletions": 33,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44676/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44676",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44727/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44727",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44676,
+    "number": 44727,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "fix(gpt2): Resolve NaN/Inf issue in lm_head on Python 3.13 with tied weights",
-    "updated_at": "2026-03-18T17:16:49Z"
+    "state": "closed",
+    "title": "fix: AutoProcessor.from_pretrained not passing kwargs to cached_file",
+    "updated_at": "2026-03-18T13:15:46Z"
   },
   {
-    "additions": 32,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "properly formats the `ContinuousBatchingConfig` below: <img width=\"976\" height=\"626\" alt=\"Screenshot 2026-03-13 at 9 09 39 AM\" src=\"https://github.com/user-attachments/assets/4390c6f7-bb63-4039-a46e-9f4ae23f5d98\" />",
-    "changed_files": 1,
+    "additions": 198,
+    "author": "LincolnBurrows2017",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Replaced bare except clause with except Exception in _safe_convert_tensor function to follow Python best practices (PEP 8).",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44675",
-    "created_at": "2026-03-13T16:10:28Z",
-    "deletions": 14,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44725",
+    "created_at": "2026-03-15T17:41:18Z",
+    "deletions": 29,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44675/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44675",
-    "labels": [],
-    "merged": true,
-    "number": 44675,
+    "files_url": "https://github.com/huggingface/transformers/pull/44725/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44725",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44725,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] cb config",
-    "updated_at": "2026-03-13T23:15:04Z"
+    "title": "fix: replace bare except with Exception in Fuyu image processing",
+    "updated_at": "2026-03-18T13:16:22Z"
   },
   {
-    "additions": 408,
-    "author": "Rocketknight1",
+    "additions": 6,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "We've had `parse_response()` in the library for a while, but it's been a soft launch / prototype feature. This PR cleans it up and documents it, making it an official feature! The API is largely unchanged from the prototype, but we drop `x\u2026",
+    "body_excerpt": "# What does this PR do? TO be explained.",
     "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44674",
-    "created_at": "2026-03-13T15:41:42Z",
-    "deletions": 34,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44724",
+    "created_at": "2026-03-15T17:14:12Z",
+    "deletions": 5,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44724/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44724",
+    "labels": [],
+    "merged": false,
+    "number": 44724,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "Fix some missing / incorrect entries in auto files",
+    "updated_at": "2026-03-16T09:59:56Z"
+  },
+  {
+    "additions": 12,
+    "author": "aashirpersonal",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary This PR fixes #44716 by exposing and forwarding `interpolate_pos_encoding` through the Pixio embedding/model call chain so the option is actually usable from `PixioModel.forward()`. ### Changes - Added `interpolate_pos_encoding:\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44723",
+    "created_at": "2026-03-15T16:52:03Z",
+    "deletions": 6,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44723/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44723",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44723,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix: propagate interpolate_pos_encoding through PixioEmbeddings and PixioModel",
+    "updated_at": "2026-03-18T15:05:52Z"
+  },
+  {
+    "additions": 38,
+    "author": "chandan11248",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Migrates the GPT-J model to use the new `@capture_outputs` and `@can_return_tuple` decorators for standardized output collection, as described in #43979. ### Changes - Added `_can_record_outputs` to `GPTJPreTrained\u2026",
+    "changed_files": 2,
+    "cluster_id": "cluster-43979-11",
+    "cluster_ids": [
+      "cluster-43979-11"
+    ],
+    "cluster_role": "member",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44722",
+    "created_at": "2026-03-15T15:33:25Z",
+    "deletions": 110,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44674/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44674",
+    "files_url": "https://github.com/huggingface/transformers/pull/44722/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44722",
     "labels": [],
-    "merged": true,
-    "number": 44674,
-    "review_comments_count": 11,
-    "state": "closed",
-    "title": "Officially launch parse_response",
-    "updated_at": "2026-03-24T15:55:05Z"
+    "merged": false,
+    "number": 44722,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Refactor gptj output tracing to use standardized decorators",
+    "updated_at": "2026-03-19T18:12:59Z"
   },
   {
-    "additions": 73,
-    "author": "remi-or",
-    "author_association": "MEMBER",
-    "body_excerpt": "This PR fixes a bug in continuous batching where non-CUDA devices cannot use the feature because some CUDA-exclusive objects are always instantiated. It also adds a test to make sure this will not break again in the future.",
-    "changed_files": 3,
+    "additions": 4,
+    "author": "rsmed31",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #44716 `PixioPatchEmbeddings.forward` already accepted `interpolate_pos_encoding` but it was silently dropped \u2014 never passed from `PixioEmbeddings.forward` or `PixioModel.forward`, making the parameter effectively unusable\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44673",
-    "created_at": "2026-03-13T15:37:01Z",
-    "deletions": 15,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44718",
+    "created_at": "2026-03-14T23:57:14Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44673/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44673",
+    "files_url": "https://github.com/huggingface/transformers/pull/44718/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44718",
     "labels": [],
-    "merged": true,
-    "number": 44673,
+    "merged": false,
+    "number": 44718,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[CB] [Bug] Fix crashes when running without cuda",
-    "updated_at": "2026-03-15T23:59:55Z"
+    "title": "Fix: propagate interpolate_pos_encoding through PixioEmbeddings and PixioModel",
+    "updated_at": "2026-03-15T17:58:58Z"
   },
   {
-    "additions": 1,
-    "author": "neo",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
+    "additions": 15,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As discussed internally, some component model classes didn't specify the correct config classes. This PR fixes them (those I could found - because the tiny model creation script fails due to those mistakes).",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44672",
-    "created_at": "2026-03-13T15:33:15Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44715",
+    "created_at": "2026-03-14T21:11:52Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44672/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44672",
+    "files_url": "https://github.com/huggingface/transformers/pull/44715/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44715",
     "labels": [],
     "merged": true,
-    "number": 44672,
+    "number": 44715,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix annotations reader for python 3.14 in `PreTrainedModel`",
-    "updated_at": "2026-03-19T13:30:48Z"
+    "title": "Fix missing / incorrect `config` class in some model class definitions",
+    "updated_at": "2026-03-15T11:19:51Z"
   },
   {
-    "additions": 6,
-    "author": "dacorvo",
-    "author_association": "MEMBER",
-    "body_excerpt": "Fixes #44678 ## Summary - Replace advanced indexing (`self.gate_up_proj[expert_ids]`) with explicit `torch.index_select(self.gate_up_proj, 0, expert_ids)` in `batched_mm_experts_forward` - 6 replacements total (3 weight tensors + 3 bias te\u2026",
-    "changed_files": 1,
+    "additions": 181,
+    "author": "LincolnBurrows2017",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes issue #44625: Qwen3.5 num_labels not propagating from core config to text_config. When calling `AutoConfig.from_pretrained(\"Qwen3.5\", num_labels=1)`, the main config gets `num_labels=1` but `text_config` still has default\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44669",
-    "created_at": "2026-03-13T14:52:22Z",
-    "deletions": 6,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44714",
+    "created_at": "2026-03-14T20:42:46Z",
+    "deletions": 26,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44669/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44669",
+    "files_url": "https://github.com/huggingface/transformers/pull/44714/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44714",
     "labels": [],
     "merged": false,
-    "number": 44669,
+    "number": 44714,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Use `index_select` instead of advanced indexing in `batched_mm_experts_forward`",
-    "updated_at": "2026-03-19T13:39:23Z"
+    "title": "fix: propagate num_labels to text_config for Qwen models",
+    "updated_at": "2026-03-18T12:56:27Z"
   },
   {
-    "additions": 18,
-    "author": "dacorvo",
-    "author_association": "MEMBER",
-    "body_excerpt": "Fixes #44677 ## Summary - Add `base_model_tp_plan` to `OlmoeConfig`, enabling `from_pretrained(tp_plan=\"auto\")` for OLMoE models - Add `TensorParallelTesterMixin` to OLMoE tests for TP validation coverage - Uses `\"colwise\"` for `q_norm` an\u2026",
+    "additions": 15,
+    "author": "kulkarni-rohan",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Applies the output tracing refactor to ColQwen2ForRetrieval as part of the broader effort tracked in issue #43979 to modernize output handling across all models in the library. Changes in both modular_colqwen2.py and modeling_colqwen2.py:\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44668",
-    "created_at": "2026-03-13T14:45:22Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44713",
+    "created_at": "2026-03-14T20:20:14Z",
+    "deletions": 28,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44668/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44668",
+    "files_url": "https://github.com/huggingface/transformers/pull/44713/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44713",
     "labels": [],
-    "merged": true,
-    "number": 44668,
-    "review_comments_count": 4,
-    "state": "closed",
-    "title": "Add `base_model_tp_plan` to `OlmoeConfig`",
-    "updated_at": "2026-03-26T13:58:58Z"
+    "merged": false,
+    "number": 44713,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "[ColQwen2] Refactor output tracing (issue #43979)",
+    "updated_at": "2026-03-14T20:21:24Z"
   },
   {
-    "additions": 412,
-    "author": "zucchini-nlp",
+    "additions": 2,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? A few find-and-replaces and cache position is deleted from the rest of models. Still have to check docs and test files, so WIP",
-    "changed_files": 120,
+    "body_excerpt": "# What does this PR do? torch 2.11 is going to be released soon, but we still use 2.9. Let's update it to 2.10 so at least a run with torch 2.10, before we update to torch 2.11 later.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44667",
-    "created_at": "2026-03-13T14:37:26Z",
-    "deletions": 1519,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44712",
+    "created_at": "2026-03-14T20:18:01Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44667/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44667",
+    "files_url": "https://github.com/huggingface/transformers/pull/44712/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44712",
     "labels": [],
-    "merged": false,
-    "number": 44667,
+    "merged": true,
+    "number": 44712,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Goodbye cache position",
-    "updated_at": "2026-03-19T11:55:04Z"
+    "title": "Update Nvidia CI docker file to use torch 2.10",
+    "updated_at": "2026-03-14T20:29:04Z"
   },
   {
-    "additions": 17,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - 2 model tester classes didn't follow the usual way we do things, which cause the tiny model creation script to fail with those model classes. - (the script initializes instances of model testers, in order to call\u2026",
-    "changed_files": 3,
+    "additions": 339,
+    "author": "anuq",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes #35141. When `tie_word_embeddings=False`, calling `resize_token_embeddings()` creates a new `nn.Linear` for the LM head via `_get_resized_lm_head()`. The new module's weight and bias tensors do **not** carry\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44666",
-    "created_at": "2026-03-13T14:24:50Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44711",
+    "created_at": "2026-03-14T19:21:21Z",
+    "deletions": 205,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44666/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44666",
-    "labels": [],
-    "merged": true,
-    "number": 44666,
+    "files_url": "https://github.com/huggingface/transformers/pull/44711/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44711",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44711,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Another (small) set of fixes required for tiny model creation",
-    "updated_at": "2026-03-13T17:20:52Z"
+    "title": "fix: mark new lm_head params as `_is_hf_initialized` after `resize_token_embeddings`",
+    "updated_at": "2026-03-20T13:36:58Z"
   },
   {
-    "additions": 1,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "Dinov3 vit was refactored to introduce a module between top level and layers to have the capture decorators work as intended. Otherwise, it would force the backbone to do manual collection. This introduced a small conversion which is now a\u2026",
+    "additions": 12,
+    "author": "he-yufeng",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes `AutoProcessor.from_pretrained` silently dropping hub kwargs like `force_download`, `cache_dir`, `token`, `revision`, etc. ### The bug The existing code on line ~300 filters kwargs using `inspect.signature(ca\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44665",
-    "created_at": "2026-03-13T13:50:26Z",
-    "deletions": 0,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44710",
+    "created_at": "2026-03-14T18:33:53Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44665/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44665",
+    "files_url": "https://github.com/huggingface/transformers/pull/44710/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44710",
     "labels": [],
     "merged": true,
-    "number": 44665,
+    "number": 44710,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[`Chmv2`] Fix conversion after capture refactor",
-    "updated_at": "2026-03-13T14:28:33Z"
+    "title": "Fix AutoProcessor.from_pretrained silently dropping hub kwargs",
+    "updated_at": "2026-03-25T18:13:14Z"
   },
   {
-    "additions": 18,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "Fixes https://github.com/huggingface/transformers/issues/44625 and https://github.com/huggingface/transformers/pull/44406#issuecomment-4203794413 We can create a clf model with LLM or VLM backbone like: ```python from transformers import A\u2026",
-    "changed_files": 8,
+    "additions": 6778,
+    "author": "LucasMa2025",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "# \ud83c\udf9b\ufe0f Add Configurable Generation Scheduler and State Machine for `generate()` ## Summary This PR introduces a **fully optional, zero-intrusion** Generation Scheduler (`GenerationScheduler`) and explicit state machine (`GenerationStateMachi\u2026",
+    "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44664",
-    "created_at": "2026-03-13T13:39:52Z",
-    "deletions": 193,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44664/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44664",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44708",
+    "created_at": "2026-03-14T17:13:34Z",
+    "deletions": 7,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44708/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44708",
     "labels": [],
     "merged": false,
-    "number": 44664,
-    "review_comments_count": 8,
-    "state": "open",
-    "title": "Generic Sequence Classifier works for multimodal models",
-    "updated_at": "2026-04-14T09:50:30Z"
+    "number": 44708,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Add Configurable Generation Scheduler and State Machine for `generate()`",
+    "updated_at": "2026-03-14T19:19:11Z"
   },
   {
-    "additions": 1,
-    "author": "Ker102",
+    "additions": 3,
+    "author": "saivedant169",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44661 \u2014 `transformers add-new-model-like` crashes with `AttributeError: 'ModelInfos' object has no attribute 'tokenizer_class'` when selecting a model that is in `TOKENIZER_MAPPING_NAMES`. ## Root Cause PR #40936 refactor\u2026",
+    "body_excerpt": "Fixes part of #32937 ## What does this PR do? Adds `position_ids` as an explicit parameter to `MptForCausalLM.forward()` and `MptModel.forward()`, bringing MPT in line with other CausalLM models. Same rationale as the Bloom PR (#44706) \u2014 M\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44663",
-    "created_at": "2026-03-13T13:25:48Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44707",
+    "created_at": "2026-03-14T17:12:16Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44663/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44663",
+    "files_url": "https://github.com/huggingface/transformers/pull/44707/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44707",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44663,
+    "number": 44707,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: restore missing `tokenizer_class` attribute in `ModelInfos.__init__`",
-    "updated_at": "2026-03-13T14:02:00Z"
+    "title": "Add position_ids to MptForCausalLM forward pass",
+    "updated_at": "2026-03-18T13:39:36Z"
   },
   {
-    "additions": 7084,
-    "author": "CyrilSterling",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR supports PenguinVL model. Paper: https://arxiv.org/abs/2603.06569 Github repo: https://github.com/tencent-ailab/Penguin-VL HuggingFace Model: https://huggingface.co/collections/tencent/ai-lab ## Before submi\u2026",
-    "changed_files": 20,
+    "additions": 3,
+    "author": "saivedant169",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes part of #32937 ## What does this PR do? Adds `position_ids` as an explicit parameter to `BloomForCausalLM.forward()` and `BloomModel.forward()`, bringing Bloom in line with other CausalLM models like Llama, Falcon, Gemma, and Mistral\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44662",
-    "created_at": "2026-03-13T13:02:26Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44706",
+    "created_at": "2026-03-14T17:09:11Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44662/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44662",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44706/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44706",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44662,
-    "review_comments_count": 95,
-    "state": "open",
-    "title": "[model] Add PenguinVL implementation",
-    "updated_at": "2026-04-01T10:53:25Z"
+    "number": 44706,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Add position_ids to BloomForCausalLM forward pass",
+    "updated_at": "2026-03-18T13:39:51Z"
   },
   {
-    "additions": 5,
-    "author": "DogWala",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #44637 This PR makes the PEFT `load_best_model_at_end` path in `Trainer` use a CPU-first adapter reload path during best-model loading. Previously, when training a PEFT model, `Trainer` could reload the best a\u2026",
+    "additions": 14,
+    "author": "saivedant169",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes part of #32937 ## What does this PR do? RoFormer introduced rotary position embeddings, but its `ForCausalLM` forward method doesn't accept `position_ids` \u2014 which means callers can't specify custom positions for packed sequences or f\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44660",
-    "created_at": "2026-03-13T12:59:25Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44705",
+    "created_at": "2026-03-14T16:48:06Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44660/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44660",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44705/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44705",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44660,
+    "number": 44705,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix: avoid late CUDA OOM in load_best_model_at_end with PEFT models",
-    "updated_at": "2026-03-13T13:18:52Z"
+    "state": "closed",
+    "title": "Add position_ids to RoFormerForCausalLM forward pass",
+    "updated_at": "2026-03-18T13:40:05Z"
   },
   {
-    "additions": 0,
-    "author": "Olexandr88",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Removes outdated use_diff entry from the docstring. The parameter is not present in the method signature or implementation.",
-    "changed_files": 1,
+    "additions": 26,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "As per title, it seems that the `cute` subfolder can be even distributed if you only install FA2 which implies something wrong. Now we check under the (normalized) distribution names",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44659",
-    "created_at": "2026-03-13T11:08:13Z",
-    "deletions": 4,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44703",
+    "created_at": "2026-03-14T14:46:02Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44659/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44659",
+    "files_url": "https://github.com/huggingface/transformers/pull/44703/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44703",
     "labels": [],
+    "merged": true,
+    "number": 44703,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "[`FA`] Fix fa detection",
+    "updated_at": "2026-03-14T17:19:07Z"
+  },
+  {
+    "additions": 148,
+    "author": "LincolnBurrows2017",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR fix? The `rms_norm_eps` parameter in `MistralConfig` was incorrectly typed as `int | None` but defaults to `1e-6` which is a float. This parameter is passed to `MistralRMSNorm` which expects `eps: float`. ### Bug Detai\u2026",
+    "changed_files": 8,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44702",
+    "created_at": "2026-03-14T14:41:15Z",
+    "deletions": 25,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44702/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44702",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44659,
+    "number": 44702,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "docs: remove outdated use_diff docstring from DistributedConfig.to_js\u2026",
-    "updated_at": "2026-03-13T13:42:29Z"
+    "state": "closed",
+    "title": "fix: Correct rms_norm_eps type hint from int to float in MistralConfig",
+    "updated_at": "2026-03-18T13:00:12Z"
   },
   {
-    "additions": 18,
-    "author": "Charly21r",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes an issue where the `.modelcard` attribute of a pipeline is not initialized. Without this fix, calling `save_pretrained` on a pipeline (e.g., ASR pipeline) raises an `AttributeError` because `.modelcard` does n\u2026",
-    "changed_files": 2,
+    "additions": 219,
+    "author": "hmellor",
+    "author_association": "MEMBER",
+    "body_excerpt": "These models have `base_model_pp_plan`s but currently do not work because the base model's forward pass depends on all the `layers` being `Qwen2VLDecoderLayer`. i.e. if one of the layers is removed/replaced with `Identity`, `decoder_layer.\u2026",
+    "changed_files": 52,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44658",
-    "created_at": "2026-03-13T10:36:22Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44699",
+    "created_at": "2026-03-14T11:44:24Z",
+    "deletions": 148,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44658/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44658",
+    "files_url": "https://github.com/huggingface/transformers/pull/44699/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44699",
     "labels": [],
-    "merged": false,
-    "number": 44658,
+    "merged": true,
+    "number": 44699,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix: fix Pipeline's save_pretrained method (issue #44655)",
-    "updated_at": "2026-03-13T14:08:27Z"
+    "title": "Fix several based models' pipeline parallel support",
+    "updated_at": "2026-03-20T13:53:27Z"
   },
   {
     "additions": 1,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@ArthurZucker @Cyrilvallez pls help review. Thx!",
+    "author": "hmellor",
+    "author_association": "MEMBER",
+    "body_excerpt": "The typo in the `elif` chain meant that `image` and `video` modalidty encoders could not be set using this method. This PR fixes the typo so that they can.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44657",
-    "created_at": "2026-03-13T10:25:07Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44698",
+    "created_at": "2026-03-14T11:18:54Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44657/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44657",
+    "files_url": "https://github.com/huggingface/transformers/pull/44698/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44698",
     "labels": [],
     "merged": true,
-    "number": 44657,
+    "number": 44698,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix bug embedding_size mismatch with hidden_size in electra model test",
-    "updated_at": "2026-04-01T08:24:54Z"
+    "title": "Fix `set_encoder`",
+    "updated_at": "2026-03-14T13:42:00Z"
   },
   {
-    "additions": 80,
-    "author": "juliendenize",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR add support to mistral-common v1.10.0. This involves: - reasoning effort feature - Python 3.14 Also add `image_sizes` in `apply_chat_template` output to match what is expected by standard processor.",
+    "additions": 75,
+    "author": "LincolnBurrows2017",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Description The `torch_float` function in `src/transformers/utils/generic.py` was incorrectly returning `int(x)` in two places where it should return `float(x)`: 1. When torch is not available (fallback case) 2. When not in a tracing co\u2026",
     "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44656",
-    "created_at": "2026-03-13T10:24:11Z",
-    "deletions": 15,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44697",
+    "created_at": "2026-03-14T10:44:12Z",
+    "deletions": 25,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44656/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44656",
+    "files_url": "https://github.com/huggingface/transformers/pull/44697/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44697",
     "labels": [],
-    "merged": true,
-    "number": 44656,
+    "merged": false,
+    "number": 44697,
     "review_comments_count": 1,
-    "state": "closed",
-    "title": "[MistralCommonBackend] Upgrade mistral-common to v1.10.0",
-    "updated_at": "2026-03-16T12:46:29Z"
+    "state": "open",
+    "title": "fix: torch_float should return float, not int",
+    "updated_at": "2026-03-17T19:29:02Z"
   },
   {
-    "additions": 13,
-    "author": "jiqing-feng",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR fixes the AWQ test suite to align with the GPTQModel migration (following #41567 and #42776). ### Changes - **Fix `replace_with_awq_linear` return value**: The function now returns the model directly instead of a tuple `(model, _)`\u2026",
+    "additions": 19,
+    "author": "hmellor",
+    "author_association": "MEMBER",
+    "body_excerpt": "In configs, `base_model_pp_plan` and `base_model_tp_plan` default to `None` In models, `_pp_plan` and `_tp_plan` _look like_ they default to `None` based on the class variables, but will actually always be a dict because of `post_init`. Th\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44654",
-    "created_at": "2026-03-13T07:31:19Z",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44696",
+    "created_at": "2026-03-14T09:41:07Z",
     "deletions": 13,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44654/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44654",
+    "files_url": "https://github.com/huggingface/transformers/pull/44696/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44696",
+    "labels": [],
+    "merged": true,
+    "number": 44696,
+    "review_comments_count": 5,
+    "state": "closed",
+    "title": "Fix `supports_{tp/pp}_plan`",
+    "updated_at": "2026-03-31T13:12:56Z"
+  },
+  {
+    "additions": 4,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? The following failing tests were identified and fixed in this PR: \u2192 **Kyutai Speech-To-Text**: [The PR [processors] Unbloating simple processors](https://github.com/huggingface/transformers/pull/40377), [refactore\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44695",
+    "created_at": "2026-03-14T09:05:35Z",
+    "deletions": 4,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44695/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44695",
     "labels": [],
     "merged": true,
-    "number": 44654,
-    "review_comments_count": 0,
+    "number": 44695,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Fix AWQ tests for GPTQModel migration",
-    "updated_at": "2026-03-13T16:28:14Z"
+    "title": "fix(testing): Fix Kyutai Speech-To-Text and LongCatFlash test failures on main CI",
+    "updated_at": "2026-04-09T15:41:05Z"
   },
   {
-    "additions": 18,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@zucchini-nlp, can you help review? Thx! unit tests to reproduce this bug: `tests/models/phi4_multimodal/test_modeling_phi4_multimodal.py::Phi4MultimodalIntegrationTest::test_audio_text_generation`",
-    "changed_files": 1,
+    "additions": 143,
+    "author": "LincolnBurrows2017",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes issue #44625: Qwen3.5 num_labels not propagated from core config to text config. When loading `AutoConfig.from_pretrained(\"Qwen3.5\", num_labels=1)`, the outer config gets `num_labels=1` but the inner `text_config` still ha\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44653",
-    "created_at": "2026-03-13T07:14:25Z",
-    "deletions": 9,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44693",
+    "created_at": "2026-03-14T05:43:00Z",
+    "deletions": 30,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44653/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44653",
+    "files_url": "https://github.com/huggingface/transformers/pull/44693/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44693",
     "labels": [],
     "merged": false,
-    "number": 44653,
-    "review_comments_count": 7,
+    "number": 44693,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix `AutoImageProcessor` to correctly detect local implementation whe\u2026",
-    "updated_at": "2026-04-09T02:32:53Z"
+    "title": "fix: Propagate num_labels to text_config in Qwen3.5",
+    "updated_at": "2026-03-18T12:56:25Z"
   },
   {
-    "additions": 8,
+    "additions": 18,
     "author": "gambletan",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44623 `processor.save_pretrained()` in v5 only saves the unified `processor_config.json` with nested sub-processor configs, but does not save standalone config files like `preprocessor_config.json` for the image processor\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary Fixes #44514. `Qwen2_5_VLProcessor.apply_chat_template` crashes with `ValueError` when called with batched input and `padding=False` (the default). The root cause is `np.array(text_inputs[\"input_ids\"])` which fails when sequence\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44652",
-    "created_at": "2026-03-13T05:38:10Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44692",
+    "created_at": "2026-03-14T04:14:38Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44652/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44652",
+    "files_url": "https://github.com/huggingface/transformers/pull/44692/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44692",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44652,
+    "number": 44692,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix processor.save_pretrained missing sub-processor config files",
-    "updated_at": "2026-03-13T12:03:37Z"
+    "title": "fix: handle ragged input_ids in Qwen2_5_VLProcessor.apply_chat_template",
+    "updated_at": "2026-03-18T12:44:18Z"
   },
   {
-    "additions": 10,
+    "additions": 23,
     "author": "gambletan",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44625 When passing `num_labels` to `AutoConfig.from_pretrained` for Qwen3.5, the value is set on the outer `Qwen3_5Config` but not propagated to `text_config`. This causes `AutoModelForSequenceClassification` to use the d\u2026",
+    "body_excerpt": "## Summary - Fixes `num_labels` (and `id2label`/`label2id`) not being propagated from the outer `Qwen3_5Config` to its inner `text_config` when passed via `AutoConfig.from_pretrained(..., num_labels=1)`. - When `text_config` is `None` or a\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44651",
-    "created_at": "2026-03-13T05:35:29Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44691",
+    "created_at": "2026-03-14T04:10:54Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44651/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44651",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44691/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44691",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44651,
+    "number": 44691,
     "review_comments_count": 0,
     "state": "closed",
     "title": "Fix Qwen3.5 num_labels not propagated to text_config",
-    "updated_at": "2026-03-13T13:39:43Z"
+    "updated_at": "2026-03-18T12:57:19Z"
   },
   {
-    "additions": 188,
-    "author": "shaealh",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Closes #44593 ## Summary - use generation_input_ids/generation_attention_mask when provided for decoder-only models - otherwise infer prompt from leading -100 labels and build left-padded prompt batch - return completion tokens for decoder\u2026",
+    "additions": 6,
+    "author": "gambletan",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #44360 The `GlmMoeDsaIndexer` is missing a ReLU activation on the per-head dot-product scores before the weighted sum across heads. The reference DeepSeek V3.2 implementation applies ReLU inside the `fp8_index` kernel: ```\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44650",
-    "created_at": "2026-03-13T05:33:59Z",
-    "deletions": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44690",
+    "created_at": "2026-03-14T03:44:37Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44650/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44650",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44690/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44690",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44650,
+    "number": 44690,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix Seq2SeqTrainer generation path for decoder-only models",
-    "updated_at": "2026-04-02T21:23:53Z"
+    "state": "closed",
+    "title": "Fix missing ReLU in GLM-MOE-DSA indexer scoring",
+    "updated_at": "2026-03-18T12:40:23Z"
   },
   {
-    "additions": 12,
-    "author": "gambletan",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44623 In v5.3.0, `ProcessorMixin.save_pretrained()` no longer calls `save_pretrained()` on non-tokenizer components (e.g. `image_processor`, `feature_extractor`). This means their individual config files (e.g. `preprocess\u2026",
-    "changed_files": 1,
+    "additions": 141,
+    "author": "LincolnBurrows2017",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes issue #44625: Qwen3.5 num_labels not propagating to text_config. When calling `AutoConfig.from_pretrained(\"Qwen3.5\", num_labels=1)`, the main config gets `num_labels=1` but text_config still has default `num_labels=2`. Thi\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44649",
-    "created_at": "2026-03-13T05:22:42Z",
-    "deletions": 207,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44688",
+    "created_at": "2026-03-14T00:40:50Z",
+    "deletions": 23,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44649/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44649",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44688/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44688",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44649,
+    "number": 44688,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: ensure processor.save_pretrained saves all component files",
-    "updated_at": "2026-03-13T05:36:54Z"
+    "title": "fix: Propagate num_labels to text_config in Qwen models",
+    "updated_at": "2026-03-18T12:56:41Z"
   },
   {
-    "additions": 0,
-    "author": "gambletan",
+    "additions": 8,
+    "author": "vxa8502",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44625 When `num_labels` or `id2label` are passed to `Qwen3_5Config` (e.g. via `AutoConfig.from_pretrained(model_name, num_labels=1)`), they are only set on the outer composite config but **not forwarded to `text_config`**\u2026",
-    "changed_files": 0,
+    "body_excerpt": "Fixes partial #32937 Adds explicit `position_ids` threading through GPT-Neo's attention layers to enable flash attention's packed sequence optimization. ## Context GPT-Neo uses learned absolute position embeddings (`wpe`) applied at the mo\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44648",
-    "created_at": "2026-03-13T05:22:03Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44687",
+    "created_at": "2026-03-13T23:28:55Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44648/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44648",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44687/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44687",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44648,
+    "number": 44687,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: propagate num_labels to text_config in Qwen3.5",
-    "updated_at": "2026-03-13T05:33:26Z"
+    "title": "Add explicit position_ids to GPT-Neo attention layers",
+    "updated_at": "2026-03-18T13:06:49Z"
   },
   {
-    "additions": 9,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@remi-or @ArthurZucker @McPatate pls help review, thx!",
-    "changed_files": 1,
+    "additions": 615,
+    "author": "tejasae-afk",
+    "author_association": "NONE",
+    "body_excerpt": "During an automated code review of src/transformers/models/marian/convert_marian_to_pytorch.py, the following issue was identified. Use safe_load in convert marian to pytorch. yaml.load on untrusted input can construct arbitrary Python obj\u2026",
+    "changed_files": 80,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44647",
-    "created_at": "2026-03-13T04:55:26Z",
-    "deletions": 6,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44686",
+    "created_at": "2026-03-13T21:22:07Z",
+    "deletions": 259,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44647/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44647",
+    "files_url": "https://github.com/huggingface/transformers/pull/44686/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44686",
     "labels": [],
     "merged": false,
-    "number": 44647,
+    "number": 44686,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add more generic device support for continuous batching",
-    "updated_at": "2026-04-09T02:32:51Z"
+    "title": "Use safe_load in convert marian to pytorch",
+    "updated_at": "2026-03-14T03:54:31Z"
   },
   {
-    "additions": 4,
-    "author": "LincolnBurrows2017",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixed typo in Qwen3.5 and Qwen3-VL-MoE model files ```diff - Since Qwen3.5 use timestamps to seperate videos + Since Qwen3.5 use timestamps to separate videos ```",
-    "changed_files": 2,
+    "additions": 10,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? For tiny model creation script - new added model test files still miss this argument ...",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44646",
-    "created_at": "2026-03-13T04:48:06Z",
-    "deletions": 4,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44685",
+    "created_at": "2026-03-13T20:53:41Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44646/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44646",
+    "files_url": "https://github.com/huggingface/transformers/pull/44685/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44685",
     "labels": [],
-    "merged": false,
-    "number": 44646,
+    "merged": true,
+    "number": 44685,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix typo: seperate -> separate",
-    "updated_at": "2026-03-13T15:27:08Z"
+    "state": "closed",
+    "title": "Fix more model tester missing `parent` issue",
+    "updated_at": "2026-03-13T21:03:46Z"
   },
   {
-    "additions": 4,
-    "author": "sywangyi",
+    "additions": 41,
+    "author": "ntenenz",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "- Intel XPU: @IlyasMoutawwakil ``` ======================================================================== FAILURES ======================================================================== _________________________________________________\u2026",
-    "changed_files": 3,
+    "body_excerpt": "\u2026 # What does this PR do? In torch versions >= 2.9.0, it requests the lse from flex_attenetion using `AuxRequest` instead of the deprecated `return_lse`, which triggers a warning and can break tracing. Fixes #44683 ## Before submitting - [\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44645",
-    "created_at": "2026-03-13T02:53:31Z",
-    "deletions": 4,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44684",
+    "created_at": "2026-03-13T20:16:35Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44645/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44645",
+    "files_url": "https://github.com/huggingface/transformers/pull/44684/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44684",
     "labels": [],
     "merged": true,
-    "number": 44645,
-    "review_comments_count": 0,
+    "number": 44684,
+    "review_comments_count": 8,
     "state": "closed",
-    "title": "fix Image.open failure in case \"tests/models/prompt_depth_anything/te\u2026",
-    "updated_at": "2026-03-27T11:11:05Z"
+    "title": "update flex attention to use `return_aux` instead of `return_lse` when torch verison >= 2.9",
+    "updated_at": "2026-03-18T11:44:18Z"
   },
   {
-    "additions": 2,
-    "author": "sywangyi",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "\u2026MXFP4PseudoquantTest::test_quantized_model fail in xpu ## Who can review? - quantization: @SunMarc - Intel XPU: @IlyasMoutawwakil",
-    "changed_files": 1,
+    "additions": 301,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Llama cpp integration in transformers serve. Minor changes to add llama.cpp integration Mostly changes on serve to fix latency for streaming and non streaming",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44644",
-    "created_at": "2026-03-13T02:02:45Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44682",
+    "created_at": "2026-03-13T18:52:41Z",
+    "deletions": 73,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44644/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44644",
+    "files_url": "https://github.com/huggingface/transformers/pull/44682/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44682",
     "labels": [],
-    "merged": true,
-    "number": 44644,
-    "review_comments_count": 6,
-    "state": "closed",
-    "title": "fix tests/quantization/fp_quant_integration/test_fp_quant.py::FPQuant\u2026",
-    "updated_at": "2026-03-27T14:14:13Z"
+    "merged": false,
+    "number": 44682,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "transformers serve + llamacpp",
+    "updated_at": "2026-03-14T07:05:29Z"
   },
   {
-    "additions": 5,
-    "author": "joshuaswanson",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "The generic `labels` docstring in `ModelArgs` says \"masked language modeling loss\" and doesn't mention that causal LM models shift labels internally. This has tripped up a lot of users who pre-shift their labels and end up training next-ne\u2026",
-    "changed_files": 1,
+    "additions": 47,
+    "author": "dacorvo",
+    "author_association": "MEMBER",
+    "body_excerpt": "Fixes #44679 ## Summary - Custom attention kernels registered via `load_and_register_attn_kernel` currently get hardcoded `flash_attention_2` mask dispatch, which produces 2D or `None` masks - Kernels that need SDPA-style 4D boolean masks\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44642",
-    "created_at": "2026-03-12T23:47:11Z",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44680",
+    "created_at": "2026-03-13T17:55:54Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44642/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44642",
+    "files_url": "https://github.com/huggingface/transformers/pull/44680/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44680",
     "labels": [],
     "merged": false,
-    "number": 44642,
-    "review_comments_count": 0,
+    "number": 44680,
+    "review_comments_count": 12,
     "state": "open",
-    "title": "Clarify that causal LM labels are shifted internally",
-    "updated_at": "2026-03-13T00:02:30Z"
+    "title": "Allow kernel modules to declare their preferred mask function",
+    "updated_at": "2026-04-14T19:29:06Z"
   },
   {
-    "additions": 1,
-    "author": "kmbhattt-aws",
+    "additions": 9,
+    "author": "JokeYoonic",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Issue: A full 4D attention mask of shape `[1, 1, seq_len, seq_len]` is being created during attention, even when not using alibi for positional embeddings. - This occupied extra memory during training. Root Cause: T\u2026",
+    "body_excerpt": "Problem: - On macOS ARM64 + Python 3.13 + transformers 5.x, GPT-2 model's lm_head forward pass produces NaN/Inf values during inference - Root cause: lm_head.weight is tied to transformer.wte.weight, and the shared memory reference causes\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44641",
-    "created_at": "2026-03-12T23:01:11Z",
-    "deletions": 1,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44641/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44641",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44676",
+    "created_at": "2026-03-13T16:28:01Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44676/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44676",
     "labels": [],
     "merged": false,
-    "number": 44641,
-    "review_comments_count": 1,
+    "number": 44676,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Conditinally passing and_mask_function arg to create_causal_mask ",
-    "updated_at": "2026-03-13T02:09:22Z"
+    "title": "fix(gpt2): Resolve NaN/Inf issue in lm_head on Python 3.13 with tied weights",
+    "updated_at": "2026-03-18T17:16:49Z"
   },
   {
-    "additions": 11,
-    "author": "michaelbenayoun",
+    "additions": 32,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - Add `\"embedding_colwise\"` tp plan - Add register methods for `ParallelInterface`. Without it, we can register plans with the register method, but not the corresponding `plan_to_weight_dim` and `plan_to_bias_dim`.",
+    "body_excerpt": "properly formats the `ContinuousBatchingConfig` below: <img width=\"976\" height=\"626\" alt=\"Screenshot 2026-03-13 at 9 09 39 AM\" src=\"https://github.com/user-attachments/assets/4390c6f7-bb63-4039-a46e-9f4ae23f5d98\" />",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44640",
-    "created_at": "2026-03-12T20:14:06Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44675",
+    "created_at": "2026-03-13T16:10:28Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44640/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44640",
+    "files_url": "https://github.com/huggingface/transformers/pull/44675/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44675",
     "labels": [],
     "merged": true,
-    "number": 44640,
+    "number": 44675,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add register method for `ParallelInterface`",
-    "updated_at": "2026-03-13T18:57:48Z"
+    "title": "[docs] cb config",
+    "updated_at": "2026-03-13T23:15:04Z"
   },
   {
-    "additions": 24099,
-    "author": "ArthurZucker",
+    "additions": 408,
+    "author": "Rocketknight1",
     "author_association": "MEMBER",
-    "body_excerpt": "## Summary Fixes bugs introduced during the `__init__` \u2192 `@dataclass` conversion in #41250. All are incorrect default values caught by automated comparison of `__init__` signatures (main) vs dataclass fields (PR). | Model | Param | Was | S\u2026",
-    "changed_files": 931,
+    "body_excerpt": "We've had `parse_response()` in the library for a while, but it's been a soft launch / prototype feature. This PR cleans it up and documents it, making it an official feature! The API is largely unchanged from the prototype, but we drop `x\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44639",
-    "created_at": "2026-03-12T16:49:54Z",
-    "deletions": 38773,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44674",
+    "created_at": "2026-03-13T15:41:42Z",
+    "deletions": 34,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44639/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44639",
+    "files_url": "https://github.com/huggingface/transformers/pull/44674/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44674",
     "labels": [],
-    "merged": false,
-    "number": 44639,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44674,
+    "review_comments_count": 11,
     "state": "closed",
-    "title": "Fix incorrect default values in config dataclass migration",
-    "updated_at": "2026-03-12T16:50:10Z"
+    "title": "Officially launch parse_response",
+    "updated_at": "2026-03-24T15:55:05Z"
   },
   {
-    "additions": 19,
-    "author": "Cyrilvallez",
+    "additions": 73,
+    "author": "remi-or",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. Supersedes https://github.com/huggingface/transformers/pull/44614. This one is the result of a long debugging session and discussion with @vasqu. The issue is as follow: - Backbone ALWAYS need to c\u2026",
+    "body_excerpt": "This PR fixes a bug in continuous batching where non-CUDA devices cannot use the feature because some CUDA-exclusive objects are always instantiated. It also adds a test to make sure this will not break again in the future.",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44638",
-    "created_at": "2026-03-12T16:19:49Z",
-    "deletions": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44673",
+    "created_at": "2026-03-13T15:37:01Z",
+    "deletions": 15,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44638/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44638",
+    "files_url": "https://github.com/huggingface/transformers/pull/44673/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44673",
     "labels": [],
     "merged": true,
-    "number": 44638,
-    "review_comments_count": 2,
+    "number": 44673,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix output capturing for Backbones",
-    "updated_at": "2026-03-12T17:11:32Z"
+    "title": "[CB] [Bug] Fix crashes when running without cuda",
+    "updated_at": "2026-03-15T23:59:55Z"
   },
   {
-    "additions": 571,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - Adds an `httpx` tracer to gather metrics about network calls - Collect and store metrics and generates an artifact in CI - Can be used locally with `DEBUG_NETWORK` - Activated in CircleCI example of local run: ```\u2026",
-    "changed_files": 4,
+    "additions": 1,
+    "author": "neo",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44636",
-    "created_at": "2026-03-12T15:25:10Z",
-    "deletions": 0,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44672",
+    "created_at": "2026-03-13T15:33:15Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44636/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44636",
+    "files_url": "https://github.com/huggingface/transformers/pull/44672/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44672",
     "labels": [],
     "merged": true,
-    "number": 44636,
-    "review_comments_count": 5,
+    "number": 44672,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "feat(ci): added a network debug report",
-    "updated_at": "2026-03-18T19:19:03Z"
+    "title": "Fix annotations reader for python 3.14 in `PreTrainedModel`",
+    "updated_at": "2026-03-19T13:30:48Z"
   },
   {
-    "additions": 111,
-    "author": "RyanMullins",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Per a conversion with @Cyrilvallez on Slack on how to make Gemma models more compatible with modular inheritance, this PR: * Updates Gemma models to use `nn.parameter.Buffer` instead of a `self.register_buffer()` fo\u2026",
-    "changed_files": 32,
+    "additions": 6,
+    "author": "dacorvo",
+    "author_association": "MEMBER",
+    "body_excerpt": "Fixes #44678 ## Summary - Replace advanced indexing (`self.gate_up_proj[expert_ids]`) with explicit `torch.index_select(self.gate_up_proj, 0, expert_ids)` in `batched_mm_experts_forward` - 6 replacements total (3 weight tensors + 3 bias te\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44635",
-    "created_at": "2026-03-12T14:47:46Z",
-    "deletions": 87,
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44669",
+    "created_at": "2026-03-13T14:52:22Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44635/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44635",
+    "files_url": "https://github.com/huggingface/transformers/pull/44669/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44669",
     "labels": [],
     "merged": false,
-    "number": 44635,
+    "number": 44669,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[Gemma] Modular-friendly buffers",
-    "updated_at": "2026-03-18T10:44:25Z"
+    "state": "closed",
+    "title": "Use `index_select` instead of advanced indexing in `batched_mm_experts_forward`",
+    "updated_at": "2026-03-19T13:39:23Z"
+  },
+  {
+    "additions": 18,
+    "author": "dacorvo",
+    "author_association": "MEMBER",
+    "body_excerpt": "Fixes #44677 ## Summary - Add `base_model_tp_plan` to `OlmoeConfig`, enabling `from_pretrained(tp_plan=\"auto\")` for OLMoE models - Add `TensorParallelTesterMixin` to OLMoE tests for TP validation coverage - Uses `\"colwise\"` for `q_norm` an\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44668",
+    "created_at": "2026-03-13T14:45:22Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44668/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44668",
+    "labels": [],
+    "merged": true,
+    "number": 44668,
+    "review_comments_count": 4,
+    "state": "closed",
+    "title": "Add `base_model_tp_plan` to `OlmoeConfig`",
+    "updated_at": "2026-03-26T13:58:58Z"
   },
   {
-    "additions": 30,
-    "author": "Cyrilvallez",
+    "additions": 412,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? A few find-and-replaces and cache position is deleted from the rest of models. Still have to check docs and test files, so WIP",
+    "changed_files": 120,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44634",
-    "created_at": "2026-03-12T14:04:36Z",
-    "deletions": 6,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44667",
+    "created_at": "2026-03-13T14:37:26Z",
+    "deletions": 1519,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44634/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44634",
+    "files_url": "https://github.com/huggingface/transformers/pull/44667/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44667",
     "labels": [],
-    "merged": true,
-    "number": 44634,
-    "review_comments_count": 1,
+    "merged": false,
+    "number": 44667,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix lfm2 kernel path",
-    "updated_at": "2026-03-12T15:00:59Z"
+    "title": "Goodbye cache position",
+    "updated_at": "2026-03-19T11:55:04Z"
   },
   {
-    "additions": 26,
-    "author": "eustlb",
+    "additions": 17,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per title \ud83e\udd17",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? - 2 model tester classes didn't follow the usual way we do things, which cause the tiny model creation script to fail with those model classes. - (the script initializes instances of model testers, in order to call\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44633",
-    "created_at": "2026-03-12T13:35:48Z",
-    "deletions": 11,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44666",
+    "created_at": "2026-03-13T14:24:50Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44633/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44633",
+    "files_url": "https://github.com/huggingface/transformers/pull/44666/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44666",
     "labels": [],
     "merged": true,
-    "number": 44633,
-    "review_comments_count": 2,
+    "number": 44666,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[medasr] doc update",
-    "updated_at": "2026-03-16T09:39:50Z"
+    "title": "Another (small) set of fixes required for tiny model creation",
+    "updated_at": "2026-03-13T17:20:52Z"
   },
   {
-    "additions": 35,
-    "author": "Abdennacer-Badaoui",
+    "additions": 1,
+    "author": "vasqu",
     "author_association": "MEMBER",
-    "body_excerpt": "Summary - Update test expectations for affected models - Add some needed dependencies - Fix TypeError: `GenerationMixin.prepare_inputs_for_generation()` got multiple values for argument 'next_sequence_length' in Qwen2.5-Omni talker by pass\u2026",
-    "changed_files": 6,
+    "body_excerpt": "Dinov3 vit was refactored to introduce a module between top level and layers to have the capture decorators work as intended. Otherwise, it would force the backbone to do manual collection. This introduced a small conversion which is now a\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44632",
-    "created_at": "2026-03-12T13:32:33Z",
-    "deletions": 22,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44665",
+    "created_at": "2026-03-13T13:50:26Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44632/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44632",
+    "files_url": "https://github.com/huggingface/transformers/pull/44665/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44665",
     "labels": [],
     "merged": true,
-    "number": 44632,
-    "review_comments_count": 6,
+    "number": 44665,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[AMD CI] Fix test failures across important models ",
-    "updated_at": "2026-03-17T14:58:10Z"
+    "title": "[`Chmv2`] Fix conversion after capture refactor",
+    "updated_at": "2026-03-13T14:28:33Z"
   },
   {
-    "additions": 33,
-    "author": "RyanMullins",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Updates the weights conversion scripts for Gemma to: * Use the new `SentencePieceExtractor` class to get the vocab and merges from the SPM * Always initialize and save the unified `GemmaTokenizer` class ## Before su\u2026",
-    "changed_files": 4,
+    "additions": 121,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "Fixes https://github.com/huggingface/transformers/issues/44625 and https://github.com/huggingface/transformers/pull/44406#issuecomment-4203794413 After long discussion, ig this is the easiest and most straightforward way for us to use `Gen\u2026",
+    "changed_files": 9,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44631",
-    "created_at": "2026-03-12T13:32:25Z",
-    "deletions": 45,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44664",
+    "created_at": "2026-03-13T13:39:52Z",
+    "deletions": 177,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44631/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44631",
+    "files_url": "https://github.com/huggingface/transformers/pull/44664/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44664",
     "labels": [],
-    "merged": true,
-    "number": 44631,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[Gemma] Update conversion scripts for Transformers v5 Comaptibility",
-    "updated_at": "2026-03-18T10:39:53Z"
+    "merged": false,
+    "number": 44664,
+    "review_comments_count": 8,
+    "state": "open",
+    "title": "Generic Sequence Classifier works for multimodal models",
+    "updated_at": "2026-04-16T14:10:47Z"
   },
   {
-    "additions": 42,
-    "author": "MaybeSam05",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes a bug where `num_labels` passed to `AutoConfig.from_pretrained` for Qwen3.5 did not propagate from the top\u2011level `Qwen3_5Config` into the `text_config`, so `AutoModelForSequenceClassification` still saw the de\u2026",
-    "changed_files": 2,
+    "additions": 1,
+    "author": "Ker102",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #44661 \u2014 `transformers add-new-model-like` crashes with `AttributeError: 'ModelInfos' object has no attribute 'tokenizer_class'` when selecting a model that is in `TOKENIZER_MAPPING_NAMES`. ## Root Cause PR #40936 refactor\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44630",
-    "created_at": "2026-03-12T13:25:54Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44663",
+    "created_at": "2026-03-13T13:25:48Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44630/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44630",
+    "files_url": "https://github.com/huggingface/transformers/pull/44663/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44663",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44630,
+    "number": 44663,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Qwen3.5 num_labels propagation to text_config (fix #44625)",
-    "updated_at": "2026-03-12T13:46:07Z"
+    "title": "fix: restore missing `tokenizer_class` attribute in `ModelInfos.__init__`",
+    "updated_at": "2026-03-13T14:02:00Z"
   },
   {
-    "additions": 15,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes \"auto\" dtype when the model is initialized `from_config` It was already fixed for `from_pretrained` in https://github.com/huggingface/transformers/pull/42990 but vLLM creates models with `AutoModel._from_confi\u2026",
-    "changed_files": 2,
+    "additions": 7084,
+    "author": "CyrilSterling",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR supports PenguinVL model. Paper: https://arxiv.org/abs/2603.06569 Github repo: https://github.com/tencent-ailab/Penguin-VL HuggingFace Model: https://huggingface.co/collections/tencent/ai-lab ## Before submi\u2026",
+    "changed_files": 20,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44629",
-    "created_at": "2026-03-12T13:07:55Z",
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44662",
+    "created_at": "2026-03-13T13:02:26Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44629/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44629",
-    "labels": [],
-    "merged": true,
-    "number": 44629,
-    "review_comments_count": 3,
-    "state": "closed",
-    "title": "Ensure same `dtype` for subconfig when `_from_config`",
-    "updated_at": "2026-03-13T11:35:10Z"
-  },
-  {
-    "additions": 37,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - `encoder_config` and `decoder_config` should return `None` for encoder / decoder config classes themselves. - The encoder / decoder model classes should have the correct config classes associated to them <!-- CURS\u2026",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44628",
-    "created_at": "2026-03-12T12:24:44Z",
-    "deletions": 5,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44628/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44628",
+    "files_url": "https://github.com/huggingface/transformers/pull/44662/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44662",
     "labels": [],
-    "merged": true,
-    "number": 44628,
-    "review_comments_count": 1,
-    "state": "closed",
-    "title": "Fix for `VibeVoiceAcousticTokenizer`",
-    "updated_at": "2026-03-12T13:33:02Z"
+    "merged": false,
+    "number": 44662,
+    "review_comments_count": 95,
+    "state": "open",
+    "title": "[model] Add PenguinVL implementation",
+    "updated_at": "2026-04-01T10:53:25Z"
   },
   {
-    "additions": 141,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The diff in revert mapping is needed, otherwise we get failures in a few models, see https://app.circleci.com/pipelines/github/huggingface/transformers/167425/workflows/fa96efe5-f810-408e-bafd-de03b7e881aa/jobs/2208\u2026",
-    "changed_files": 78,
+    "additions": 5,
+    "author": "DogWala",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #44637 This PR makes the PEFT `load_best_model_at_end` path in `Trainer` use a CPU-first adapter reload path during best-model loading. Previously, when training a PEFT model, `Trainer` could reload the best a\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 16,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44627",
-    "created_at": "2026-03-12T12:00:31Z",
-    "deletions": 367,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44660",
+    "created_at": "2026-03-13T12:59:25Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44627/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44627",
+    "files_url": "https://github.com/huggingface/transformers/pull/44660/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44660",
     "labels": [],
-    "merged": true,
-    "number": 44627,
-    "review_comments_count": 8,
-    "state": "closed",
-    "title": "Move VLM conversions to the main mapping",
-    "updated_at": "2026-03-17T10:13:03Z"
+    "merged": false,
+    "number": 44660,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix: avoid late CUDA OOM in load_best_model_at_end with PEFT models",
+    "updated_at": "2026-03-13T13:18:52Z"
   },
   {
-    "additions": 11,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Adds a missing branch. I don't really know if this is worth it, can't find a model online that enforces the flag to `True`",
+    "additions": 0,
+    "author": "Olexandr88",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Removes outdated use_diff entry from the docstring. The parameter is not present in the method signature or implementation.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44626",
-    "created_at": "2026-03-12T11:23:21Z",
-    "deletions": 5,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44659",
+    "created_at": "2026-03-13T11:08:13Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44626/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44626",
+    "files_url": "https://github.com/huggingface/transformers/pull/44659/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44659",
     "labels": [],
     "merged": false,
-    "number": 44626,
-    "review_comments_count": 1,
+    "number": 44659,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "don't break legacy behavior when enforced!",
-    "updated_at": "2026-03-12T11:32:46Z"
+    "title": "docs: remove outdated use_diff docstring from DistributedConfig.to_js\u2026",
+    "updated_at": "2026-03-13T13:42:29Z"
   },
   {
-    "additions": 34,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Follow-up of #44549",
-    "changed_files": 16,
+    "additions": 18,
+    "author": "Charly21r",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes an issue where the `.modelcard` attribute of a pipeline is not initialized. Without this fix, calling `save_pretrained` on a pipeline (e.g., ASR pipeline) raises an `AttributeError` because `.modelcard` does n\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44624",
-    "created_at": "2026-03-12T09:26:17Z",
-    "deletions": 34,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44658",
+    "created_at": "2026-03-13T10:36:22Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44624/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44624",
+    "files_url": "https://github.com/huggingface/transformers/pull/44658/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44658",
     "labels": [],
-    "merged": true,
-    "number": 44624,
+    "merged": false,
+    "number": 44658,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix more wrong HF hub checkpoint names",
-    "updated_at": "2026-03-12T09:59:12Z"
+    "title": "Fix: fix Pipeline's save_pretrained method (issue #44655)",
+    "updated_at": "2026-03-13T14:08:27Z"
   },
   {
-    "additions": 17,
-    "author": "LysandreJik",
-    "author_association": "MEMBER",
-    "body_excerpt": "CB temporarily disabled on non-cuda devices as it's currently using cuda streams by default.",
-    "changed_files": 3,
+    "additions": 1,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@ArthurZucker @Cyrilvallez pls help review. Thx!",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44622",
-    "created_at": "2026-03-12T08:11:10Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44657",
+    "created_at": "2026-03-13T10:25:07Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44622/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44622",
+    "files_url": "https://github.com/huggingface/transformers/pull/44657/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44657",
     "labels": [],
-    "merged": false,
-    "number": 44622,
+    "merged": true,
+    "number": 44657,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "CB temporarily disabled on non-cuda devices",
-    "updated_at": "2026-03-16T00:00:20Z"
+    "title": "fix bug embedding_size mismatch with hidden_size in electra model test",
+    "updated_at": "2026-04-01T08:24:54Z"
   },
   {
-    "additions": 0,
-    "author": "KoichiYasuoka",
+    "additions": 80,
+    "author": "juliendenize",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #43170 for `modelcard` removal Quick reproduce: ``` from transformers import pipeline fmp=pipeline(\"fill-mask\",\"google-bert/bert-base-cased\") fmp.save_pretrained(\"tmpdir\") ``` ## Before submitting - [ ] This P\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? This PR add support to mistral-common v1.10.0. This involves: - reasoning effort feature - Python 3.14 Also add `image_sizes` in `apply_chat_template` output to match what is expected by standard processor.",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44621",
-    "created_at": "2026-03-12T08:04:29Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44656",
+    "created_at": "2026-03-13T10:24:11Z",
+    "deletions": 15,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44621/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44621",
+    "files_url": "https://github.com/huggingface/transformers/pull/44656/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44656",
     "labels": [],
     "merged": true,
-    "number": 44621,
-    "review_comments_count": 0,
+    "number": 44656,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "pipelines do not have modelcard",
-    "updated_at": "2026-03-13T14:28:48Z"
+    "title": "[MistralCommonBackend] Upgrade mistral-common to v1.10.0",
+    "updated_at": "2026-03-16T12:46:29Z"
   },
   {
-    "additions": 15,
-    "author": "LysandreJik",
-    "author_association": "MEMBER",
-    "body_excerpt": "FastAPI doesn't play well with `from __future__ import annotations`. This PR reverts this change and correctly guards against unprotected optional imports. Reverts https://github.com/huggingface/transformers/pull/44256",
+    "additions": 13,
+    "author": "jiqing-feng",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "This PR fixes the AWQ test suite to align with the GPTQModel migration (following #41567 and #42776). ### Changes - **Fix `replace_with_awq_linear` return value**: The function now returns the model directly instead of a tuple `(model, _)`\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44620",
-    "created_at": "2026-03-12T07:56:55Z",
-    "deletions": 16,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44654",
+    "created_at": "2026-03-13T07:31:19Z",
+    "deletions": 13,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44620/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44620",
+    "files_url": "https://github.com/huggingface/transformers/pull/44654/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44654",
     "labels": [],
     "merged": true,
-    "number": 44620,
+    "number": 44654,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix transformers serve's 422 unprocessable entity",
-    "updated_at": "2026-03-16T13:41:44Z"
+    "title": "Fix AWQ tests for GPTQModel migration",
+    "updated_at": "2026-03-13T16:28:14Z"
   },
   {
-    "additions": 43,
-    "author": "yunhaoli24",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Set `add_bos_token=True` and `add_eos_token=True` by default in `DebertaV2Tokenizer` to fix the regression where `add_special_tokens=True` doesn't add BOS/EOS tokens for `microsoft/mdeberta-v3-base` tokenizer in transformers >=5\u2026",
-    "changed_files": 2,
+    "additions": 18,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@zucchini-nlp, can you help review? Thx! unit tests to reproduce this bug: `tests/models/phi4_multimodal/test_modeling_phi4_multimodal.py::Phi4MultimodalIntegrationTest::test_audio_text_generation`",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44618",
-    "created_at": "2026-03-12T04:46:19Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44653",
+    "created_at": "2026-03-13T07:14:25Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44618/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44618",
+    "files_url": "https://github.com/huggingface/transformers/pull/44653/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44653",
     "labels": [],
     "merged": false,
-    "number": 44618,
-    "review_comments_count": 0,
+    "number": 44653,
+    "review_comments_count": 7,
     "state": "closed",
-    "title": "fix: Add BOS/EOS tokens by default for DeBERTa v2 tokenizer",
-    "updated_at": "2026-03-16T05:28:25Z"
+    "title": "Fix `AutoImageProcessor` to correctly detect local implementation whe\u2026",
+    "updated_at": "2026-04-09T02:32:53Z"
   },
   {
-    "additions": 7,
-    "author": "s-zx",
+    "additions": 8,
+    "author": "gambletan",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Add fallback to bfloat16 when Float8 dtype fails to set, preventing TypeError when loading FP8 models on PyTorch builds without Float8_e4m3fnStorage support. ## Root Cause `torch.set_default_dtype(dtype)` raises `TypeError: coul\u2026",
+    "body_excerpt": "## Summary Fixes #44623 `processor.save_pretrained()` in v5 only saves the unified `processor_config.json` with nested sub-processor configs, but does not save standalone config files like `preprocessor_config.json` for the image processor\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44616",
-    "created_at": "2026-03-11T23:00:15Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44652",
+    "created_at": "2026-03-13T05:38:10Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44616/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44616",
+    "files_url": "https://github.com/huggingface/transformers/pull/44652/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44652",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44616,
+    "number": 44652,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: add Float8 dtype fallback in modeling_utils.py",
-    "updated_at": "2026-03-18T16:02:54Z"
-  },
-  {
-    "additions": 35,
-    "author": "MaybeSam05",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Restores `is_torch_fx_available` in `transformers.utils.import_utils` as a backwards-compatibility shim so that Hub models loaded with `trust_remote_code=True` that still import this symbol no longer raise `ImportEr\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44615",
-    "created_at": "2026-03-11T22:52:23Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44615/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44615",
-    "labels": [],
-    "merged": false,
-    "number": 44615,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Restore is_torch_fx_available for trust_remote_code backwards compatibility",
-    "updated_at": "2026-03-12T10:33:43Z"
+    "title": "Fix processor.save_pretrained missing sub-processor config files",
+    "updated_at": "2026-03-13T12:03:37Z"
   },
   {
-    "additions": 19,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "Only detected in #43590, so it can only be detected there for `test_sdpa_can_compile_dynamic` (`lw_detr`). Core issue: Dynamo can cache the attribute and ignore it across frames which means that updates/reads are not working as expected. T\u2026",
+    "additions": 10,
+    "author": "gambletan",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #44625 When passing `num_labels` to `AutoConfig.from_pretrained` for Qwen3.5, the value is set on the outer `Qwen3_5Config` but not propagated to `text_config`. This causes `AutoModelForSequenceClassification` to use the d\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44614",
-    "created_at": "2026-03-11T20:49:51Z",
-    "deletions": 22,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44614/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44614",
-    "labels": [],
-    "merged": false,
-    "number": 44614,
-    "review_comments_count": 10,
-    "state": "closed",
-    "title": "[`Compile`] Fix capture outputs during compile",
-    "updated_at": "2026-03-13T02:15:21Z"
-  },
-  {
-    "additions": 105,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "adds docs for tensor parallelism for training",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44613",
-    "created_at": "2026-03-11T20:43:53Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44651",
+    "created_at": "2026-03-13T05:35:29Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44613/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44613",
+    "files_url": "https://github.com/huggingface/transformers/pull/44651/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44651",
     "labels": [],
-    "merged": true,
-    "number": 44613,
+    "merged": false,
+    "number": 44651,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] tp training",
-    "updated_at": "2026-04-09T22:20:47Z"
+    "title": "Fix Qwen3.5 num_labels not propagated to text_config",
+    "updated_at": "2026-03-13T13:39:43Z"
   },
   {
-    "additions": 1,
-    "author": "michaelbenayoun",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The output function hook in `EmbeddingParallel` casts the mask to fp32. It breaks things for neuron devices. Suggested fix: cast to the outputs' dtype.",
-    "changed_files": 1,
+    "additions": 188,
+    "author": "shaealh",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Closes #44593 ## Summary - use generation_input_ids/generation_attention_mask when provided for decoder-only models - otherwise infer prompt from leading -100 labels and build left-padded prompt batch - return completion tokens for decoder\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44612",
-    "created_at": "2026-03-11T20:09:41Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44650",
+    "created_at": "2026-03-13T05:33:59Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44612/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44612",
+    "files_url": "https://github.com/huggingface/transformers/pull/44650/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44650",
     "labels": [],
-    "merged": true,
-    "number": 44612,
-    "review_comments_count": 3,
-    "state": "closed",
-    "title": "fix: cast to proper dtype in EmbeddingParallel",
-    "updated_at": "2026-03-12T21:08:04Z"
+    "merged": false,
+    "number": 44650,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Fix Seq2SeqTrainer generation path for decoder-only models",
+    "updated_at": "2026-04-02T21:23:53Z"
   },
   {
-    "additions": 15,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following issue was identified and fixed in this PR: \u2192 [This PR (\ud83d\udea8 Delete duplicate code in backbone utils)](https://github.com/huggingface/transformers/pull/43323) structured config loading to use [BackboneMi\u2026",
-    "changed_files": 2,
+    "additions": 12,
+    "author": "gambletan",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #44623 In v5.3.0, `ProcessorMixin.save_pretrained()` no longer calls `save_pretrained()` on non-tokenizer components (e.g. `image_processor`, `feature_extractor`). This means their individual config files (e.g. `preprocess\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44611",
-    "created_at": "2026-03-11T20:02:14Z",
-    "deletions": 3,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44649",
+    "created_at": "2026-03-13T05:22:42Z",
+    "deletions": 207,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44611/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44611",
+    "files_url": "https://github.com/huggingface/transformers/pull/44649/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44649",
     "labels": [],
-    "merged": true,
-    "number": 44611,
-    "review_comments_count": 12,
+    "merged": false,
+    "number": 44649,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(models): Forward timm model kwargs to timm.create_model for OmDet-Turbo",
-    "updated_at": "2026-03-13T11:57:20Z"
+    "title": "fix: ensure processor.save_pretrained saves all component files",
+    "updated_at": "2026-03-13T05:36:54Z"
   },
   {
     "additions": 0,
-    "author": "michaelbenayoun",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? By default, `initialize_tensor_parallelism` hides stdout and stderr for ranks > 0. While convenient, this is not perfect for dev and debugging. I suggest we simply add a flag to be able to disable this feature if wa\u2026",
-    "changed_files": 1,
+    "author": "gambletan",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #44625 When `num_labels` or `id2label` are passed to `Qwen3_5Config` (e.g. via `AutoConfig.from_pretrained(model_name, num_labels=1)`), they are only set on the outer composite config but **not forwarded to `text_config`**\u2026",
+    "changed_files": 0,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44608",
-    "created_at": "2026-03-11T18:57:01Z",
-    "deletions": 7,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44648",
+    "created_at": "2026-03-13T05:22:03Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44608/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44608",
+    "files_url": "https://github.com/huggingface/transformers/pull/44648/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44648",
     "labels": [],
-    "merged": true,
-    "number": 44608,
-    "review_comments_count": 2,
+    "merged": false,
+    "number": 44648,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Allow to disable stdout hiding for TP",
-    "updated_at": "2026-03-12T19:36:06Z"
+    "title": "fix: propagate num_labels to text_config in Qwen3.5",
+    "updated_at": "2026-03-13T05:33:26Z"
   },
   {
     "additions": 9,
-    "author": "gabe-l-hart",
+    "author": "kaixuanliu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR fixes generation for models using the `Idefics3ForConditionalGeneration` architecture with `use_cache=False`. ## Testing <details> <summary>docling_repro.py</summary> ```py import os import torch import time\u2026",
-    "changed_files": 3,
+    "body_excerpt": "@remi-or @ArthurZucker @McPatate pls help review, thx!",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44607",
-    "created_at": "2026-03-11T18:41:58Z",
-    "deletions": 3,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44647",
+    "created_at": "2026-03-13T04:55:26Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44607/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44607",
+    "files_url": "https://github.com/huggingface/transformers/pull/44647/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44647",
     "labels": [],
-    "merged": true,
-    "number": 44607,
+    "merged": false,
+    "number": 44647,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Idefics3 without cache fix",
-    "updated_at": "2026-03-16T15:23:34Z"
+    "title": "Add more generic device support for continuous batching",
+    "updated_at": "2026-04-09T02:32:51Z"
   },
   {
-    "additions": 26,
-    "author": "itazap",
-    "author_association": "MEMBER",
-    "body_excerpt": "In v5, we enforce creating a model-specific tokenizer (ex. LlamaTokenizer, Qwen2Tokenizer, et .) object when specified. 1. For instance, when `tokenizer_class` is set in `tokenization_config.json` 2. Or when using the auto_mapped `tokenize\u2026",
-    "changed_files": 1,
+    "additions": 4,
+    "author": "LincolnBurrows2017",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixed typo in Qwen3.5 and Qwen3-VL-MoE model files ```diff - Since Qwen3.5 use timestamps to seperate videos + Since Qwen3.5 use timestamps to separate videos ```",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44606",
-    "created_at": "2026-03-11T17:29:12Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44646",
+    "created_at": "2026-03-13T04:48:06Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44606/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44606",
+    "files_url": "https://github.com/huggingface/transformers/pull/44646/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44646",
     "labels": [],
     "merged": false,
-    "number": 44606,
+    "number": 44646,
     "review_comments_count": 0,
     "state": "open",
-    "title": "optionally override tokenizer class with serialized tokenizer ",
-    "updated_at": "2026-03-17T16:03:19Z"
+    "title": "Fix typo: seperate -> separate",
+    "updated_at": "2026-03-13T15:27:08Z"
   },
   {
-    "additions": 564,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "* refactors DeepSpeed ZeRO doc: - moves the troubleshooting section to the more general Debugging guide to keep everything in one place - moves the sequence parallelism section into a new doc to give it more visibility - update to be more\u2026",
-    "changed_files": 4,
+    "additions": 4,
+    "author": "sywangyi",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "- Intel XPU: @IlyasMoutawwakil ``` ======================================================================== FAILURES ======================================================================== _________________________________________________\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44605",
-    "created_at": "2026-03-11T17:26:12Z",
-    "deletions": 1163,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44645",
+    "created_at": "2026-03-13T02:53:31Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44605/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44605",
+    "files_url": "https://github.com/huggingface/transformers/pull/44645/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44645",
     "labels": [],
     "merged": true,
-    "number": 44605,
-    "review_comments_count": 6,
+    "number": 44645,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] zero + sequence parallelism",
-    "updated_at": "2026-04-10T16:46:49Z"
+    "title": "fix Image.open failure in case \"tests/models/prompt_depth_anything/te\u2026",
+    "updated_at": "2026-03-27T11:11:05Z"
   },
   {
-    "additions": 415,
-    "author": "SunMarc",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR fix the quantization CI : - [x] autoround - [x] bnb - [x] fp_quant_integration - [x] metal - [x] mxfp4 - [x] quark_integration - [x] torchao_integration One major point in this PR is that I bump the min vers\u2026",
-    "changed_files": 19,
+    "additions": 2,
+    "author": "sywangyi",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "\u2026MXFP4PseudoquantTest::test_quantized_model fail in xpu ## Who can review? - quantization: @SunMarc - Intel XPU: @IlyasMoutawwakil",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 33,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44604",
-    "created_at": "2026-03-11T16:44:38Z",
-    "deletions": 912,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44644",
+    "created_at": "2026-03-13T02:02:45Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44604/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44604",
+    "files_url": "https://github.com/huggingface/transformers/pull/44644/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44644",
     "labels": [],
     "merged": true,
-    "number": 44604,
-    "review_comments_count": 0,
+    "number": 44644,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "Bump torchao >=0.15 and fix quantization CI",
-    "updated_at": "2026-03-16T16:07:12Z"
+    "title": "fix tests/quantization/fp_quant_integration/test_fp_quant.py::FPQuant\u2026",
+    "updated_at": "2026-03-27T14:14:13Z"
   },
   {
-    "additions": 17,
-    "author": "michalrzak",
+    "additions": 5,
+    "author": "joshuaswanson",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? A quick fix that updates the Dockerfile to run on `arm64` systems (such as the NVIDIA Spark). The previous version of the Dockerfile fails on `arm64` systems due to `SudachiPy`, which only provides wheels for `x86_6\u2026",
+    "body_excerpt": "The generic `labels` docstring in `ModelArgs` says \"masked language modeling loss\" and doesn't mention that causal LM models shift labels internally. This has tripped up a lot of users who pre-shift their labels and end up training next-ne\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44603",
-    "created_at": "2026-03-11T16:42:30Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44642",
+    "created_at": "2026-03-12T23:47:11Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44603/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44603",
+    "files_url": "https://github.com/huggingface/transformers/pull/44642/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44642",
     "labels": [],
     "merged": false,
-    "number": 44603,
+    "number": 44642,
     "review_comments_count": 0,
     "state": "open",
-    "title": "fixed dockerfile for arm64 systems",
-    "updated_at": "2026-03-11T16:42:30Z"
-  },
-  {
-    "additions": 218,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. Follow up of https://github.com/huggingface/transformers/pull/44330 Also take the opportunity to simplify t5 and its children, because the way they compute`position_bias` was super convoluted/overc\u2026",
-    "changed_files": 61,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44602",
-    "created_at": "2026-03-11T16:19:43Z",
-    "deletions": 1083,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44602/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44602",
-    "labels": [],
-    "merged": true,
-    "number": 44602,
-    "review_comments_count": 14,
-    "state": "closed",
-    "title": "Remove `cache_position` in more models (2)",
-    "updated_at": "2026-03-12T22:38:15Z"
+    "title": "Clarify that causal LM labels are shifted internally",
+    "updated_at": "2026-03-13T00:02:30Z"
   },
   {
-    "additions": 510,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Goal is to be able to run a model with both PP and TP.",
-    "changed_files": 4,
+    "additions": 1,
+    "author": "kmbhattt-aws",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Issue: A full 4D attention mask of shape `[1, 1, seq_len, seq_len]` is being created during attention, even when not using alibi for positional embeddings. - This occupied extra memory during training. Root Cause: T\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44601",
-    "created_at": "2026-03-11T15:56:51Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44641",
+    "created_at": "2026-03-12T23:01:11Z",
+    "deletions": 1,
     "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44601/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44601",
+    "files_url": "https://github.com/huggingface/transformers/pull/44641/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44641",
     "labels": [],
     "merged": false,
-    "number": 44601,
-    "review_comments_count": 0,
+    "number": 44641,
+    "review_comments_count": 1,
     "state": "open",
-    "title": "[Distributed] Add PP support natively",
-    "updated_at": "2026-03-12T11:53:24Z"
+    "title": "Conditinally passing and_mask_function arg to create_causal_mask ",
+    "updated_at": "2026-03-13T02:09:22Z"
   },
   {
-    "additions": 0,
-    "author": "Cyrilvallez",
+    "additions": 11,
+    "author": "michaelbenayoun",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Small mistake in https://github.com/huggingface/transformers/pull/44432. cc @zucchini-nlp, was it intended to remove the scaling? (I assume so since the embedding now has the saling baked in, and I guess paligemma a\u2026",
+    "body_excerpt": "# What does this PR do? - Add `\"embedding_colwise\"` tp plan - Add register methods for `ParallelInterface`. Without it, we can register plans with the register method, but not the corresponding `plan_to_weight_dim` and `plan_to_bias_dim`.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44600",
-    "created_at": "2026-03-11T15:40:16Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44640",
+    "created_at": "2026-03-12T20:14:06Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44600/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44600",
+    "files_url": "https://github.com/huggingface/transformers/pull/44640/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44640",
     "labels": [],
     "merged": true,
-    "number": 44600,
+    "number": 44640,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Remove useless identity assignment",
-    "updated_at": "2026-03-12T10:21:23Z"
+    "title": "Add register method for `ParallelInterface`",
+    "updated_at": "2026-03-13T18:57:48Z"
   },
   {
-    "additions": 3,
-    "author": "hf-security-analysis[bot]",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Update `.github/workflows/self-comment-ci.yml` workflow configuration. cc @vasqu Closes huggingface/tracking-issues#17",
-    "changed_files": 1,
+    "additions": 24099,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "## Summary Fixes bugs introduced during the `__init__` \u2192 `@dataclass` conversion in #41250. All are incorrect default values caught by automated comparison of `__init__` signatures (main) vs dataclass fields (PR). | Model | Param | Was | S\u2026",
+    "changed_files": 931,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44599",
-    "created_at": "2026-03-11T15:39:30Z",
-    "deletions": 161,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44639",
+    "created_at": "2026-03-12T16:49:54Z",
+    "deletions": 38773,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44599/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44599",
+    "files_url": "https://github.com/huggingface/transformers/pull/44639/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44639",
     "labels": [],
     "merged": false,
-    "number": 44599,
+    "number": 44639,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "chore: update self-comment-ci.yml",
-    "updated_at": "2026-04-02T09:05:25Z"
+    "title": "Fix incorrect default values in config dataclass migration",
+    "updated_at": "2026-03-12T16:50:10Z"
   },
   {
-    "additions": 1,
-    "author": "vasqu",
+    "additions": 19,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "As per title, updated to remove former members as well cc @yonigozlan @ydshieh",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? As per the title. Supersedes https://github.com/huggingface/transformers/pull/44614. This one is the result of a long debugging session and discussion with @vasqu. The issue is as follow: - Backbone ALWAYS need to c\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44598",
-    "created_at": "2026-03-11T15:13:25Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44638",
+    "created_at": "2026-03-12T16:19:49Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44598/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44598",
+    "files_url": "https://github.com/huggingface/transformers/pull/44638/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44638",
     "labels": [],
     "merged": true,
-    "number": 44598,
-    "review_comments_count": 0,
+    "number": 44638,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Add Yoni to run-slow workflow",
-    "updated_at": "2026-03-11T15:38:10Z"
+    "title": "Fix output capturing for Backbones",
+    "updated_at": "2026-03-12T17:11:32Z"
   },
   {
-    "additions": 1,
-    "author": "ydshieh",
+    "additions": 571,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Our beautiful Dashboard is missing ..... damm",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? - Adds an `httpx` tracer to gather metrics about network calls - Collect and store metrics and generates an artifact in CI - Can be used locally with `DEBUG_NETWORK` - Activated in CircleCI example of local run: ```\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44597",
-    "created_at": "2026-03-11T13:53:02Z",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44636",
+    "created_at": "2026-03-12T15:25:10Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44597/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44597",
+    "files_url": "https://github.com/huggingface/transformers/pull/44636/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44636",
     "labels": [],
     "merged": true,
-    "number": 44597,
-    "review_comments_count": 0,
+    "number": 44636,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "Fix CircleCI summary report not showing due to missing dependency",
-    "updated_at": "2026-03-20T07:33:38Z"
+    "title": "feat(ci): added a network debug report",
+    "updated_at": "2026-03-18T19:19:03Z"
   },
   {
-    "additions": 26,
-    "author": "Desel72",
-    "author_association": "NONE",
-    "body_excerpt": "Fix TypeError when loading float8 models by falling back to bfloat16 in local_torch_dtype # What does this PR do? When loading FP8 models (e.g. `Qwen/Qwen3.5-35B-A3B-FP8`) with `dtype=\"auto\"`, the auto-detected dtype from checkpoint weight\u2026",
-    "changed_files": 2,
+    "additions": 111,
+    "author": "RyanMullins",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Per a conversion with @Cyrilvallez on Slack on how to make Gemma models more compatible with modular inheritance, this PR: * Updates Gemma models to use `nn.parameter.Buffer` instead of a `self.register_buffer()` fo\u2026",
+    "changed_files": 32,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44596",
-    "created_at": "2026-03-11T13:03:19Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44635",
+    "created_at": "2026-03-12T14:47:46Z",
+    "deletions": 87,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44596/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44596",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44635/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44635",
+    "labels": [],
     "merged": false,
-    "number": 44596,
+    "number": 44635,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix TypeError when loading float8 models by falling back to bfloat16 in local_torch_dtype",
-    "updated_at": "2026-03-11T14:00:39Z"
+    "state": "open",
+    "title": "[Gemma] Modular-friendly buffers",
+    "updated_at": "2026-03-18T10:44:25Z"
   },
   {
-    "additions": 2324,
-    "author": "yonigozlan",
+    "additions": 30,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Add [CHMv2 ](https://arxiv.org/abs/2603.06382) to Transformers",
-    "changed_files": 23,
+    "body_excerpt": "# What does this PR do? As per the title",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44595",
-    "created_at": "2026-03-11T12:38:44Z",
-    "deletions": 23,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44634",
+    "created_at": "2026-03-12T14:04:36Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44595/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44595",
-    "labels": [
-      "New model",
-      "run-slow"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44634/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44634",
+    "labels": [],
     "merged": true,
-    "number": 44595,
-    "review_comments_count": 30,
+    "number": 44634,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Add CHMv2",
-    "updated_at": "2026-03-11T16:00:03Z"
+    "title": "Fix lfm2 kernel path",
+    "updated_at": "2026-03-12T15:00:59Z"
   },
   {
-    "additions": 271,
-    "author": "vimal-crypto",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What this PR does This PR brings `ObjectDetectionPipeline` in line with its sister pipelines (`ZeroShotObjectDetectionPipeline`, `ImageClassificationPipeline`) by adding four enhancements to the postprocessing stage. ### Changes **1. Sc\u2026",
-    "changed_files": 2,
+    "additions": 26,
+    "author": "eustlb",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per title \ud83e\udd17",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44594",
-    "created_at": "2026-03-11T12:37:46Z",
-    "deletions": 40,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44633",
+    "created_at": "2026-03-12T13:35:48Z",
+    "deletions": 11,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44594/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44594",
+    "files_url": "https://github.com/huggingface/transformers/pull/44633/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44633",
     "labels": [],
-    "merged": false,
-    "number": 44594,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "[Pipeline] Add top_k, label filtering, box_format and score sorting to ObjectDetectionPipeline",
-    "updated_at": "2026-03-11T12:37:46Z"
+    "merged": true,
+    "number": 44633,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "[medasr] doc update",
+    "updated_at": "2026-03-16T09:39:50Z"
   },
   {
-    "additions": 15,
-    "author": "BenjaminBossan",
+    "additions": 35,
+    "author": "Abdennacer-Badaoui",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Multiple PEFT tests are failing due to recent changes in transformers. - hf_device_map attribute may not exist in some cases - respect inference_mode in load_adapter - new model loading requires changes for bnb (SCB\u2026",
-    "changed_files": 3,
+    "body_excerpt": "Summary - Update test expectations for affected models - Add some needed dependencies - Fix TypeError: `GenerationMixin.prepare_inputs_for_generation()` got multiple values for argument 'next_sequence_length' in Qwen2.5-Omni talker by pass\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44592",
-    "created_at": "2026-03-11T10:41:51Z",
-    "deletions": 6,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44632",
+    "created_at": "2026-03-12T13:32:33Z",
+    "deletions": 22,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44592/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44592",
+    "files_url": "https://github.com/huggingface/transformers/pull/44632/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44632",
     "labels": [],
     "merged": true,
-    "number": 44592,
-    "review_comments_count": 0,
+    "number": 44632,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "FIX Multiple PEFT errors after v5 transition",
-    "updated_at": "2026-03-11T12:24:05Z"
+    "title": "[AMD CI] Fix test failures across important models ",
+    "updated_at": "2026-03-17T14:58:10Z"
   },
   {
-    "additions": 60,
-    "author": "moktamd",
-    "author_association": "NONE",
-    "body_excerpt": "Adds `_apply_mps_fixes` in `sdpa_attention.py` to handle two upstream PyTorch MPS bugs: 1. **pytorch/pytorch#176767** (fixed in PyTorch 2.12): pads value tensor when `v_head_dim != q_head_dim` to avoid corrupted output. Affects DeepSeek mo\u2026",
-    "changed_files": 1,
+    "additions": 33,
+    "author": "RyanMullins",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Updates the weights conversion scripts for Gemma to: * Use the new `SentencePieceExtractor` class to get the vocab and merges from the SPM * Always initialize and save the unified `GemmaTokenizer` class ## Before su\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44591",
-    "created_at": "2026-03-11T10:32:26Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44631",
+    "created_at": "2026-03-12T13:32:25Z",
+    "deletions": 45,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44591/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44591",
+    "files_url": "https://github.com/huggingface/transformers/pull/44631/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44631",
     "labels": [],
-    "merged": false,
-    "number": 44591,
+    "merged": true,
+    "number": 44631,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add MPS SDPA workarounds for value head dim and bidirectional attention",
-    "updated_at": "2026-03-11T13:37:15Z"
+    "title": "[Gemma] Update conversion scripts for Transformers v5 Comaptibility",
+    "updated_at": "2026-03-18T10:39:53Z"
   },
   {
-    "additions": 2,
-    "author": "pranay-3108",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes incorrect documentation for `position_ids` in `masking_utils.py`. The docstring previously described `position_ids` as `torch.Tensor`. This PR updates it to `torch.LongTensor` and aligns the description with the standard wording used\u2026",
-    "changed_files": 1,
+    "additions": 42,
+    "author": "MaybeSam05",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes a bug where `num_labels` passed to `AutoConfig.from_pretrained` for Qwen3.5 did not propagate from the top\u2011level `Qwen3_5Config` into the `text_config`, so `AutoModelForSequenceClassification` still saw the de\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44590",
-    "created_at": "2026-03-11T05:13:57Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44630",
+    "created_at": "2026-03-12T13:25:54Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44590/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44590",
+    "files_url": "https://github.com/huggingface/transformers/pull/44630/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44630",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44590,
+    "number": 44630,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix incorrect docstring for position_ids",
-    "updated_at": "2026-03-11T21:08:42Z"
+    "title": "Fix Qwen3.5 num_labels propagation to text_config (fix #44625)",
+    "updated_at": "2026-03-12T13:46:07Z"
   },
   {
-    "additions": 1,
-    "author": "jiqing-feng",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes Llama4 model loading under BitsAndBytes (BNB) quantization mode. Router quantized incorrectly causes shape mismatch: Llama4Router inherits from nn.Linear, so BNB quantizes its weight into a packed format. However, super().forward() c\u2026",
-    "changed_files": 1,
+    "additions": 15,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes \"auto\" dtype when the model is initialized `from_config` It was already fixed for `from_pretrained` in https://github.com/huggingface/transformers/pull/42990 but vLLM creates models with `AutoModel._from_confi\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 15,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44588",
-    "created_at": "2026-03-11T01:42:33Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44629",
+    "created_at": "2026-03-12T13:07:55Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44588/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44588",
+    "files_url": "https://github.com/huggingface/transformers/pull/44629/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44629",
     "labels": [],
     "merged": true,
-    "number": 44588,
-    "review_comments_count": 4,
+    "number": 44629,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Fix llama4 bnb mode",
-    "updated_at": "2026-03-27T14:19:14Z"
+    "title": "Ensure same `dtype` for subconfig when `_from_config`",
+    "updated_at": "2026-03-13T11:35:10Z"
   },
   {
-    "additions": 32,
-    "author": "kmbhattt-aws",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 2,
+    "additions": 37,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? - `encoder_config` and `decoder_config` should return `None` for encoder / decoder config classes themselves. - The encoder / decoder model classes should have the correct config classes associated to them <!-- CURS\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44587",
-    "created_at": "2026-03-11T01:01:18Z",
-    "deletions": 20,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44587/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44587",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44628",
+    "created_at": "2026-03-12T12:24:44Z",
+    "deletions": 5,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44628/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44628",
     "labels": [],
-    "merged": false,
-    "number": 44587,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix: Handling fused qkv result tensor slicing for tp sharded qkv weights",
-    "updated_at": "2026-03-12T21:31:29Z"
+    "merged": true,
+    "number": 44628,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "Fix for `VibeVoiceAcousticTokenizer`",
+    "updated_at": "2026-03-12T13:33:02Z"
   },
   {
-    "additions": 91,
-    "author": "mvanhorn",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Decouples router logits collection from output visibility in Mixtral's `ForCausalLM`. Previously, `output_router_logits=False` (the default) prevented `aux_loss` from being computed, meaning load balancing was silen\u2026",
-    "changed_files": 13,
+    "additions": 141,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? The diff in revert mapping is needed, otherwise we get failures in a few models, see https://app.circleci.com/pipelines/github/huggingface/transformers/167425/workflows/fa96efe5-f810-408e-bafd-de03b7e881aa/jobs/2208\u2026",
+    "changed_files": 78,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44586",
-    "created_at": "2026-03-11T00:24:07Z",
-    "deletions": 39,
+    "comments_count": 16,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44627",
+    "created_at": "2026-03-12T12:00:31Z",
+    "deletions": 367,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44586/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44586",
+    "files_url": "https://github.com/huggingface/transformers/pull/44627/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44627",
     "labels": [],
-    "merged": false,
-    "number": 44586,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44627,
+    "review_comments_count": 8,
     "state": "closed",
-    "title": "Fix Mixtral aux_loss not computed when output_router_logits=False",
-    "updated_at": "2026-03-11T14:31:21Z"
+    "title": "Move VLM conversions to the main mapping",
+    "updated_at": "2026-03-17T10:13:03Z"
   },
   {
-    "additions": 10,
-    "author": "mvanhorn",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Passes `eps=config.rms_norm_eps` to both `q_a_layernorm` and `kv_a_layernorm` in the DeepseekV3 MLA attention module. Without this, these layernorms default to `eps=1e-5` instead of the config value (`1e-6`), causin\u2026",
-    "changed_files": 5,
+    "additions": 11,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Adds a missing branch. I don't really know if this is worth it, can't find a model online that enforces the flag to `True`",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44585",
-    "created_at": "2026-03-11T00:20:54Z",
-    "deletions": 10,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44626",
+    "created_at": "2026-03-12T11:23:21Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44585/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44585",
+    "files_url": "https://github.com/huggingface/transformers/pull/44626/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44626",
     "labels": [],
     "merged": false,
-    "number": 44585,
-    "review_comments_count": 0,
+    "number": 44626,
+    "review_comments_count": 1,
     "state": "open",
-    "title": "Fix missing rms_norm_eps in DeepseekV3 MLA layernorms",
-    "updated_at": "2026-03-12T14:39:12Z"
+    "title": "don't break legacy behavior when enforced!",
+    "updated_at": "2026-03-12T11:32:46Z"
   },
   {
-    "additions": 1,
-    "author": "mvanhorn",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes an off-by-one error in `decode_spans()` in the document question answering pipeline that causes a `ValueError: kth(=N) out of bounds` crash when `len(scores_flat) == topk`. The boundary check on line 97 uses `\u2026",
-    "changed_files": 1,
+    "additions": 34,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Follow-up of #44549",
+    "changed_files": 16,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44584",
-    "created_at": "2026-03-10T23:52:51Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44624",
+    "created_at": "2026-03-12T09:26:17Z",
+    "deletions": 34,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44584/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44584",
+    "files_url": "https://github.com/huggingface/transformers/pull/44624/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44624",
     "labels": [],
     "merged": true,
-    "number": 44584,
+    "number": 44624,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix off-by-one in decode_spans boundary check",
-    "updated_at": "2026-03-12T13:22:10Z"
+    "title": "Fix more wrong HF hub checkpoint names",
+    "updated_at": "2026-03-12T09:59:12Z"
   },
   {
-    "additions": 45,
-    "author": "wilnn",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "\u2026kpoint when `save_strategy` is `best` # What does this PR do? fix load_best_model_checkpoint_at_end do not load the best model checkpoint at the end when `save_strategy` is `\"best\"` Fixes # (issue) fix load_best_model_checkpoint_at_end do\u2026",
-    "changed_files": 5,
+    "additions": 17,
+    "author": "LysandreJik",
+    "author_association": "MEMBER",
+    "body_excerpt": "CB temporarily disabled on non-cuda devices as it's currently using cuda streams by default.",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44583",
-    "created_at": "2026-03-10T22:37:36Z",
-    "deletions": 16,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44622",
+    "created_at": "2026-03-12T08:11:10Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44583/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44583",
+    "files_url": "https://github.com/huggingface/transformers/pull/44622/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44622",
     "labels": [],
-    "merged": true,
-    "number": 44583,
-    "review_comments_count": 2,
+    "merged": false,
+    "number": 44622,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix load_best_model_checkpoint_at_end do not load the best model chec\u2026",
-    "updated_at": "2026-03-24T15:42:26Z"
+    "title": "CB temporarily disabled on non-cuda devices",
+    "updated_at": "2026-03-16T00:00:20Z"
   },
   {
-    "additions": 3,
-    "author": "yonigozlan",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix an issue introduced in #42564 . The refactor embedded raw image tokens instead of BPE tokens, causing the model to output gibberish. This fix adds back the image tokens to BPE tokens conversion before embedding.\u2026",
+    "additions": 0,
+    "author": "KoichiYasuoka",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #43170 for `modelcard` removal Quick reproduce: ``` from transformers import pipeline fmp=pipeline(\"fill-mask\",\"google-bert/bert-base-cased\") fmp.save_pretrained(\"tmpdir\") ``` ## Before submitting - [ ] This P\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44582",
-    "created_at": "2026-03-10T21:00:30Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44582/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44582",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44621",
+    "created_at": "2026-03-12T08:04:29Z",
+    "deletions": 3,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44621/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44621",
     "labels": [],
     "merged": true,
-    "number": 44582,
+    "number": 44621,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix missing BPE token conversion step in Chameleon",
-    "updated_at": "2026-03-11T11:26:49Z"
+    "title": "pipelines do not have modelcard",
+    "updated_at": "2026-03-13T14:28:48Z"
   },
   {
-    "additions": 9,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following issues were identified and fixed in this PR: \u2192 **Llama-4 Vision:** [freqs_ci is stored as a plain attr](https://github.com/huggingface/transformers/blob/153894c013/src/transformers/models/llama4/mode\u2026",
-    "changed_files": 2,
+    "additions": 15,
+    "author": "LysandreJik",
+    "author_association": "MEMBER",
+    "body_excerpt": "FastAPI doesn't play well with `from __future__ import annotations`. This PR reverts this change and correctly guards against unprotected optional imports. Reverts https://github.com/huggingface/transformers/pull/44256",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44581",
-    "created_at": "2026-03-10T19:33:51Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44620",
+    "created_at": "2026-03-12T07:56:55Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44581/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44581",
+    "files_url": "https://github.com/huggingface/transformers/pull/44620/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44620",
     "labels": [],
     "merged": true,
-    "number": 44581,
-    "review_comments_count": 2,
+    "number": 44620,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(models, testing): Fix Llama4 vision rotary meta tensor initialization and MyT5 get_tokenizer signature",
-    "updated_at": "2026-03-13T16:13:09Z"
+    "title": "Fix transformers serve's 422 unprocessable entity",
+    "updated_at": "2026-03-16T13:41:44Z"
   },
   {
-    "additions": 16,
-    "author": "rabbierabbie",
+    "additions": 43,
+    "author": "yunhaoli24",
     "author_association": "NONE",
-    "body_excerpt": "This PR clarifies references to the **Transformers library** in the README. While reading the documentation as a new user, I initially found the term **\"Transformers\"** ambiguous because it could refer either to the **Transformer architect\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary Set `add_bos_token=True` and `add_eos_token=True` by default in `DebertaV2Tokenizer` to fix the regression where `add_special_tokens=True` doesn't add BOS/EOS tokens for `microsoft/mdeberta-v3-base` tokenizer in transformers >=5\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44580",
-    "created_at": "2026-03-10T18:21:55Z",
-    "deletions": 16,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44618",
+    "created_at": "2026-03-12T04:46:19Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44580/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44580",
+    "files_url": "https://github.com/huggingface/transformers/pull/44618/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44618",
     "labels": [],
     "merged": false,
-    "number": 44580,
+    "number": 44618,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Clarify references to the Transformers library in README",
-    "updated_at": "2026-03-11T13:24:58Z"
+    "title": "fix: Add BOS/EOS tokens by default for DeBERTa v2 tokenizer",
+    "updated_at": "2026-03-16T05:28:25Z"
   },
   {
-    "additions": 1,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? We need to fetch the specific commit (the so called merge commit created by Github itself)",
+    "additions": 7,
+    "author": "s-zx",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Add fallback to bfloat16 when Float8 dtype fails to set, preventing TypeError when loading FP8 models on PyTorch builds without Float8_e4m3fnStorage support. ## Root Cause `torch.set_default_dtype(dtype)` raises `TypeError: coul\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44579",
-    "created_at": "2026-03-10T16:58:09Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44616",
+    "created_at": "2026-03-11T23:00:15Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44579/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44579",
-    "labels": [],
-    "merged": true,
-    "number": 44579,
+    "files_url": "https://github.com/huggingface/transformers/pull/44616/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44616",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44616,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix PR comment CI for quantization job",
-    "updated_at": "2026-03-10T17:07:11Z"
+    "title": "fix: add Float8 dtype fallback in modeling_utils.py",
+    "updated_at": "2026-03-18T16:02:54Z"
   },
   {
-    "additions": 449,
-    "author": "remi-or",
-    "author_association": "MEMBER",
-    "body_excerpt": "# Summary This PR adds three attributes to the compile config, to have granularity over how varlen (handles mixed prefil and decode batches) and decode (only decode batches) are compiled. We want to have this kind of granularity because va\u2026",
-    "changed_files": 11,
+    "additions": 35,
+    "author": "MaybeSam05",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Restores `is_torch_fx_available` in `transformers.utils.import_utils` as a backwards-compatibility shim so that Hub models loaded with `trust_remote_code=True` that still import this symbol no longer raise `ImportEr\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44578",
-    "created_at": "2026-03-10T16:31:20Z",
-    "deletions": 121,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44615",
+    "created_at": "2026-03-11T22:52:23Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44578/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44578",
+    "files_url": "https://github.com/huggingface/transformers/pull/44615/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44615",
     "labels": [],
-    "merged": true,
-    "number": 44578,
-    "review_comments_count": 4,
-    "state": "closed",
-    "title": "[CB] Better parametrization for compile",
-    "updated_at": "2026-03-19T11:50:08Z"
+    "merged": false,
+    "number": 44615,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Restore is_torch_fx_available for trust_remote_code backwards compatibility",
+    "updated_at": "2026-03-12T10:33:43Z"
   },
   {
-    "additions": 5,
-    "author": "tarekziade",
+    "additions": 19,
+    "author": "vasqu",
     "author_association": "MEMBER",
-    "body_excerpt": "- Fix quantizer_aqlm.py to use renamed modules_to_not_convert parameter instead of removed linear_weights_not_to_quantize - Update test to match new function signature: no tuple return, module names instead of weight names",
+    "body_excerpt": "Only detected in #43590, so it can only be detected there for `test_sdpa_can_compile_dynamic` (`lw_detr`). Core issue: Dynamo can cache the attribute and ignore it across frames which means that updates/reads are not working as expected. T\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44577",
-    "created_at": "2026-03-10T15:57:36Z",
-    "deletions": 5,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44614",
+    "created_at": "2026-03-11T20:49:51Z",
+    "deletions": 22,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44577/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44577",
+    "files_url": "https://github.com/huggingface/transformers/pull/44614/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44614",
     "labels": [],
-    "merged": true,
-    "number": 44577,
-    "review_comments_count": 0,
+    "merged": false,
+    "number": 44614,
+    "review_comments_count": 10,
     "state": "closed",
-    "title": "Fix: AQLM quantizer to match updated replace_with_aqlm_linear signature",
-    "updated_at": "2026-03-10T17:48:00Z"
+    "title": "[`Compile`] Fix capture outputs during compile",
+    "updated_at": "2026-03-13T02:15:21Z"
   },
   {
-    "additions": 16,
-    "author": "SunMarc",
+    "additions": 105,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/44387. This PR disable async loading when we want to quantize the model. it is actually faster than doing a semaphore. If a quantizer happens to quantize fast\u2026",
-    "changed_files": 1,
+    "body_excerpt": "adds docs for tensor parallelism for training",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44576",
-    "created_at": "2026-03-10T15:07:01Z",
-    "deletions": 5,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44613",
+    "created_at": "2026-03-11T20:43:53Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44576/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44576",
+    "files_url": "https://github.com/huggingface/transformers/pull/44613/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44613",
     "labels": [],
     "merged": true,
-    "number": 44576,
-    "review_comments_count": 1,
+    "number": 44613,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Disable async loading when quantizing on the fly",
-    "updated_at": "2026-03-16T16:36:42Z"
+    "title": "[docs] tp training",
+    "updated_at": "2026-04-09T22:20:47Z"
   },
   {
-    "additions": 13,
-    "author": "Rocketknight1",
+    "additions": 1,
+    "author": "michaelbenayoun",
     "author_association": "MEMBER",
-    "body_excerpt": "Some parameters in Tapas are initialized in `__init__()` and not reinitialized in `_init_weights()`, which means that if the model is created on the `meta` device, those parameters do not get a weight initialization. This causes a crash la\u2026",
+    "body_excerpt": "# What does this PR do? The output function hook in `EmbeddingParallel` casts the mask to fp32. It breaks things for neuron devices. Suggested fix: cast to the outputs' dtype.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44575",
-    "created_at": "2026-03-10T14:42:40Z",
-    "deletions": 16,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44612",
+    "created_at": "2026-03-11T20:09:41Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44575/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44575",
+    "files_url": "https://github.com/huggingface/transformers/pull/44612/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44612",
     "labels": [],
     "merged": true,
-    "number": 44575,
-    "review_comments_count": 0,
+    "number": 44612,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Correct Tapas initialization",
-    "updated_at": "2026-03-10T15:14:37Z"
+    "title": "fix: cast to proper dtype in EmbeddingParallel",
+    "updated_at": "2026-03-12T21:08:04Z"
   },
   {
-    "additions": 33,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "Fixes #44448",
-    "changed_files": 4,
+    "additions": 15,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? The following issue was identified and fixed in this PR: \u2192 [This PR (\ud83d\udea8 Delete duplicate code in backbone utils)](https://github.com/huggingface/transformers/pull/43323) structured config loading to use [BackboneMi\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44571",
-    "created_at": "2026-03-10T14:28:22Z",
-    "deletions": 12,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44611",
+    "created_at": "2026-03-11T20:02:14Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44571/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44571",
+    "files_url": "https://github.com/huggingface/transformers/pull/44611/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44611",
     "labels": [],
     "merged": true,
-    "number": 44571,
-    "review_comments_count": 0,
+    "number": 44611,
+    "review_comments_count": 12,
     "state": "closed",
-    "title": "Fix pegasus conversion",
-    "updated_at": "2026-03-18T09:55:00Z"
+    "title": "fix(models): Forward timm model kwargs to timm.create_model for OmDet-Turbo",
+    "updated_at": "2026-03-13T11:57:20Z"
   },
   {
-    "additions": 30,
-    "author": "umbilnm",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? In transformers v5, `DebertaV2Tokenizer` was rewritten to use `TokenizersBackend`, but the `post_processor` responsible for adding `[CLS]`/`[SEP]` tokens was never set. This causes `add_special_tokens=True` to silen\u2026",
-    "changed_files": 2,
+    "additions": 0,
+    "author": "michaelbenayoun",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? By default, `initialize_tensor_parallelism` hides stdout and stderr for ranks > 0. While convenient, this is not perfect for dev and debugging. I suggest we simply add a flag to be able to disable this feature if wa\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44570",
-    "created_at": "2026-03-10T13:37:06Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44608",
+    "created_at": "2026-03-11T18:57:01Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44570/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44570",
+    "files_url": "https://github.com/huggingface/transformers/pull/44608/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44608",
     "labels": [],
     "merged": true,
-    "number": 44570,
-    "review_comments_count": 1,
+    "number": 44608,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix missing post_processor in DebertaV2Tokenizer causing no special t\u2026",
-    "updated_at": "2026-03-24T09:40:44Z"
+    "title": "Allow to disable stdout hiding for TP",
+    "updated_at": "2026-03-12T19:36:06Z"
   },
   {
-    "additions": 267,
-    "author": "aashay-sarvam",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Adds native support for the `sarvam_mla` model type (`sarvamai/sarvam-105b`) to HuggingFace Transformers using the **modular pattern**, inheriting from DeepSeek V3. ### Model Architecture SarvamMLA is a **105B para\u2026",
-    "changed_files": 10,
+    "additions": 9,
+    "author": "gabe-l-hart",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR fixes generation for models using the `Idefics3ForConditionalGeneration` architecture with `use_cache=False`. ## Testing <details> <summary>docling_repro.py</summary> ```py import os import torch import time\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44569",
-    "created_at": "2026-03-10T11:55:01Z",
-    "deletions": 0,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44607",
+    "created_at": "2026-03-11T18:41:58Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44569/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44569",
+    "files_url": "https://github.com/huggingface/transformers/pull/44607/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44607",
     "labels": [],
-    "merged": false,
-    "number": 44569,
-    "review_comments_count": 20,
-    "state": "open",
-    "title": "Add SarvamMLA model (sarvamai/sarvam-105b)",
-    "updated_at": "2026-03-18T13:58:18Z"
+    "merged": true,
+    "number": 44607,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Idefics3 without cache fix",
+    "updated_at": "2026-03-16T15:23:34Z"
   },
   {
-    "additions": 2,
-    "author": "tomaarsen",
+    "additions": 26,
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Currently, when using Apertus (or rather, it's `XIELUActivation`), and you don't have `xielu` installed, then you'll fall to this `except:` https://github.com/huggingface/transformers/blob/5a098a1e01034095f037c8a37f\u2026",
+    "body_excerpt": "In v5, we enforce creating a model-specific tokenizer (ex. LlamaTokenizer, Qwen2Tokenizer, et .) object when specified. 1. For instance, when `tokenizer_class` is set in `tokenization_config.json` 2. Or when using the auto_mapped `tokenize\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44567",
-    "created_at": "2026-03-10T11:39:20Z",
-    "deletions": 3,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44606",
+    "created_at": "2026-03-11T17:29:12Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44567/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44567",
+    "files_url": "https://github.com/huggingface/transformers/pull/44606/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44606",
     "labels": [],
-    "merged": true,
-    "number": 44567,
+    "merged": false,
+    "number": 44606,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[`fix`] Prevent crash with Apertus without xielu installed",
-    "updated_at": "2026-03-10T13:24:11Z"
+    "state": "open",
+    "title": "optionally override tokenizer class with serialized tokenizer ",
+    "updated_at": "2026-03-17T16:03:19Z"
   },
   {
-    "additions": 181,
-    "author": "tarekziade",
+    "additions": 564,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "This patch extends `ty` check to `src/transformers/cli` Based on https://github.com/huggingface/transformers/pull/44412",
-    "changed_files": 8,
+    "body_excerpt": "* refactors DeepSpeed ZeRO doc: - moves the troubleshooting section to the more general Debugging guide to keep everything in one place - moves the sequence parallelism section into a new doc to give it more visibility - update to be more\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44566",
-    "created_at": "2026-03-10T10:40:13Z",
-    "deletions": 86,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44605",
+    "created_at": "2026-03-11T17:26:12Z",
+    "deletions": 1163,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44566/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44566",
+    "files_url": "https://github.com/huggingface/transformers/pull/44605/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44605",
     "labels": [],
-    "merged": false,
-    "number": 44566,
-    "review_comments_count": 21,
+    "merged": true,
+    "number": 44605,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "chore(typing): extend typing to `src/transformers/cli` ",
-    "updated_at": "2026-04-01T16:05:57Z"
+    "title": "[docs] zero + sequence parallelism",
+    "updated_at": "2026-04-10T16:46:49Z"
   },
   {
-    "additions": 36,
-    "author": "Cyrilvallez",
+    "additions": 415,
+    "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. The unused memory is taken into account too late, which leads to different device_map for the same hardware and models, and even random cuda OOM!! Basically, the max memory needs to be adjusted BEF\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? This PR fix the quantization CI : - [x] autoround - [x] bnb - [x] fp_quant_integration - [x] metal - [x] mxfp4 - [x] quark_integration - [x] torchao_integration One major point in this PR is that I bump the min vers\u2026",
+    "changed_files": 19,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44565",
-    "created_at": "2026-03-10T10:31:10Z",
-    "deletions": 22,
+    "comments_count": 33,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44604",
+    "created_at": "2026-03-11T16:44:38Z",
+    "deletions": 912,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44565/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44565",
+    "files_url": "https://github.com/huggingface/transformers/pull/44604/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44604",
     "labels": [],
     "merged": true,
-    "number": 44565,
-    "review_comments_count": 2,
+    "number": 44604,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[device_map] Fix device_map computation by correctly adjusting memory available",
-    "updated_at": "2026-03-10T17:16:01Z"
+    "title": "Bump torchao >=0.15 and fix quantization CI",
+    "updated_at": "2026-03-16T16:07:12Z"
   },
   {
-    "additions": 2,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes #44360",
-    "changed_files": 2,
+    "additions": 17,
+    "author": "michalrzak",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? A quick fix that updates the Dockerfile to run on `arm64` systems (such as the NVIDIA Spark). The previous version of the Dockerfile fails on `arm64` systems due to `SudachiPy`, which only provides wheels for `x86_6\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44564",
-    "created_at": "2026-03-10T10:14:21Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44603",
+    "created_at": "2026-03-11T16:42:30Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44564/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44564",
+    "files_url": "https://github.com/huggingface/transformers/pull/44603/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44603",
     "labels": [],
-    "merged": true,
-    "number": 44564,
+    "merged": false,
+    "number": 44603,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix glm dsa",
-    "updated_at": "2026-03-19T15:13:36Z"
+    "state": "open",
+    "title": "fixed dockerfile for arm64 systems",
+    "updated_at": "2026-03-11T16:42:30Z"
   },
   {
-    "additions": 165,
-    "author": "zucchini-nlp",
+    "additions": 218,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Split out `mm_token_type_id` creation to a separate utility and just call it in VLMs. Also make sure that `mm_token_type_id` can be created even when `padding=False` and the inputs are of different length. As long a\u2026",
-    "changed_files": 37,
+    "body_excerpt": "# What does this PR do? As per the title. Follow up of https://github.com/huggingface/transformers/pull/44330 Also take the opportunity to simplify t5 and its children, because the way they compute`position_bias` was super convoluted/overc\u2026",
+    "changed_files": 61,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44563",
-    "created_at": "2026-03-10T10:13:29Z",
-    "deletions": 267,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44602",
+    "created_at": "2026-03-11T16:19:43Z",
+    "deletions": 1083,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44563/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44563",
+    "files_url": "https://github.com/huggingface/transformers/pull/44602/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44602",
     "labels": [],
     "merged": true,
-    "number": 44563,
-    "review_comments_count": 4,
+    "number": 44602,
+    "review_comments_count": 14,
     "state": "closed",
-    "title": "Allow `mm_token_type` be non-padded lists ",
-    "updated_at": "2026-03-25T11:33:46Z"
+    "title": "Remove `cache_position` in more models (2)",
+    "updated_at": "2026-03-12T22:38:15Z"
   },
   {
-    "additions": 54,
-    "author": "SamArun28",
-    "author_association": "NONE",
-    "body_excerpt": "#Standardizing the BERT model card as part of issue #36979 Changes made: - Added friendly description of BERT - Added Pipeline and AutoModel code examples - Added Notes section with helpful tips - Added Resources section with links @stevhl\u2026",
-    "changed_files": 1,
+    "additions": 510,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Goal is to be able to run a model with both PP and TP.",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44562",
-    "created_at": "2026-03-10T09:58:19Z",
-    "deletions": 111,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44562/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44562",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44601",
+    "created_at": "2026-03-11T15:56:51Z",
+    "deletions": 2,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44601/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44601",
     "labels": [],
     "merged": false,
-    "number": 44562,
+    "number": 44601,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "docs: standardize BERT model card",
-    "updated_at": "2026-03-10T16:00:43Z"
+    "state": "open",
+    "title": "[Distributed] Add PP support natively",
+    "updated_at": "2026-03-12T11:53:24Z"
   },
   {
-    "additions": 368,
-    "author": "rain-1",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "> Adds support for the legacy text completions endpoint, which accepts a freeform text prompt (no chat template) and returns generated text in choices[].text. Supports both streaming and non-streaming modes, suffix for fill-in-the-middle i\u2026",
-    "changed_files": 3,
+    "additions": 0,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Small mistake in https://github.com/huggingface/transformers/pull/44432. cc @zucchini-nlp, was it intended to remove the scaling? (I assume so since the embedding now has the saling baked in, and I guess paligemma a\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44558",
-    "created_at": "2026-03-10T07:09:07Z",
-    "deletions": 0,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44600",
+    "created_at": "2026-03-11T15:40:16Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44558/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44558",
+    "files_url": "https://github.com/huggingface/transformers/pull/44600/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44600",
     "labels": [],
-    "merged": false,
-    "number": 44558,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "Add /v1/completions endpoint (OpenAI legacy completions API) to `transformers serve`",
-    "updated_at": "2026-04-15T15:16:40Z"
+    "merged": true,
+    "number": 44600,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Remove useless identity assignment",
+    "updated_at": "2026-03-12T10:21:23Z"
   },
   {
-    "additions": 1,
-    "author": "black-yt",
-    "author_association": "NONE",
-    "body_excerpt": "This PR fixes a runtime `TypeError` encountered during model initialization when using Qwen3.5 configurations with recent `transformers` versions. The error occurs in `modeling_rope_utils.py` during RoPE parameter validation: ``` TypeError\u2026",
+    "additions": 3,
+    "author": "hf-security-analysis[bot]",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Update `.github/workflows/self-comment-ci.yml` workflow configuration. cc @vasqu Closes huggingface/tracking-issues#17",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 16,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44555",
-    "created_at": "2026-03-10T01:52:18Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44599",
+    "created_at": "2026-03-11T15:39:30Z",
+    "deletions": 161,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44555/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44555",
+    "files_url": "https://github.com/huggingface/transformers/pull/44599/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44599",
     "labels": [],
     "merged": false,
-    "number": 44555,
+    "number": 44599,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix TypeError in RoPE validation when ignore_keys_at_rope_validation is a list",
-    "updated_at": "2026-03-28T10:04:35Z"
+    "title": "chore: update self-comment-ci.yml",
+    "updated_at": "2026-04-02T09:05:25Z"
   },
   {
-    "additions": 233,
+    "additions": 1,
     "author": "vasqu",
     "author_association": "MEMBER",
-    "body_excerpt": "As per title, wip and really unsure if we really want this",
-    "changed_files": 16,
+    "body_excerpt": "As per title, updated to remove former members as well cc @yonigozlan @ydshieh",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44553",
-    "created_at": "2026-03-10T01:04:49Z",
-    "deletions": 263,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44553/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44553",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44598",
+    "created_at": "2026-03-11T15:13:25Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44598/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44598",
     "labels": [],
-    "merged": false,
-    "number": 44553,
-    "review_comments_count": 1,
-    "state": "open",
-    "title": "[`FA`] Refactor FA CB kwargs",
-    "updated_at": "2026-03-17T09:14:21Z"
+    "merged": true,
+    "number": 44598,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Add Yoni to run-slow workflow",
+    "updated_at": "2026-03-11T15:38:10Z"
   },
   {
-    "additions": 0,
+    "additions": 1,
     "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "Reverts huggingface/transformers#44529",
+    "body_excerpt": "# What does this PR do? Our beautiful Dashboard is missing ..... damm",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44552",
-    "created_at": "2026-03-09T21:05:46Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44597",
+    "created_at": "2026-03-11T13:53:02Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44552/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44552",
+    "files_url": "https://github.com/huggingface/transformers/pull/44597/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44597",
     "labels": [],
     "merged": true,
-    "number": 44552,
+    "number": 44597,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Revert \"test merge queue 1\"",
-    "updated_at": "2026-03-09T21:15:55Z"
+    "title": "Fix CircleCI summary report not showing due to missing dependency",
+    "updated_at": "2026-03-20T07:33:38Z"
   },
   {
-    "additions": 12,
-    "author": "echarlaix",
-    "author_association": "MEMBER",
-    "body_excerpt": "`self.rotary_emb` is always called since https://github.com/huggingface/transformers/pull/39847 while only being initialized when `config.use_mem_rope` is True inference failing since v5 for models `config.use_mem_rope=False` ``` Attribute\u2026",
+    "additions": 26,
+    "author": "Desel72",
+    "author_association": "NONE",
+    "body_excerpt": "Fix TypeError when loading float8 models by falling back to bfloat16 in local_torch_dtype # What does this PR do? When loading FP8 models (e.g. `Qwen/Qwen3.5-35B-A3B-FP8`) with `dtype=\"auto\"`, the auto-detected dtype from checkpoint weight\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44551",
-    "created_at": "2026-03-09T18:15:17Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44596",
+    "created_at": "2026-03-11T13:03:19Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44551/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44551",
-    "labels": [],
-    "merged": true,
-    "number": 44551,
+    "files_url": "https://github.com/huggingface/transformers/pull/44596/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44596",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44596,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix zamba2 rotary embedding call when use_mem_rope is False",
-    "updated_at": "2026-03-10T11:43:36Z"
+    "title": "Fix TypeError when loading float8 models by falling back to bfloat16 in local_torch_dtype",
+    "updated_at": "2026-03-11T14:00:39Z"
   },
   {
-    "additions": 6,
-    "author": "himani27301",
+    "additions": 2324,
+    "author": "yonigozlan",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Add [CHMv2 ](https://arxiv.org/abs/2603.06382) to Transformers",
+    "changed_files": 23,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44595",
+    "created_at": "2026-03-11T12:38:44Z",
+    "deletions": 23,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44595/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44595",
+    "labels": [
+      "New model",
+      "run-slow"
+    ],
+    "merged": true,
+    "number": 44595,
+    "review_comments_count": 30,
+    "state": "closed",
+    "title": "Add CHMv2",
+    "updated_at": "2026-03-11T16:00:03Z"
+  },
+  {
+    "additions": 271,
+    "author": "vimal-crypto",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Improved wording and grammar in the Auto Classes documentation to enhance readability without changing functionality. # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, you\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## What this PR does This PR brings `ObjectDetectionPipeline` in line with its sister pipelines (`ZeroShotObjectDetectionPipeline`, `ImageClassificationPipeline`) by adding four enhancements to the postprocessing stage. ### Changes **1. Sc\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44550",
-    "created_at": "2026-03-09T16:12:59Z",
-    "deletions": 16,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44594",
+    "created_at": "2026-03-11T12:37:46Z",
+    "deletions": 40,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44550/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44550",
+    "files_url": "https://github.com/huggingface/transformers/pull/44594/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44594",
     "labels": [],
     "merged": false,
-    "number": 44550,
-    "review_comments_count": 1,
+    "number": 44594,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Improve clarity and grammar in Auto Classes documentation",
-    "updated_at": "2026-03-09T16:32:29Z"
+    "title": "[Pipeline] Add top_k, label filtering, box_format and score sorting to ObjectDetectionPipeline",
+    "updated_at": "2026-03-11T12:37:46Z"
   },
   {
-    "additions": 158,
-    "author": "ydshieh",
+    "additions": 15,
+    "author": "BenjaminBossan",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The previous values are simply wrong. (which also cause problems for tiny model creation)",
-    "changed_files": 51,
+    "body_excerpt": "# What does this PR do? Multiple PEFT tests are failing due to recent changes in transformers. - hf_device_map attribute may not exist in some cases - respect inference_mode in load_adapter - new model loading requires changes for bnb (SCB\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44549",
-    "created_at": "2026-03-09T15:44:36Z",
-    "deletions": 126,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44592",
+    "created_at": "2026-03-11T10:41:51Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44549/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44549",
+    "files_url": "https://github.com/huggingface/transformers/pull/44592/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44592",
     "labels": [],
     "merged": true,
-    "number": 44549,
+    "number": 44592,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix wrong (non-existing) checkpoints",
-    "updated_at": "2026-03-11T15:16:56Z"
+    "title": "FIX Multiple PEFT errors after v5 transition",
+    "updated_at": "2026-03-11T12:24:05Z"
   },
   {
-    "additions": 12,
-    "author": "mvanhorn",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes #43935 ## Summary - Added `eval_on_end` argument to `TrainingArguments` (default: `False`) - Added conditional evaluation at the end of training in `Trainer.train()`, symmetric to `eval_on_start` - Implementation mirrors the existing\u2026",
-    "changed_files": 2,
+    "additions": 60,
+    "author": "moktamd",
+    "author_association": "NONE",
+    "body_excerpt": "Adds `_apply_mps_fixes` in `sdpa_attention.py` to handle two upstream PyTorch MPS bugs: 1. **pytorch/pytorch#176767** (fixed in PyTorch 2.12): pads value tensor when `v_head_dim != q_head_dim` to avoid corrupted output. Affects DeepSeek mo\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44548",
-    "created_at": "2026-03-09T15:03:27Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44591",
+    "created_at": "2026-03-11T10:32:26Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44548/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44548",
+    "files_url": "https://github.com/huggingface/transformers/pull/44591/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44591",
     "labels": [],
     "merged": false,
-    "number": 44548,
+    "number": 44591,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add eval_on_end flag to Trainer",
-    "updated_at": "2026-03-09T19:31:59Z"
+    "title": "Add MPS SDPA workarounds for value head dim and bidirectional attention",
+    "updated_at": "2026-03-11T13:37:15Z"
   },
   {
     "additions": 2,
-    "author": "mvanhorn",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes #44373 ## Summary - Corrected the docstring for `position_ids` parameter in `prepare_fa_kwargs_from_position_ids` and `_prepare_from_posids` which incorrectly described attention mask semantics (\"Boolean or int tensor... 1 means vali\u2026",
+    "author": "pranay-3108",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes incorrect documentation for `position_ids` in `masking_utils.py`. The docstring previously described `position_ids` as `torch.Tensor`. This PR updates it to `torch.LongTensor` and aligns the description with the standard wording used\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44547",
-    "created_at": "2026-03-09T14:59:27Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44590",
+    "created_at": "2026-03-11T05:13:57Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44547/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44547",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44590/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44590",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44547,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "Fix position_ids docstring in modeling_flash_attention_utils.py",
-    "updated_at": "2026-03-09T16:08:29Z"
+    "number": 44590,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix incorrect docstring for position_ids",
+    "updated_at": "2026-03-11T21:08:42Z"
   },
   {
-    "additions": 4,
-    "author": "Abdennacer-Badaoui",
-    "author_association": "MEMBER",
-    "body_excerpt": "The AMD Docker image build (`latest-pytorch-amd`) has been failing since early February due to the 6h GitHub Actions job time limit being exceeded. The root cause is that Flash Attention is cloned and compiled from the latest commit on the\u2026",
+    "additions": 1,
+    "author": "jiqing-feng",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fixes Llama4 model loading under BitsAndBytes (BNB) quantization mode. Router quantized incorrectly causes shape mismatch: Llama4Router inherits from nn.Linear, so BNB quantizes its weight into a packed format. However, super().forward() c\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44546",
-    "created_at": "2026-03-09T14:15:51Z",
-    "deletions": 2,
+    "comments_count": 15,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44588",
+    "created_at": "2026-03-11T01:42:33Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44546/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44546",
+    "files_url": "https://github.com/huggingface/transformers/pull/44588/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44588",
     "labels": [],
     "merged": true,
-    "number": 44546,
-    "review_comments_count": 0,
+    "number": 44588,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Fix AMD Docker image build timeout by pinning Flash Attention commit",
-    "updated_at": "2026-03-09T14:37:50Z"
+    "title": "Fix llama4 bnb mode",
+    "updated_at": "2026-03-27T14:19:14Z"
   },
   {
-    "additions": 18,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. Fixes https://github.com/huggingface/transformers/issues/44336",
-    "changed_files": 1,
+    "additions": 32,
+    "author": "kmbhattt-aws",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44544",
-    "created_at": "2026-03-09T11:38:09Z",
-    "deletions": 18,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44544/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44544",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44587",
+    "created_at": "2026-03-11T01:01:18Z",
+    "deletions": 20,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44587/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44587",
     "labels": [],
-    "merged": true,
-    "number": 44544,
+    "merged": false,
+    "number": 44587,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix ansi codes in loading reports when not connected to terminal",
-    "updated_at": "2026-03-09T11:52:16Z"
+    "state": "open",
+    "title": "Fix: Handling fused qkv result tensor slicing for tp sharded qkv weights",
+    "updated_at": "2026-03-12T21:31:29Z"
   },
   {
-    "additions": 154,
-    "author": "umbilnm",
+    "additions": 91,
+    "author": "mvanhorn",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #44521 `apply_chat_template` with `return_assistant_tokens_mask=True` returns all-zero masks when multimodal inputs (images/videos) are present. ## Root cause `generation_indices` (character-level positions of\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Decouples router logits collection from output visibility in Mixtral's `ForCausalLM`. Previously, `output_router_logits=False` (the default) prevented `aux_loss` from being computed, meaning load balancing was silen\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44543",
-    "created_at": "2026-03-09T10:47:05Z",
-    "deletions": 21,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44586",
+    "created_at": "2026-03-11T00:24:07Z",
+    "deletions": 39,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44543/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44543",
+    "files_url": "https://github.com/huggingface/transformers/pull/44586/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44586",
     "labels": [],
     "merged": false,
-    "number": 44543,
+    "number": 44586,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix assistant_masks for multimodal inputs in apply_chat_template",
-    "updated_at": "2026-03-09T10:47:05Z"
+    "state": "closed",
+    "title": "Fix Mixtral aux_loss not computed when output_router_logits=False",
+    "updated_at": "2026-03-11T14:31:21Z"
   },
   {
-    "additions": 3,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title",
-    "changed_files": 2,
+    "additions": 10,
+    "author": "mvanhorn",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Passes `eps=config.rms_norm_eps` to both `q_a_layernorm` and `kv_a_layernorm` in the DeepseekV3 MLA attention module. Without this, these layernorms default to `eps=1e-5` instead of the config value (`1e-6`), causin\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44542",
-    "created_at": "2026-03-09T10:00:29Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44585",
+    "created_at": "2026-03-11T00:20:54Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44542/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44542",
+    "files_url": "https://github.com/huggingface/transformers/pull/44585/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44585",
     "labels": [],
-    "merged": true,
-    "number": 44542,
+    "merged": false,
+    "number": 44585,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix backend dependency",
-    "updated_at": "2026-03-09T10:10:24Z"
+    "state": "open",
+    "title": "Fix missing rms_norm_eps in DeepseekV3 MLA layernorms",
+    "updated_at": "2026-03-12T14:39:12Z"
   },
   {
     "additions": 1,
-    "author": "YangKai0616",
+    "author": "mvanhorn",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Per the title, an error occurs when `tp_plan` is empty due to [here](https://github.com/huggingface/transformers/blob/701628527ae1ef37473f05f5d94fac7f457a3f8f/src/transformers/core_model_loading.py#L1120): ``` [rank0]: Traceback (most rece\u2026",
+    "body_excerpt": "# What does this PR do? Fixes an off-by-one error in `decode_spans()` in the document question answering pipeline that causes a `ValueError: kth(=N) out of bounds` crash when `len(scores_flat) == topk`. The boundary check on line 97 uses `\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44540",
-    "created_at": "2026-03-09T09:37:36Z",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44584",
+    "created_at": "2026-03-10T23:52:51Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44540/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44540",
+    "files_url": "https://github.com/huggingface/transformers/pull/44584/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44584",
     "labels": [],
     "merged": true,
-    "number": 44540,
+    "number": 44584,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix UnboundLocalError for tp_plan_alt when tp_plan is empty",
-    "updated_at": "2026-03-11T13:42:39Z"
+    "title": "Fix off-by-one in decode_spans boundary check",
+    "updated_at": "2026-03-12T13:22:10Z"
   },
   {
-    "additions": 1333,
-    "author": "kmswin1",
-    "author_association": "NONE",
-    "body_excerpt": "A.X K1 \ubaa8\ub378 \ucd94\uac00\ud569\ub2c8\ub2e4. \uae30\uc874 inference \uc640 \ub3d9\uc77c\ud55c \uac83 \ud655\uc778\ud588\uc2b5\ub2c8\ub2e4.",
-    "changed_files": 8,
+    "additions": 45,
+    "author": "wilnn",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "\u2026kpoint when `save_strategy` is `best` # What does this PR do? fix load_best_model_checkpoint_at_end do not load the best model checkpoint at the end when `save_strategy` is `\"best\"` Fixes # (issue) fix load_best_model_checkpoint_at_end do\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44539",
-    "created_at": "2026-03-09T09:03:10Z",
-    "deletions": 0,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44583",
+    "created_at": "2026-03-10T22:37:36Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44539/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44539",
+    "files_url": "https://github.com/huggingface/transformers/pull/44583/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44583",
     "labels": [],
-    "merged": false,
-    "number": 44539,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44583,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Add A.X K1",
-    "updated_at": "2026-03-09T09:04:11Z"
+    "title": "fix load_best_model_checkpoint_at_end do not load the best model chec\u2026",
+    "updated_at": "2026-03-24T15:42:26Z"
   },
   {
-    "additions": 14,
-    "author": "ydshieh",
+    "additions": 3,
+    "author": "yonigozlan",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Follow-up of #44532: we need to change the required status check to the new added job `doc_build_status_check` added in this PR, otherwise the merge queue won't get the required (passing) status and will eventually\u2026",
+    "body_excerpt": "# What does this PR do? Fix an issue introduced in #42564 . The refactor embedded raw image tokens instead of BPE tokens, causing the model to output gibberish. This fix adds back the image tokens to BPE tokens conversion before embedding.\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44538",
-    "created_at": "2026-03-09T08:59:12Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44582",
+    "created_at": "2026-03-10T21:00:30Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44538/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44538",
+    "files_url": "https://github.com/huggingface/transformers/pull/44582/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44582",
     "labels": [],
     "merged": true,
-    "number": 44538,
+    "number": 44582,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add a new job in `build_pr_documentation.yml` (will be the new required job)",
-    "updated_at": "2026-03-09T09:12:25Z"
+    "title": "Fix missing BPE token conversion step in Chameleon",
+    "updated_at": "2026-03-11T11:26:49Z"
   },
   {
-    "additions": 2,
-    "author": "YangKai0616",
+    "additions": 9,
+    "author": "harshaljanjani",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Per the title. Supplement skip logic for XPU in the CPU-only tp tests. Hi @IlyasMoutawwakil, please help review, thanks!",
-    "changed_files": 1,
+    "body_excerpt": "### What does this PR do? The following issues were identified and fixed in this PR: \u2192 **Llama-4 Vision:** [freqs_ci is stored as a plain attr](https://github.com/huggingface/transformers/blob/153894c013/src/transformers/models/llama4/mode\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44536",
-    "created_at": "2026-03-09T08:09:41Z",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44581",
+    "created_at": "2026-03-10T19:33:51Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44536/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44536",
+    "files_url": "https://github.com/huggingface/transformers/pull/44581/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44581",
     "labels": [],
     "merged": true,
-    "number": 44536,
-    "review_comments_count": 0,
+    "number": 44581,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Supplement skip logic for XPU in the CPU-only tp tests",
-    "updated_at": "2026-03-09T10:10:49Z"
-  },
-  {
-    "additions": 53,
-    "author": "Anakintano",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Problem `Qwen2_5_VLProcessor.apply_chat_template` raises `ValueError: setting an array element with a sequence` when called with a batch of \u22652 conversations that include images under the default `padding=False` setting. **Root cause:**\u2026",
-    "changed_files": 3,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44535",
-    "created_at": "2026-03-09T07:54:58Z",
-    "deletions": 10,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44535/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44535",
-    "labels": [],
-    "merged": false,
-    "number": 44535,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix crash in Qwen2_5_VLProcessor when using batched input with padding=False",
-    "updated_at": "2026-03-09T12:44:00Z"
+    "title": "fix(models, testing): Fix Llama4 vision rotary meta tensor initialization and MyT5 get_tokenizer signature",
+    "updated_at": "2026-03-13T16:13:09Z"
   },
   {
-    "additions": 0,
-    "author": "stargazerwh",
+    "additions": 16,
+    "author": "rabbierabbie",
     "author_association": "NONE",
-    "body_excerpt": "## Description The 'transformers run' command was removed in Transformers v5, but the documentation still contained references in 62 model documentation files. This PR removes all `<hfoption id='transformers CLI'>` and `<hfoption id='trans\u2026",
-    "changed_files": 62,
+    "body_excerpt": "This PR clarifies references to the **Transformers library** in the README. While reading the documentation as a new user, I initially found the term **\"Transformers\"** ambiguous because it could refer either to the **Transformer architect\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44533",
-    "created_at": "2026-03-09T01:39:32Z",
-    "deletions": 439,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44580",
+    "created_at": "2026-03-10T18:21:55Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44533/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44533",
+    "files_url": "https://github.com/huggingface/transformers/pull/44580/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44580",
     "labels": [],
     "merged": false,
-    "number": 44533,
+    "number": 44580,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "docs: Remove references to removed 'transformers run' CLI command",
-    "updated_at": "2026-03-09T02:33:08Z"
+    "title": "Clarify references to the Transformers library in README",
+    "updated_at": "2026-03-11T13:24:58Z"
   },
   {
-    "additions": 9,
+    "additions": 1,
     "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? So we can use `Require Merge Queue` functionoality",
+    "body_excerpt": "# What does this PR do? We need to fetch the specific commit (the so called merge commit created by Github itself)",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44532",
-    "created_at": "2026-03-08T20:34:05Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44579",
+    "created_at": "2026-03-10T16:58:09Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44532/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44532",
+    "files_url": "https://github.com/huggingface/transformers/pull/44579/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44579",
     "labels": [],
     "merged": true,
-    "number": 44532,
+    "number": 44579,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Update `build_pr_documentation` workflow for `merge_group` event",
-    "updated_at": "2026-03-08T20:42:57Z"
+    "title": "Fix PR comment CI for quantization job",
+    "updated_at": "2026-03-10T17:07:11Z"
   },
   {
-    "additions": 16,
-    "author": "s-zx",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes `Qwen2_5_VLProcessor.apply_chat_template` crashing with a `ValueError` when called with a batch of conversations with different prompt lengths and `padding=False` (the default). ### Root cause In the `mm_toke\u2026",
+    "additions": 449,
+    "author": "remi-or",
+    "author_association": "MEMBER",
+    "body_excerpt": "# Summary This PR adds three attributes to the compile config, to have granularity over how varlen (handles mixed prefil and decode batches) and decode (only decode batches) are compiled. We want to have this kind of granularity because va\u2026",
+    "changed_files": 11,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44578",
+    "created_at": "2026-03-10T16:31:20Z",
+    "deletions": 121,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44578/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44578",
+    "labels": [],
+    "merged": true,
+    "number": 44578,
+    "review_comments_count": 4,
+    "state": "closed",
+    "title": "[CB] Better parametrization for compile",
+    "updated_at": "2026-03-19T11:50:08Z"
+  },
+  {
+    "additions": 5,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "- Fix quantizer_aqlm.py to use renamed modules_to_not_convert parameter instead of removed linear_weights_not_to_quantize - Update test to match new function signature: no tuple return, module names instead of weight names",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44531",
-    "created_at": "2026-03-08T19:38:00Z",
-    "deletions": 10,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44577",
+    "created_at": "2026-03-10T15:57:36Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44531/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44531",
+    "files_url": "https://github.com/huggingface/transformers/pull/44577/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44577",
     "labels": [],
-    "merged": false,
-    "number": 44531,
+    "merged": true,
+    "number": 44577,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Qwen2_5_VLProcessor.apply_chat_template crash on unpadded batched input",
-    "updated_at": "2026-03-09T13:14:02Z"
+    "title": "Fix: AQLM quantizer to match updated replace_with_aqlm_linear signature",
+    "updated_at": "2026-03-10T17:48:00Z"
   },
   {
-    "additions": 3,
-    "author": "ydshieh2",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Merge queue is enabled, but we need to see if it works well with CircleCI",
+    "additions": 16,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/44387. This PR disable async loading when we want to quantize the model. it is actually faster than doing a semaphore. If a quantizer happens to quantize fast\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44529",
-    "created_at": "2026-03-08T18:12:54Z",
-    "deletions": 0,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44576",
+    "created_at": "2026-03-10T15:07:01Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44529/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44529",
+    "files_url": "https://github.com/huggingface/transformers/pull/44576/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44576",
     "labels": [],
     "merged": true,
-    "number": 44529,
-    "review_comments_count": 0,
+    "number": 44576,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "test merge queue 1",
-    "updated_at": "2026-03-09T21:01:09Z"
+    "title": "Disable async loading when quantizing on the fly",
+    "updated_at": "2026-03-16T16:36:42Z"
   },
   {
-    "additions": 3,
-    "author": "ydshieh",
+    "additions": 13,
+    "author": "Rocketknight1",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Merge queue is enabled, but we need to see if it works well with CircleCI",
+    "body_excerpt": "Some parameters in Tapas are initialized in `__init__()` and not reinitialized in `_init_weights()`, which means that if the model is created on the `meta` device, those parameters do not get a weight initialization. This causes a crash la\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44528",
-    "created_at": "2026-03-08T17:54:32Z",
-    "deletions": 0,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44575",
+    "created_at": "2026-03-10T14:42:40Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44528/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44528",
+    "files_url": "https://github.com/huggingface/transformers/pull/44575/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44575",
     "labels": [],
-    "merged": false,
-    "number": 44528,
+    "merged": true,
+    "number": 44575,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "test merge queue 1",
-    "updated_at": "2026-03-09T20:39:15Z"
+    "title": "Correct Tapas initialization",
+    "updated_at": "2026-03-10T15:14:37Z"
   },
   {
-    "additions": 3,
-    "author": "Sai-Suraj-27",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes these failing [MusicgenStereoIntegrationTests](https://github.com/huggingface/transformers/actions/runs/22606636929/job/65500451700#step:14:7870) ## Before submitting - [ ] This PR fixes a typo or improves the\u2026",
-    "changed_files": 1,
+    "additions": 33,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "Fixes #44448",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44527",
-    "created_at": "2026-03-08T14:26:02Z",
-    "deletions": 1,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44571",
+    "created_at": "2026-03-10T14:28:22Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44527/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44527",
+    "files_url": "https://github.com/huggingface/transformers/pull/44571/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44571",
     "labels": [],
     "merged": true,
-    "number": 44527,
+    "number": 44571,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix failing `MusicgenStereo` integration tests",
-    "updated_at": "2026-03-10T12:28:39Z"
+    "title": "Fix pegasus conversion",
+    "updated_at": "2026-03-18T09:55:00Z"
   },
   {
-    "additions": 90,
-    "author": "JoursBleu",
+    "additions": 30,
+    "author": "umbilnm",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Add GGUF loading support for MiniMax-M2.1 (456B MoE) model. MiniMax-M2.1 is a large Mixture-of-Experts model with 456B total parameters (45.9B active), 256 experts and 8 experts per token. This PR enables loading it\u2026",
+    "body_excerpt": "# What does this PR do? In transformers v5, `DebertaV2Tokenizer` was rewritten to use `TokenizersBackend`, but the `post_processor` responsible for adding `[CLS]`/`[SEP]` tokens was never set. This causes `add_special_tokens=True` to silen\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44526",
-    "created_at": "2026-03-08T09:57:38Z",
-    "deletions": 0,
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44570",
+    "created_at": "2026-03-10T13:37:06Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44526/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44526",
+    "files_url": "https://github.com/huggingface/transformers/pull/44570/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44570",
     "labels": [],
     "merged": true,
-    "number": 44526,
+    "number": 44570,
     "review_comments_count": 1,
     "state": "closed",
-    "title": "Add GGUF support for MiniMax-M2.1 model",
-    "updated_at": "2026-03-18T14:39:20Z"
+    "title": "Fix missing post_processor in DebertaV2Tokenizer causing no special t\u2026",
+    "updated_at": "2026-03-24T09:40:44Z"
   },
   {
-    "additions": 1,
-    "author": "jnMetaCode",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes a `KeyError` crash in `_parse_type_hint` in `chat_template_utils.py` (line 117). When processing Union types, the code accesses `subtype[\"type\"]` without checking the key exists. `_get_json_schema_type(Any)` returns `{}` (\u2026",
+    "additions": 267,
+    "author": "aashay-sarvam",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Adds native support for the `sarvam_mla` model type (`sarvamai/sarvam-105b`) to HuggingFace Transformers using the **modular pattern**, inheriting from DeepSeek V3. ### Model Architecture SarvamMLA is a **105B para\u2026",
+    "changed_files": 10,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44569",
+    "created_at": "2026-03-10T11:55:01Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44569/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44569",
+    "labels": [],
+    "merged": false,
+    "number": 44569,
+    "review_comments_count": 20,
+    "state": "open",
+    "title": "Add SarvamMLA model (sarvamai/sarvam-105b)",
+    "updated_at": "2026-03-18T13:58:18Z"
+  },
+  {
+    "additions": 2,
+    "author": "tomaarsen",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Currently, when using Apertus (or rather, it's `XIELUActivation`), and you don't have `xielu` installed, then you'll fall to this `except:` https://github.com/huggingface/transformers/blob/5a098a1e01034095f037c8a37f\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44525",
-    "created_at": "2026-03-08T09:21:27Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44567",
+    "created_at": "2026-03-10T11:39:20Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44525/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44525",
+    "files_url": "https://github.com/huggingface/transformers/pull/44567/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44567",
     "labels": [],
     "merged": true,
-    "number": 44525,
+    "number": 44567,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix KeyError in _parse_type_hint when Union contains Any",
-    "updated_at": "2026-03-09T13:43:23Z"
+    "title": "[`fix`] Prevent crash with Apertus without xielu installed",
+    "updated_at": "2026-03-10T13:24:11Z"
   },
   {
-    "additions": 1,
-    "author": "jnMetaCode",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes a bug in `AssistantTracker.is_active()` in `chat_template_utils.py`. After activation via `activate_tracker()`, `_rendered_blocks` and `_generation_indices` are set to list arguments which may be empty `[]`. The `is_active\u2026",
-    "changed_files": 1,
+    "additions": 181,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "This patch extends `ty` check to `src/transformers/cli` Based on https://github.com/huggingface/transformers/pull/44412",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44524",
-    "created_at": "2026-03-08T09:21:25Z",
-    "deletions": 1,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44566",
+    "created_at": "2026-03-10T10:40:13Z",
+    "deletions": 86,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44524/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44524",
+    "files_url": "https://github.com/huggingface/transformers/pull/44566/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44566",
     "labels": [],
-    "merged": true,
-    "number": 44524,
-    "review_comments_count": 0,
+    "merged": false,
+    "number": 44566,
+    "review_comments_count": 21,
     "state": "closed",
-    "title": "Fix AssistantTracker.is_active() returning False after activation with empty lists",
-    "updated_at": "2026-03-09T13:36:19Z"
+    "title": "chore(typing): extend typing to `src/transformers/cli` ",
+    "updated_at": "2026-04-01T16:05:57Z"
   },
   {
-    "additions": 2,
-    "author": "jnMetaCode",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes two small bugs in `load_sharded_checkpoint` in `trainer_utils.py`: **Bug 1 \u2014 Copy-paste error in error message (line 1108):** When reporting unexpected keys, the error message incorrectly says \"Missing key(s)\" instead of \"\u2026",
+    "additions": 36,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title. The unused memory is taken into account too late, which leads to different device_map for the same hardware and models, and even random cuda OOM!! Basically, the max memory needs to be adjusted BEF\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44523",
-    "created_at": "2026-03-08T09:21:22Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44565",
+    "created_at": "2026-03-10T10:31:10Z",
+    "deletions": 22,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44523/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44523",
+    "files_url": "https://github.com/huggingface/transformers/pull/44565/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44565",
     "labels": [],
     "merged": true,
-    "number": 44523,
-    "review_comments_count": 0,
+    "number": 44565,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix error message label and docstring default in load_sharded_checkpoint",
-    "updated_at": "2026-03-10T15:48:41Z"
+    "title": "[device_map] Fix device_map computation by correctly adjusting memory available",
+    "updated_at": "2026-03-10T17:16:01Z"
   },
   {
-    "additions": 41,
-    "author": "nakigami",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? This PR introduces initial unit test coverage for the `transformers-cli` tool, specifically focusing on diagnostic and model utility commands. Currently, these CLI entry points lack automated tests. These new tests\u2026",
-    "changed_files": 1,
+    "additions": 2,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes #44360",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44520",
-    "created_at": "2026-03-08T01:30:39Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44564",
+    "created_at": "2026-03-10T10:14:21Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44520/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44520",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44520,
+    "files_url": "https://github.com/huggingface/transformers/pull/44564/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44564",
+    "labels": [],
+    "merged": true,
+    "number": 44564,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "test(cli): add unit tests for env and model utility commands",
-    "updated_at": "2026-03-09T13:19:15Z"
+    "title": "Fix glm dsa",
+    "updated_at": "2026-03-19T15:13:36Z"
   },
   {
-    "additions": 3,
-    "author": "Sai-Suraj-27",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes these failing [MarianIntegrationTests](https://github.com/huggingface/transformers/actions/runs/22606636929/job/65500458014#step:14:6186) <img width=\"2378\" height=\"657\" alt=\"image\" src=\"https://github.com/user\u2026",
-    "changed_files": 1,
+    "additions": 165,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Split out `mm_token_type_id` creation to a separate utility and just call it in VLMs. Also make sure that `mm_token_type_id` can be created even when `padding=False` and the inputs are of different length. As long a\u2026",
+    "changed_files": 37,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44519",
-    "created_at": "2026-03-07T19:53:23Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44563",
+    "created_at": "2026-03-10T10:13:29Z",
+    "deletions": 267,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44519/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44519",
+    "files_url": "https://github.com/huggingface/transformers/pull/44563/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44563",
     "labels": [],
     "merged": true,
-    "number": 44519,
-    "review_comments_count": 0,
+    "number": 44563,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Fix failing `MarianIntegrationTests`",
-    "updated_at": "2026-03-09T14:11:12Z"
+    "title": "Allow `mm_token_type` be non-padded lists ",
+    "updated_at": "2026-03-25T11:33:46Z"
   },
   {
-    "additions": 12,
-    "author": "KartikPawade",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #44514 `Qwen2_5_VLProcessor.__call__` crashed with a `ValueError` when processing a batch of conversations with different lengths and `padding=False` (the default). **Root cause:** The `mm_token_type_ids` blo\u2026",
-    "changed_files": 2,
+    "additions": 54,
+    "author": "SamArun28",
+    "author_association": "NONE",
+    "body_excerpt": "#Standardizing the BERT model card as part of issue #36979 Changes made: - Added friendly description of BERT - Added Pipeline and AutoModel code examples - Added Notes section with helpful tips - Added Resources section with links @stevhl\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44518",
-    "created_at": "2026-03-07T19:22:40Z",
-    "deletions": 10,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44562",
+    "created_at": "2026-03-10T09:58:19Z",
+    "deletions": 111,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44518/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44518",
+    "files_url": "https://github.com/huggingface/transformers/pull/44562/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44562",
     "labels": [],
     "merged": false,
-    "number": 44518,
+    "number": 44562,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: Qwen2_5_VLProcessor crashes on batched input when padding=False \u2026",
-    "updated_at": "2026-03-10T18:57:10Z"
+    "title": "docs: standardize BERT model card",
+    "updated_at": "2026-03-10T16:00:43Z"
   },
   {
-    "additions": 12637,
-    "author": "ShahVandit",
+    "additions": 368,
+    "author": "rain-1",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Adds Qwen3-TTS, a series of text-to-speech models by the Qwen team (Alibaba Group), to Transformers. **Architecture:** - `Qwen3TTSForConditionalGeneration` \u2014 text to multi-codebook speech codes (talker) - `Qwen3TTS\u2026",
-    "changed_files": 16,
+    "body_excerpt": "> Adds support for the legacy text completions endpoint, which accepts a freeform text prompt (no chat template) and returns generated text in choices[].text. Supports both streaming and non-streaming modes, suffix for fill-in-the-middle i\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44517",
-    "created_at": "2026-03-07T18:48:04Z",
-    "deletions": 24,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44558",
+    "created_at": "2026-03-10T07:09:07Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44517/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44517",
-    "labels": [
-      "New model",
-      "Audio"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44558/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44558",
+    "labels": [],
     "merged": false,
-    "number": 44517,
-    "review_comments_count": 19,
+    "number": 44558,
+    "review_comments_count": 3,
     "state": "open",
-    "title": "Add qwen3 tts",
-    "updated_at": "2026-04-13T07:53:06Z"
+    "title": "Add /v1/completions endpoint (OpenAI legacy completions API) to `transformers serve`",
+    "updated_at": "2026-04-15T15:16:40Z"
   },
   {
-    "additions": 65,
-    "author": "JasonCZMeng",
+    "additions": 1,
+    "author": "black-yt",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Fix `Qwen2_5_VLProcessor.apply_chat_template` crashing with `ValueError` when called with batched inputs of different sequence lengths (ragged lists) and `padding=False` (the default). Fixes #44514 ## Root Cause The `mm_token_ty\u2026",
-    "changed_files": 3,
+    "body_excerpt": "This PR fixes a runtime `TypeError` encountered during model initialization when using Qwen3.5 configurations with recent `transformers` versions. The error occurs in `modeling_rope_utils.py` during RoPE parameter validation: ``` TypeError\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44516",
-    "created_at": "2026-03-07T18:33:40Z",
-    "deletions": 10,
+    "comments_count": 16,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44555",
+    "created_at": "2026-03-10T01:52:18Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44516/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44516",
+    "files_url": "https://github.com/huggingface/transformers/pull/44555/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44555",
     "labels": [],
     "merged": false,
-    "number": 44516,
+    "number": 44555,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(qwen2_5_vl): handle ragged batched input in apply_chat_template",
-    "updated_at": "2026-03-09T13:14:22Z"
+    "title": "Fix TypeError in RoPE validation when ignore_keys_at_rope_validation is a list",
+    "updated_at": "2026-03-28T10:04:35Z"
   },
   {
-    "additions": 1,
-    "author": "Sai-Suraj-27",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes this failing [GPTNeoModelLanguageGenerationTest](https://github.com/huggingface/transformers/actions/runs/22606636929/job/65500455894#step:14:1483) <img width=\"2363\" height=\"239\" alt=\"image\" src=\"https://githu\u2026",
+    "additions": 233,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "As per title, wip and really unsure if we really want this",
+    "changed_files": 16,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44553",
+    "created_at": "2026-03-10T01:04:49Z",
+    "deletions": 263,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44553/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44553",
+    "labels": [],
+    "merged": false,
+    "number": 44553,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "[`FA`] Refactor FA CB kwargs",
+    "updated_at": "2026-03-17T09:14:21Z"
+  },
+  {
+    "additions": 0,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "Reverts huggingface/transformers#44529",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44515",
-    "created_at": "2026-03-07T18:16:35Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44552",
+    "created_at": "2026-03-09T21:05:46Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44515/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44515",
+    "files_url": "https://github.com/huggingface/transformers/pull/44552/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44552",
     "labels": [],
     "merged": true,
-    "number": 44515,
+    "number": 44552,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix failing `GPTNeoModelLanguageGenerationTest`",
-    "updated_at": "2026-03-09T14:11:21Z"
+    "title": "Revert \"test merge queue 1\"",
+    "updated_at": "2026-03-09T21:15:55Z"
   },
   {
-    "additions": 1,
-    "author": "math-hiyoko",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 69,
+    "additions": 12,
+    "author": "echarlaix",
+    "author_association": "MEMBER",
+    "body_excerpt": "`self.rotary_emb` is always called since https://github.com/huggingface/transformers/pull/39847 while only being initialized when `config.use_mem_rope` is True inference failing since v5 for models `config.use_mem_rope=False` ``` Attribute\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44513",
-    "created_at": "2026-03-07T16:11:55Z",
-    "deletions": 492,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44551",
+    "created_at": "2026-03-09T18:15:17Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44513/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44513",
+    "files_url": "https://github.com/huggingface/transformers/pull/44551/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44551",
     "labels": [],
     "merged": true,
-    "number": 44513,
+    "number": 44551,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix: Remove references to transformers run command",
-    "updated_at": "2026-03-09T15:37:16Z"
+    "title": "Fix zamba2 rotary embedding call when use_mem_rope is False",
+    "updated_at": "2026-03-10T11:43:36Z"
   },
   {
-    "additions": 4,
-    "author": "04cb",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #44355. The inspect.getsource() call raises TypeError when running compiled Python files with Cython-compiled functions. Added try-except block to gracefully handle this case by returning a default indentation level of 4.",
+    "additions": 6,
+    "author": "himani27301",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Improved wording and grammar in the Auto Classes documentation to enhance readability without changing functionality. # What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, you\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44511",
-    "created_at": "2026-03-07T05:36:25Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44550",
+    "created_at": "2026-03-09T16:12:59Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44511/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44511",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44550/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44550",
+    "labels": [],
     "merged": false,
-    "number": 44511,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix get_docstring_indentation_level to handle compiled functions",
-    "updated_at": "2026-03-09T13:10:06Z"
+    "number": 44550,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "Improve clarity and grammar in Auto Classes documentation",
+    "updated_at": "2026-03-09T16:32:29Z"
   },
   {
-    "additions": 3,
-    "author": "math-hiyoko",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 27,
+    "additions": 158,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? The previous values are simply wrong. (which also cause problems for tiny model creation)",
+    "changed_files": 51,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44510",
-    "created_at": "2026-03-06T23:37:51Z",
-    "deletions": 358,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44549",
+    "created_at": "2026-03-09T15:44:36Z",
+    "deletions": 126,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44510/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44510",
+    "files_url": "https://github.com/huggingface/transformers/pull/44549/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44549",
     "labels": [],
     "merged": true,
-    "number": 44510,
-    "review_comments_count": 1,
+    "number": 44549,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix: Remove references to `text2text-generation`, `summarization` and `translation` pipeline tasks",
-    "updated_at": "2026-03-10T00:39:30Z"
+    "title": "Fix wrong (non-existing) checkpoints",
+    "updated_at": "2026-03-11T15:16:56Z"
   },
   {
-    "additions": 8,
-    "author": "KartikPawade",
+    "additions": 12,
+    "author": "mvanhorn",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Older OwlViT checkpoints stored `position_ids` as buffers in the text and vision embedding modules. These tensors are simple integer ranges (0 \u2192 max sequence length) and are now recomputed dynamically during initial\u2026",
+    "body_excerpt": "Fixes #43935 ## Summary - Added `eval_on_end` argument to `TrainingArguments` (default: `False`) - Added conditional evaluation at the end of training in `Trainer.train()`, symmetric to `eval_on_start` - Implementation mirrors the existing\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44508",
-    "created_at": "2026-03-06T18:49:59Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44548",
+    "created_at": "2026-03-09T15:03:27Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44508/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44508",
+    "files_url": "https://github.com/huggingface/transformers/pull/44548/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44548",
     "labels": [],
-    "merged": true,
-    "number": 44508,
+    "merged": false,
+    "number": 44548,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix unexpected `position_ids` keys when loading OwlViT models",
-    "updated_at": "2026-03-18T18:30:48Z"
+    "title": "Add eval_on_end flag to Trainer",
+    "updated_at": "2026-03-09T19:31:59Z"
   },
   {
-    "additions": 4,
-    "author": "0xDELUXA",
+    "additions": 2,
+    "author": "mvanhorn",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? `torch.distributed.fsdp` is not available in all PyTorch builds (for example, Windows ROCm). Importing it unconditionally at the top level causes an immediate crash with: ``` ModuleNotFoundError: No module named 'to\u2026",
+    "body_excerpt": "Fixes #44373 ## Summary - Corrected the docstring for `position_ids` parameter in `prepare_fa_kwargs_from_position_ids` and `_prepare_from_posids` which incorrectly described attention mask semantics (\"Boolean or int tensor... 1 means vali\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44507",
-    "created_at": "2026-03-06T18:03:49Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44547",
+    "created_at": "2026-03-09T14:59:27Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44507/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44507",
+    "files_url": "https://github.com/huggingface/transformers/pull/44547/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44547",
+    "labels": [],
+    "merged": false,
+    "number": 44547,
+    "review_comments_count": 3,
+    "state": "open",
+    "title": "Fix position_ids docstring in modeling_flash_attention_utils.py",
+    "updated_at": "2026-03-09T16:08:29Z"
+  },
+  {
+    "additions": 4,
+    "author": "Abdennacer-Badaoui",
+    "author_association": "MEMBER",
+    "body_excerpt": "The AMD Docker image build (`latest-pytorch-amd`) has been failing since early February due to the 6h GitHub Actions job time limit being exceeded. The root cause is that Flash Attention is cloned and compiled from the latest commit on the\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44546",
+    "created_at": "2026-03-09T14:15:51Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44546/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44546",
     "labels": [],
     "merged": true,
-    "number": 44507,
+    "number": 44546,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix: Conditionally import `torch.distributed.fsdp` in `trainer_seq2seq.py`",
-    "updated_at": "2026-03-13T10:17:56Z"
+    "title": "Fix AMD Docker image build timeout by pinning Flash Attention commit",
+    "updated_at": "2026-03-09T14:37:50Z"
   },
   {
-    "additions": 1,
-    "author": "michaelbenayoun",
+    "additions": 18,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The current implementation does not work with the `mps` device and TP. ## Example script script.py ``` import os os.environ[\"PYTORCH_ENABLE_MPS_FALLBACK\"] = \"1\" import torch from transformers import AutoModelForCaus\u2026",
+    "body_excerpt": "# What does this PR do? As per the title. Fixes https://github.com/huggingface/transformers/issues/44336",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44506",
-    "created_at": "2026-03-06T18:03:33Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44544",
+    "created_at": "2026-03-09T11:38:09Z",
+    "deletions": 18,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44506/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44506",
+    "files_url": "https://github.com/huggingface/transformers/pull/44544/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44544",
     "labels": [],
     "merged": true,
-    "number": 44506,
+    "number": 44544,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Tensor Parallelism and `mps` device",
-    "updated_at": "2026-03-11T15:16:49Z"
+    "title": "Fix ansi codes in loading reports when not connected to terminal",
+    "updated_at": "2026-03-09T11:52:16Z"
   },
   {
-    "additions": 16,
-    "author": "kushalkkb",
-    "author_association": "NONE",
-    "body_excerpt": "This PR improves error handling in the load_vocab function. Changes: - Added validation to ensure vocab_file is a string path - Added check for file existence - Raised clearer FileNotFoundError when vocabulary file is missing This improves\u2026",
-    "changed_files": 1,
+    "additions": 154,
+    "author": "umbilnm",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #44521 `apply_chat_template` with `return_assistant_tokens_mask=True` returns all-zero masks when multimodal inputs (images/videos) are present. ## Root cause `generation_indices` (character-level positions of\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44505",
-    "created_at": "2026-03-06T17:47:37Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44543",
+    "created_at": "2026-03-09T10:47:05Z",
+    "deletions": 21,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44505/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44505",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44543/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44543",
+    "labels": [],
     "merged": false,
-    "number": 44505,
+    "number": 44543,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Improve error handling in load_vocab for invalid vocabulary path",
-    "updated_at": "2026-03-10T04:14:31Z"
+    "state": "open",
+    "title": "Fix assistant_masks for multimodal inputs in apply_chat_template",
+    "updated_at": "2026-03-09T10:47:05Z"
   },
   {
-    "additions": 13,
-    "author": "kushalkkb",
-    "author_association": "NONE",
-    "body_excerpt": "This PR improves error handling in the load_vocab function. Changes: - Added validation to ensure vocab_file is a string path - Added check for file existence - Raised clearer FileNotFoundError when vocabulary file is missing This improves\u2026",
-    "changed_files": 1,
+    "additions": 3,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44504",
-    "created_at": "2026-03-06T17:24:10Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44542",
+    "created_at": "2026-03-09T10:00:29Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44504/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44504",
+    "files_url": "https://github.com/huggingface/transformers/pull/44542/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44542",
     "labels": [],
-    "merged": false,
-    "number": 44504,
+    "merged": true,
+    "number": 44542,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Improve error handling in load_vocab for invalid vocabulary path",
-    "updated_at": "2026-03-06T17:46:17Z"
+    "title": "Fix backend dependency",
+    "updated_at": "2026-03-09T10:10:24Z"
   },
   {
-    "additions": 8,
-    "author": "harshaljanjani",
+    "additions": 1,
+    "author": "YangKai0616",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following failing tests were identified and fixed in this PR: \u2192 **Moonshine:** In [MoonshineEncoder.forward](https://github.com/huggingface/transformers/blob/main/src/transformers/models/moonshine/modular_moon\u2026",
-    "changed_files": 4,
+    "body_excerpt": "Per the title, an error occurs when `tp_plan` is empty due to [here](https://github.com/huggingface/transformers/blob/701628527ae1ef37473f05f5d94fac7f457a3f8f/src/transformers/core_model_loading.py#L1120): ``` [rank0]: Traceback (most rece\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44540",
+    "created_at": "2026-03-09T09:37:36Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44540/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44540",
+    "labels": [],
+    "merged": true,
+    "number": 44540,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix UnboundLocalError for tp_plan_alt when tp_plan is empty",
+    "updated_at": "2026-03-11T13:42:39Z"
+  },
+  {
+    "additions": 1333,
+    "author": "kmswin1",
+    "author_association": "NONE",
+    "body_excerpt": "A.X K1 \ubaa8\ub378 \ucd94\uac00\ud569\ub2c8\ub2e4. \uae30\uc874 inference \uc640 \ub3d9\uc77c\ud55c \uac83 \ud655\uc778\ud588\uc2b5\ub2c8\ub2e4.",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44503",
-    "created_at": "2026-03-06T17:08:00Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44539",
+    "created_at": "2026-03-09T09:03:10Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44503/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44503",
+    "files_url": "https://github.com/huggingface/transformers/pull/44539/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44539",
     "labels": [],
-    "merged": true,
-    "number": 44503,
+    "merged": false,
+    "number": 44539,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(testing): Fix MoonshineEncoder UnboundLocalError and Florence2VisionBackbone dtype mismatch",
-    "updated_at": "2026-03-09T18:06:17Z"
+    "title": "Add A.X K1",
+    "updated_at": "2026-03-09T09:04:11Z"
   },
   {
-    "additions": 1,
-    "author": "Cyrilvallez",
+    "additions": 14,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. Introduced in https://github.com/huggingface/transformers/pull/44381, not sure why the CI passed",
+    "body_excerpt": "# What does this PR do? Follow-up of #44532: we need to change the required status check to the new added job `doc_build_status_check` added in this PR, otherwise the merge queue won't get the required (passing) status and will eventually\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44502",
-    "created_at": "2026-03-06T17:03:17Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44538",
+    "created_at": "2026-03-09T08:59:12Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44502/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44502",
+    "files_url": "https://github.com/huggingface/transformers/pull/44538/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44538",
     "labels": [],
     "merged": true,
-    "number": 44502,
+    "number": 44538,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix type checker",
-    "updated_at": "2026-03-06T17:09:37Z"
+    "title": "Add a new job in `build_pr_documentation.yml` (will be the new required job)",
+    "updated_at": "2026-03-09T09:12:25Z"
   },
   {
-    "additions": 1,
-    "author": "frogNotToad",
+    "additions": 2,
+    "author": "YangKai0616",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Changes the word \"maximize\" to \"minimize\" in the docs Fixes #44492 ## Before submitting - [x] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). ## Who can review? Anyon\u2026",
+    "body_excerpt": "Per the title. Supplement skip logic for XPU in the CPU-only tp tests. Hi @IlyasMoutawwakil, please help review, thanks!",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44501",
-    "created_at": "2026-03-06T16:58:14Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44536",
+    "created_at": "2026-03-09T08:09:41Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44501/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44501",
+    "files_url": "https://github.com/huggingface/transformers/pull/44536/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44536",
     "labels": [],
     "merged": true,
-    "number": 44501,
+    "number": 44536,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fixed typo in docs/source/en/kv_cache.md",
-    "updated_at": "2026-03-06T20:05:36Z"
+    "title": "Supplement skip logic for XPU in the CPU-only tp tests",
+    "updated_at": "2026-03-09T10:10:49Z"
   },
   {
-    "additions": 18,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? - Do proper type check in case jax is installed. - Make sure older torch versions don't raise typing issues",
-    "changed_files": 4,
+    "additions": 53,
+    "author": "Anakintano",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Problem `Qwen2_5_VLProcessor.apply_chat_template` raises `ValueError: setting an array element with a sequence` when called with a batch of \u22652 conversations that include images under the default `padding=False` setting. **Root cause:**\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44500",
-    "created_at": "2026-03-06T16:56:12Z",
-    "deletions": 16,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44535",
+    "created_at": "2026-03-09T07:54:58Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44500/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44500",
+    "files_url": "https://github.com/huggingface/transformers/pull/44535/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44535",
     "labels": [],
-    "merged": true,
-    "number": 44500,
+    "merged": false,
+    "number": 44535,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Follow-up typing checking fixes",
-    "updated_at": "2026-03-09T10:47:31Z"
+    "state": "open",
+    "title": "Fix crash in Qwen2_5_VLProcessor when using batched input with padding=False",
+    "updated_at": "2026-03-09T12:44:00Z"
   },
   {
-    "additions": 11,
-    "author": "michaelbenayoun",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR is just a small cleanup. The `TensorParallelLayer` class defines `_prepare_input_fn` and `_prepare_output_fn` as static methods. But then these methods end-up being instance or static methods in the sub clas\u2026",
-    "changed_files": 1,
+    "additions": 0,
+    "author": "stargazerwh",
+    "author_association": "NONE",
+    "body_excerpt": "## Description The 'transformers run' command was removed in Transformers v5, but the documentation still contained references in 62 model documentation files. This PR removes all `<hfoption id='transformers CLI'>` and `<hfoption id='trans\u2026",
+    "changed_files": 62,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44499",
-    "created_at": "2026-03-06T16:46:18Z",
-    "deletions": 18,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44533",
+    "created_at": "2026-03-09T01:39:32Z",
+    "deletions": 439,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44499/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44499",
+    "files_url": "https://github.com/huggingface/transformers/pull/44533/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44533",
     "labels": [],
-    "merged": true,
-    "number": 44499,
+    "merged": false,
+    "number": 44533,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Make `_prepare_input_fn` and `_prepare_output_fn` instance methods",
-    "updated_at": "2026-03-10T13:53:18Z"
+    "title": "docs: Remove references to removed 'transformers run' CLI command",
+    "updated_at": "2026-03-09T02:33:08Z"
   },
   {
-    "additions": 1,
-    "author": "michaelbenayoun",
+    "additions": 9,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Add the `neuron` backend for initialization in TP.",
+    "body_excerpt": "# What does this PR do? So we can use `Require Merge Queue` functionoality",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44498",
-    "created_at": "2026-03-06T16:23:18Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44532",
+    "created_at": "2026-03-08T20:34:05Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44498/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44498",
+    "files_url": "https://github.com/huggingface/transformers/pull/44532/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44532",
     "labels": [],
     "merged": true,
-    "number": 44498,
+    "number": 44532,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "feat: add neuron in tensor parallelism initialization",
-    "updated_at": "2026-03-12T18:07:52Z"
+    "title": "Update `build_pr_documentation` workflow for `merge_group` event",
+    "updated_at": "2026-03-08T20:42:57Z"
   },
   {
-    "additions": 43,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. Fixes https://github.com/huggingface/transformers/issues/44466 and avoid issues with torch `.bin` checkpoints which always contain both keys!",
+    "additions": 16,
+    "author": "s-zx",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes `Qwen2_5_VLProcessor.apply_chat_template` crashing with a `ValueError` when called with a batch of conversations with different prompt lengths and `padding=False` (the default). ### Root cause In the `mm_toke\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44497",
-    "created_at": "2026-03-06T16:21:14Z",
-    "deletions": 14,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44531",
+    "created_at": "2026-03-08T19:38:00Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44497/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44497",
+    "files_url": "https://github.com/huggingface/transformers/pull/44531/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44531",
     "labels": [],
-    "merged": true,
-    "number": 44497,
+    "merged": false,
+    "number": 44531,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[tie weights] \ud83d\udea8 If both weights are present with same weights, still tie them",
-    "updated_at": "2026-03-09T15:00:25Z"
+    "title": "Fix Qwen2_5_VLProcessor.apply_chat_template crash on unpadded batched input",
+    "updated_at": "2026-03-09T13:14:02Z"
   },
   {
-    "additions": 69,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "As per title, WIP",
-    "changed_files": 354,
+    "additions": 3,
+    "author": "ydshieh2",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Merge queue is enabled, but we need to see if it works well with CircleCI",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44495",
-    "created_at": "2026-03-06T13:57:04Z",
-    "deletions": 521,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44495/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44495",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44529",
+    "created_at": "2026-03-08T18:12:54Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44529/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44529",
     "labels": [],
-    "merged": false,
-    "number": 44495,
+    "merged": true,
+    "number": 44529,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[`Gradient Ckpting`] Remove unnecessary attribute definitions",
-    "updated_at": "2026-03-06T13:58:22Z"
+    "state": "closed",
+    "title": "test merge queue 1",
+    "updated_at": "2026-03-09T21:01:09Z"
   },
   {
-    "additions": 13,
-    "author": "tarekziade",
+    "additions": 3,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "- updates `ty` to `0.2.0` - pinned regex package (older versions did not have typing stubs) - fixed a couple of typing failures that went through via other parallel branches",
-    "changed_files": 6,
+    "body_excerpt": "# What does this PR do? Merge queue is enabled, but we need to see if it works well with CircleCI",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44494",
-    "created_at": "2026-03-06T12:57:25Z",
-    "deletions": 11,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44528",
+    "created_at": "2026-03-08T17:54:32Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44494/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44494",
+    "files_url": "https://github.com/huggingface/transformers/pull/44528/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44528",
     "labels": [],
-    "merged": true,
-    "number": 44494,
-    "review_comments_count": 3,
+    "merged": false,
+    "number": 44528,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Update `ty` to 0.0.20",
-    "updated_at": "2026-03-06T13:30:25Z"
+    "title": "test merge queue 1",
+    "updated_at": "2026-03-09T20:39:15Z"
   },
   {
-    "additions": 439,
-    "author": "SunMarc",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Since I removed some folders (fsdp, deepspeed) related to training, I need to modify the workflows !",
-    "changed_files": 18,
+    "additions": 3,
+    "author": "Sai-Suraj-27",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes these failing [MusicgenStereoIntegrationTests](https://github.com/huggingface/transformers/actions/runs/22606636929/job/65500451700#step:14:7870) ## Before submitting - [ ] This PR fixes a typo or improves the\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44491",
-    "created_at": "2026-03-06T11:15:42Z",
-    "deletions": 647,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44527",
+    "created_at": "2026-03-08T14:26:02Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44491/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44491",
+    "files_url": "https://github.com/huggingface/transformers/pull/44527/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44527",
     "labels": [],
     "merged": true,
-    "number": 44491,
-    "review_comments_count": 3,
+    "number": 44527,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix training ci and clean some tests",
-    "updated_at": "2026-03-11T16:27:57Z"
+    "title": "Fix failing `MusicgenStereo` integration tests",
+    "updated_at": "2026-03-10T12:28:39Z"
   },
   {
-    "additions": 4,
-    "author": "kaixuanliu",
+    "additions": 90,
+    "author": "JoursBleu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@ArthurZucker @Cyrilvallez pls help review, thx! This PR fixes failed test case: `pytest -rA tests/models/eurobert/test_modeling_eurobert.py::EuroBertModelTest::test_model_parallelism`",
+    "body_excerpt": "# What does this PR do? Add GGUF loading support for MiniMax-M2.1 (456B MoE) model. MiniMax-M2.1 is a large Mixture-of-Experts model with 456B total parameters (45.9B active), 256 experts and 8 experts per token. This PR enables loading it\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44490",
-    "created_at": "2026-03-06T10:56:48Z",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44526",
+    "created_at": "2026-03-08T09:57:38Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44490/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44490",
+    "files_url": "https://github.com/huggingface/transformers/pull/44526/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44526",
     "labels": [],
     "merged": true,
-    "number": 44490,
-    "review_comments_count": 0,
+    "number": 44526,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "fix model parallelism bug for eurobert model",
-    "updated_at": "2026-04-01T08:25:28Z"
+    "title": "Add GGUF support for MiniMax-M2.1 model",
+    "updated_at": "2026-03-18T14:39:20Z"
   },
   {
-    "additions": 310,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "This PR makes `.ai` the single source of truth for agent templates and skills, and adds explicit `Makefile` targets to generate `Codex` and `Claude Code` specific artifacts. It contains a first skill aimed at properly dealing with typing e\u2026",
-    "changed_files": 7,
+    "additions": 1,
+    "author": "jnMetaCode",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes a `KeyError` crash in `_parse_type_hint` in `chat_template_utils.py` (line 117). When processing Union types, the code accesses `subtype[\"type\"]` without checking the key exists. `_get_json_schema_type(Any)` returns `{}` (\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44489",
-    "created_at": "2026-03-06T08:42:12Z",
-    "deletions": 62,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44525",
+    "created_at": "2026-03-08T09:21:27Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44489/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44489",
+    "files_url": "https://github.com/huggingface/transformers/pull/44525/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44525",
     "labels": [],
     "merged": true,
-    "number": 44489,
-    "review_comments_count": 2,
+    "number": 44525,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Centralize AI agent templates in `.ai`",
-    "updated_at": "2026-03-18T14:17:22Z"
+    "title": "Fix KeyError in _parse_type_hint when Union contains Any",
+    "updated_at": "2026-03-09T13:43:23Z"
   },
   {
-    "additions": 482,
-    "author": "abhijeet-dhumal",
+    "additions": 1,
+    "author": "jnMetaCode",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #44486 Adds `KubeflowCallback` to enable automatic progress and metrics reporting for training jobs running on [Kubeflow Trainer](https://github.com/kubeflow/trainer). When training runs inside a Kubeflow Trai\u2026",
-    "changed_files": 6,
+    "body_excerpt": "## Summary Fixes a bug in `AssistantTracker.is_active()` in `chat_template_utils.py`. After activation via `activate_tracker()`, `_rendered_blocks` and `_generation_indices` are set to list arguments which may be empty `[]`. The `is_active\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44487",
-    "created_at": "2026-03-06T08:31:30Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44524",
+    "created_at": "2026-03-08T09:21:25Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44487/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44487",
+    "files_url": "https://github.com/huggingface/transformers/pull/44524/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44524",
     "labels": [],
     "merged": true,
-    "number": 44487,
-    "review_comments_count": 8,
+    "number": 44524,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "feat(integration): Add KubeflowCallback to enable automatic progress \u2026",
-    "updated_at": "2026-03-18T14:58:23Z"
+    "title": "Fix AssistantTracker.is_active() returning False after activation with empty lists",
+    "updated_at": "2026-03-09T13:36:19Z"
   },
   {
-    "additions": 691,
-    "author": "kaixuanliu",
+    "additions": 2,
+    "author": "jnMetaCode",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@IlyasMoutawwakil pls help review, thx!",
+    "body_excerpt": "## Summary Fixes two small bugs in `load_sharded_checkpoint` in `trainer_utils.py`: **Bug 1 \u2014 Copy-paste error in error message (line 1108):** When reporting unexpected keys, the error message incorrectly says \"Missing key(s)\" instead of \"\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44482",
-    "created_at": "2026-03-06T02:39:41Z",
-    "deletions": 332,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44523",
+    "created_at": "2026-03-08T09:21:22Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44482/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44482",
+    "files_url": "https://github.com/huggingface/transformers/pull/44523/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44523",
     "labels": [],
     "merged": true,
-    "number": 44482,
+    "number": 44523,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "add XPU Expectations for higgs_audio_v2 tests",
-    "updated_at": "2026-04-09T02:32:48Z"
+    "title": "Fix error message label and docstring default in load_sharded_checkpoint",
+    "updated_at": "2026-03-10T15:48:41Z"
   },
   {
-    "additions": 2353,
-    "author": "XingyuHu109",
+    "additions": 41,
+    "author": "nakigami",
     "author_association": "NONE",
-    "body_excerpt": "## Summary This PR adds native Transformers support for DeepSeek-V3.2. It introduces a new `deepseek_v32` model family so the official checkpoints resolve through the standard auto classes without `trust_remote_code`. The implementation ke\u2026",
-    "changed_files": 19,
+    "body_excerpt": "# What does this PR do? This PR introduces initial unit test coverage for the `transformers-cli` tool, specifically focusing on diagnostic and model utility commands. Currently, these CLI entry points lack automated tests. These new tests\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44481",
-    "created_at": "2026-03-05T21:14:38Z",
-    "deletions": 30,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44520",
+    "created_at": "2026-03-08T01:30:39Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44481/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44481",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44520/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44520",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44481,
-    "review_comments_count": 4,
+    "number": 44520,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add native DeepSeek-V3.2 support",
-    "updated_at": "2026-04-14T21:34:03Z"
+    "title": "test(cli): add unit tests for env and model utility commands",
+    "updated_at": "2026-03-09T13:19:15Z"
   },
   {
     "additions": 3,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? add `diffusers` to docker file for `VibeVoice` (added in PR #40546).",
+    "author": "Sai-Suraj-27",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes these failing [MarianIntegrationTests](https://github.com/huggingface/transformers/actions/runs/22606636929/job/65500458014#step:14:6186) <img width=\"2378\" height=\"657\" alt=\"image\" src=\"https://github.com/user\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44480",
-    "created_at": "2026-03-05T20:54:07Z",
-    "deletions": 0,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44519",
+    "created_at": "2026-03-07T19:53:23Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44480/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44480",
+    "files_url": "https://github.com/huggingface/transformers/pull/44519/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44519",
     "labels": [],
     "merged": true,
-    "number": 44480,
+    "number": 44519,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add `diffusers` to CI docker file",
-    "updated_at": "2026-03-05T21:11:17Z"
+    "title": "Fix failing `MarianIntegrationTests`",
+    "updated_at": "2026-03-09T14:11:12Z"
   },
   {
-    "additions": 116,
-    "author": "BenjaminBossan",
-    "author_association": "MEMBER",
-    "body_excerpt": "Required fixes: - some code was using unordered data structures, making weight order random - adjust alpha to offset increased rank from fusion - import functions from PEFT if available See https://github.com/huggingface/peft/pull/3083.",
-    "changed_files": 4,
+    "additions": 12,
+    "author": "KartikPawade",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #44514 `Qwen2_5_VLProcessor.__call__` crashed with a `ValueError` when processing a batch of conversations with different lengths and `padding=False` (the default). **Root cause:** The `mm_token_type_ids` blo\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44478",
-    "created_at": "2026-03-05T17:19:31Z",
-    "deletions": 26,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44518",
+    "created_at": "2026-03-07T19:22:40Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44478/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44478",
+    "files_url": "https://github.com/huggingface/transformers/pull/44518/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44518",
     "labels": [],
-    "merged": true,
-    "number": 44478,
-    "review_comments_count": 1,
+    "merged": false,
+    "number": 44518,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[WIP] FIX Make Mixtral LoRA loading work",
-    "updated_at": "2026-03-11T17:44:20Z"
+    "title": "fix: Qwen2_5_VLProcessor crashes on batched input when padding=False \u2026",
+    "updated_at": "2026-03-10T18:57:10Z"
   },
   {
-    "additions": 1,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. It's quite a random rule to fix https://huggingface.co/fixie-ai/ultravox-v0_5-llama-3_2-1b to be honest",
-    "changed_files": 1,
+    "additions": 12637,
+    "author": "ShahVandit",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Adds Qwen3-TTS, a series of text-to-speech models by the Qwen team (Alibaba Group), to Transformers. **Architecture:** - `Qwen3TTSForConditionalGeneration` \u2014 text to multi-codebook speech codes (talker) - `Qwen3TTS\u2026",
+    "changed_files": 16,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44477",
-    "created_at": "2026-03-05T16:58:29Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44517",
+    "created_at": "2026-03-07T18:48:04Z",
+    "deletions": 24,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44477/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44477",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44517/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44517",
+    "labels": [
+      "New model",
+      "Audio"
+    ],
     "merged": false,
-    "number": 44477,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "[vllm compat] Fix remote code inits",
-    "updated_at": "2026-03-11T10:34:06Z"
-  },
-  {
-    "additions": 4,
-    "author": "Rocketknight1",
-    "author_association": "MEMBER",
-    "body_excerpt": "I made an oversight in the fix at #43981 - I didn't realize the dim order changed for torch, so the test was still flaky for torch tensors. The fix reduced the flaky frequency a lot so I thought it had been fixed, but actually it's still t\u2026",
-    "changed_files": 1,
+    "number": 44517,
+    "review_comments_count": 19,
+    "state": "open",
+    "title": "Add qwen3 tts",
+    "updated_at": "2026-04-13T07:53:06Z"
+  },
+  {
+    "additions": 65,
+    "author": "JasonCZMeng",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fix `Qwen2_5_VLProcessor.apply_chat_template` crashing with `ValueError` when called with batched inputs of different sequence lengths (ragged lists) and `padding=False` (the default). Fixes #44514 ## Root Cause The `mm_token_ty\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44476",
-    "created_at": "2026-03-05T16:39:44Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44516",
+    "created_at": "2026-03-07T18:33:40Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44476/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44476",
+    "files_url": "https://github.com/huggingface/transformers/pull/44516/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44516",
     "labels": [],
-    "merged": true,
-    "number": 44476,
+    "merged": false,
+    "number": 44516,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Llava tests for torch too!",
-    "updated_at": "2026-03-11T16:47:05Z"
+    "title": "fix(qwen2_5_vl): handle ragged batched input in apply_chat_template",
+    "updated_at": "2026-03-09T13:14:22Z"
   },
   {
     "additions": 1,
-    "author": "itazap",
-    "author_association": "MEMBER",
-    "body_excerpt": "chameleon added to MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS",
+    "author": "Sai-Suraj-27",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes this failing [GPTNeoModelLanguageGenerationTest](https://github.com/huggingface/transformers/actions/runs/22606636929/job/65500455894#step:14:1483) <img width=\"2363\" height=\"239\" alt=\"image\" src=\"https://githu\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44475",
-    "created_at": "2026-03-05T16:29:18Z",
-    "deletions": 0,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44515",
+    "created_at": "2026-03-07T18:16:35Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44475/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44475",
+    "files_url": "https://github.com/huggingface/transformers/pull/44515/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44515",
     "labels": [],
     "merged": true,
-    "number": 44475,
+    "number": 44515,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "chameleon added to MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS",
-    "updated_at": "2026-03-09T22:33:20Z"
+    "title": "Fix failing `GPTNeoModelLanguageGenerationTest`",
+    "updated_at": "2026-03-09T14:11:21Z"
   },
   {
-    "additions": 875,
-    "author": "JJJYmmm",
+    "additions": 1,
+    "author": "math-hiyoko",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fix https://github.com/QwenLM/Qwen3.5/issues/58. In the latest code, Qwen3VL and Qwen3.5 use the same `get_rope_index` func of Qwen2VL. But they should be different since Qwen3VL/Qwen3.5 introduce text timestamps. T\u2026",
-    "changed_files": 9,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44474",
-    "created_at": "2026-03-05T15:46:09Z",
-    "deletions": 107,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44474/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44474",
-    "labels": [],
-    "merged": true,
-    "number": 44474,
-    "review_comments_count": 10,
-    "state": "closed",
-    "title": "[Bugfix] fix video inference of qwen3vl and qwen3.5 series",
-    "updated_at": "2026-03-10T09:52:44Z"
-  },
-  {
-    "additions": 137,
-    "author": "winglian",
-    "author_association": "COLLABORATOR",
-    "body_excerpt": "# What does this PR do? supersedes #44446 on `main`, when loading to cpu and using meta devices for non-rank0 processes, it now re-initializes weights on those processes as well as uses more CPU memory. In testing with loading llama3-8b. m\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 69,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44473",
-    "created_at": "2026-03-05T14:52:15Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44513",
+    "created_at": "2026-03-07T16:11:55Z",
+    "deletions": 492,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44473/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44473",
+    "files_url": "https://github.com/huggingface/transformers/pull/44513/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44513",
     "labels": [],
     "merged": true,
-    "number": 44473,
-    "review_comments_count": 4,
+    "number": 44513,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix FSDP loading with meta devices",
-    "updated_at": "2026-03-09T15:46:22Z"
+    "title": "Fix: Remove references to transformers run command",
+    "updated_at": "2026-03-09T15:37:16Z"
   },
   {
-    "additions": 13,
-    "author": "jblox26",
+    "additions": 4,
+    "author": "04cb",
     "author_association": "NONE",
-    "body_excerpt": "## What does this fix? Running video inference with any `Qwen3VL` model raises `StopIteration` during `model.generate()`: ``` File \".../transformers/models/qwen3_vl/modeling_qwen3_vl.py\", line 1126, in get_rope_index grid_thw = next(grid_i\u2026",
+    "body_excerpt": "Fixes #44355. The inspect.getsource() call raises TypeError when running compiled Python files with Cython-compiled functions. Added try-except block to gracefully handle this case by returning a default indentation level of 4.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44472",
-    "created_at": "2026-03-05T14:50:06Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44511",
+    "created_at": "2026-03-07T05:36:25Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44472/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44472",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44511/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44511",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44472,
+    "number": 44511,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Qwen3VL get_rope_index StopIteration with per-frame video tokens",
-    "updated_at": "2026-03-06T15:15:58Z"
+    "title": "Fix get_docstring_indentation_level to handle compiled functions",
+    "updated_at": "2026-03-09T13:10:06Z"
   },
   {
-    "additions": 50,
-    "author": "weiguangli-io",
+    "additions": 3,
+    "author": "math-hiyoko",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #44466 After `.to(device)`, PyTorch's `Module._apply` may create new `Parameter` objects that no longer share storage with tied weights. This caused `remove_tied_weights_from_state_dict` to fail to detect and\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 27,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44471",
-    "created_at": "2026-03-05T14:30:17Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44510",
+    "created_at": "2026-03-06T23:37:51Z",
+    "deletions": 358,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44471/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44471",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44471,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44510/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44510",
+    "labels": [],
+    "merged": true,
+    "number": 44510,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Fix tied weights serialization being device-dependent",
-    "updated_at": "2026-03-06T14:03:18Z"
+    "title": "Fix: Remove references to `text2text-generation`, `summarization` and `translation` pipeline tasks",
+    "updated_at": "2026-03-10T00:39:30Z"
   },
   {
     "additions": 8,
-    "author": "weiguangli-io",
+    "author": "KartikPawade",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes #44360 The reference `fp8_index` kernel clamps per-head q\u00b7k scores with `T.max(logits, 0)` before the weighted sum across heads ([kernel.py#L241](https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp/blob/main/inference/kernel.py#L241\u2026",
+    "body_excerpt": "# What does this PR do? Older OwlViT checkpoints stored `position_ids` as buffers in the text and vision embedding modules. These tensors are simple integer ranges (0 \u2192 max sequence length) and are now recomputed dynamically during initial\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44470",
-    "created_at": "2026-03-05T14:02:05Z",
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44508",
+    "created_at": "2026-03-06T18:49:59Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44470/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44470",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44470,
+    "files_url": "https://github.com/huggingface/transformers/pull/44508/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44508",
+    "labels": [],
+    "merged": true,
+    "number": 44508,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add missing ReLU in GlmMoeDsaIndexer",
-    "updated_at": "2026-03-05T15:39:38Z"
+    "title": "Fix unexpected `position_ids` keys when loading OwlViT models",
+    "updated_at": "2026-03-18T18:30:48Z"
   },
   {
     "additions": 4,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? For remote code that behave correctly with tied weights, we need to keep the same behavior as for the main lib, i.e. not remove them from tied weights (as tied weights are marked as missing to avoid inits!!)",
+    "author": "0xDELUXA",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? `torch.distributed.fsdp` is not available in all PyTorch builds (for example, Windows ROCm). Importing it unconditionally at the top level causes an immediate crash with: ``` ModuleNotFoundError: No module named 'to\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44469",
-    "created_at": "2026-03-05T13:51:55Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44507",
+    "created_at": "2026-03-06T18:03:49Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44469/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44469",
+    "files_url": "https://github.com/huggingface/transformers/pull/44507/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44507",
     "labels": [],
     "merged": true,
-    "number": 44469,
+    "number": 44507,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[remote code/vllm] Fix incorrect tied weights",
-    "updated_at": "2026-03-05T15:07:56Z"
+    "title": "Fix: Conditionally import `torch.distributed.fsdp` in `trainer_seq2seq.py`",
+    "updated_at": "2026-03-13T10:17:56Z"
   },
   {
-    "additions": 13,
-    "author": "itazap",
+    "additions": 1,
+    "author": "michaelbenayoun",
     "author_association": "MEMBER",
-    "body_excerpt": "Replace placeholder tokens as specified in added_tokens_decoder if we have added_tokens_decoder with specific token_ids, we need to overwrite them in spm model ! example: [UNUSED_TOKEN_146] -> <|im_start|> see internlm2: https://huggingfac\u2026",
+    "body_excerpt": "# What does this PR do? The current implementation does not work with the `mps` device and TP. ## Example script script.py ``` import os os.environ[\"PYTORCH_ENABLE_MPS_FALLBACK\"] = \"1\" import torch from transformers import AutoModelForCaus\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44468",
-    "created_at": "2026-03-05T13:48:56Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44506",
+    "created_at": "2026-03-06T18:03:33Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44468/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44468",
+    "files_url": "https://github.com/huggingface/transformers/pull/44506/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44506",
     "labels": [],
     "merged": true,
-    "number": 44468,
+    "number": 44506,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Replace placeholder tokens as specified in added_tokens_decoder",
-    "updated_at": "2026-03-05T16:29:13Z"
-  },
-  {
-    "additions": 346,
-    "author": "itazap",
-    "author_association": "MEMBER",
-    "body_excerpt": "Replace placeholder tokens as specified in added_tokens_decoder if we have `added_tokens_decoder` with specific token_ids, we need to overwrite them in spm model ! `example: [UNUSED_TOKEN_146] -> <|im_start|>` see internlm2: https://huggin\u2026",
-    "changed_files": 24,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44467",
-    "created_at": "2026-03-05T13:44:54Z",
-    "deletions": 204,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44467/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44467",
-    "labels": [],
-    "merged": false,
-    "number": 44467,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Placeholder tokens update",
-    "updated_at": "2026-03-05T13:47:28Z"
+    "title": "Tensor Parallelism and `mps` device",
+    "updated_at": "2026-03-11T15:16:49Z"
   },
   {
-    "additions": 20,
-    "author": "kashif",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix the loss calculation; we should calculate it on scaled targets. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with\u2026",
-    "changed_files": 2,
+    "additions": 16,
+    "author": "kushalkkb",
+    "author_association": "NONE",
+    "body_excerpt": "This PR improves error handling in the load_vocab function. Changes: - Added validation to ensure vocab_file is a string path - Added check for file existence - Raised clearer FileNotFoundError when vocabulary file is missing This improves\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44465",
-    "created_at": "2026-03-05T12:59:23Z",
-    "deletions": 14,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44505",
+    "created_at": "2026-03-06T17:47:37Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44465/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44465",
+    "files_url": "https://github.com/huggingface/transformers/pull/44505/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44505",
     "labels": [
-      "bug"
+      "Code agent slop"
     ],
-    "merged": true,
-    "number": 44465,
+    "merged": false,
+    "number": 44505,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[timesfm2_5] fix loss scaling",
-    "updated_at": "2026-03-05T14:50:26Z"
+    "title": "Improve error handling in load_vocab for invalid vocabulary path",
+    "updated_at": "2026-03-10T04:14:31Z"
   },
   {
-    "additions": 16,
-    "author": "weiguangli-io",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #44462 When a model's `model_type` (e.g. `\"llama\"`) has no entry in `TOKENIZER_MAPPING_NAMES`, `AutoTokenizer.from_pretrained` falls through to loading the tokenizer class declared in `tokenizer_config.json`\u2026",
+    "additions": 13,
+    "author": "kushalkkb",
+    "author_association": "NONE",
+    "body_excerpt": "This PR improves error handling in the load_vocab function. Changes: - Added validation to ensure vocab_file is a string path - Added check for file existence - Raised clearer FileNotFoundError when vocabulary file is missing This improves\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44463",
-    "created_at": "2026-03-05T12:45:57Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44504",
+    "created_at": "2026-03-06T17:24:10Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44463/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44463",
+    "files_url": "https://github.com/huggingface/transformers/pull/44504/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44504",
     "labels": [],
     "merged": false,
-    "number": 44463,
+    "number": 44504,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix AutoTokenizer ignoring tokenizer.json for unregistered model types",
-    "updated_at": "2026-03-07T13:50:44Z"
+    "title": "Improve error handling in load_vocab for invalid vocabulary path",
+    "updated_at": "2026-03-06T17:46:17Z"
   },
   {
-    "additions": 12,
-    "author": "kashif",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? for SP loss we do not have torch device mesh but rather a deepspeed only. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes\u2026",
-    "changed_files": 1,
+    "additions": 8,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? The following failing tests were identified and fixed in this PR: \u2192 **Moonshine:** In [MoonshineEncoder.forward](https://github.com/huggingface/transformers/blob/main/src/transformers/models/moonshine/modular_moon\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44461",
-    "created_at": "2026-03-05T11:39:02Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44503",
+    "created_at": "2026-03-06T17:08:00Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44461/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44461",
+    "files_url": "https://github.com/huggingface/transformers/pull/44503/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44503",
     "labels": [],
     "merged": true,
-    "number": 44461,
-    "review_comments_count": 1,
+    "number": 44503,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[Trainer] fix SP loss",
-    "updated_at": "2026-03-05T13:00:40Z"
+    "title": "fix(testing): Fix MoonshineEncoder UnboundLocalError and Florence2VisionBackbone dtype mismatch",
+    "updated_at": "2026-03-09T18:06:17Z"
   },
   {
     "additions": 1,
-    "author": "3outeille",
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": null,
+    "body_excerpt": "# What does this PR do? As per the title. Introduced in https://github.com/huggingface/transformers/pull/44381, not sure why the CI passed",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44460",
-    "created_at": "2026-03-05T10:53:07Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44502",
+    "created_at": "2026-03-06T17:03:17Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44460/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44460",
+    "files_url": "https://github.com/huggingface/transformers/pull/44502/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44502",
     "labels": [],
     "merged": true,
-    "number": 44460,
+    "number": 44502,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "trigger tensor parallel utils test in the CI",
-    "updated_at": "2026-03-05T11:25:51Z"
+    "title": "Fix type checker",
+    "updated_at": "2026-03-06T17:09:37Z"
   },
   {
-    "additions": 95,
-    "author": "weiguangli-io",
+    "additions": 1,
+    "author": "frogNotToad",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #44458 PR #42848 introduced a regression where `torch.compile` on `MllamaForConditionalGeneration` fails with a C++ compile error from the torch inductor backend (`'tmp2' was not declared in this scope`). The root cause is\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Changes the word \"maximize\" to \"minimize\" in the docs Fixes #44492 ## Before submitting - [x] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). ## Who can review? Anyon\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44459",
-    "created_at": "2026-03-05T07:58:28Z",
-    "deletions": 7,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44501",
+    "created_at": "2026-03-06T16:58:14Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44459/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44459",
+    "files_url": "https://github.com/huggingface/transformers/pull/44501/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44501",
     "labels": [],
-    "merged": false,
-    "number": 44459,
-    "review_comments_count": 2,
+    "merged": true,
+    "number": 44501,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: make Mllama cross attention mask compatible with torch.compile",
-    "updated_at": "2026-03-07T13:50:40Z"
+    "title": "Fixed typo in docs/source/en/kv_cache.md",
+    "updated_at": "2026-03-06T20:05:36Z"
   },
   {
-    "additions": 1,
-    "author": "Sai-Suraj-27",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes this failing [DepthProModelIntegrationTest](https://github.com/huggingface/transformers/actions/runs/22606636929/job/65500453624#step:14:4893). <img width=\"2231\" height=\"99\" alt=\"image\" src=\"https://github.com\u2026",
-    "changed_files": 1,
+    "additions": 18,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? - Do proper type check in case jax is installed. - Make sure older torch versions don't raise typing issues",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44456",
-    "created_at": "2026-03-05T06:01:06Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44500",
+    "created_at": "2026-03-06T16:56:12Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44456/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44456",
+    "files_url": "https://github.com/huggingface/transformers/pull/44500/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44500",
     "labels": [],
     "merged": true,
-    "number": 44456,
+    "number": 44500,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix failing `DepthProModelIntegrationTest`",
-    "updated_at": "2026-03-05T14:52:40Z"
+    "title": "Follow-up typing checking fixes",
+    "updated_at": "2026-03-09T10:47:31Z"
   },
   {
-    "additions": 3,
-    "author": "jiqing-feng",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Check if accelerator exists before using `pin_memory`. reproduce it on a CPU only node: `python examples/pytorch/continuous_batching_simple.py` output: ``` File \"/home/jiqingfe/transformers/src/transformers/generation/continuous_batching/i\u2026",
+    "additions": 11,
+    "author": "michaelbenayoun",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR is just a small cleanup. The `TensorParallelLayer` class defines `_prepare_input_fn` and `_prepare_output_fn` as static methods. But then these methods end-up being instance or static methods in the sub clas\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44455",
-    "created_at": "2026-03-05T05:20:13Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44499",
+    "created_at": "2026-03-06T16:46:18Z",
+    "deletions": 18,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44455/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44455",
+    "files_url": "https://github.com/huggingface/transformers/pull/44499/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44499",
     "labels": [],
     "merged": true,
-    "number": 44455,
-    "review_comments_count": 6,
+    "number": 44499,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix pin_memory for contiguous batching",
-    "updated_at": "2026-03-09T13:49:30Z"
+    "title": "Make `_prepare_input_fn` and `_prepare_output_fn` instance methods",
+    "updated_at": "2026-03-10T13:53:18Z"
   },
   {
-    "additions": 17,
-    "author": "weiguangli-io",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary - Fixes the `_init_weights` method in `PegasusPreTrainedModel` and `MarianPreTrainedModel` to handle sinusoidal position embeddings before calling `super()._init_weights(module)`, preventing the generic `nn.Embedding` branch fro\u2026",
-    "changed_files": 3,
+    "additions": 1,
+    "author": "michaelbenayoun",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Add the `neuron` backend for initialization in TP.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44454",
-    "created_at": "2026-03-05T03:51:38Z",
-    "deletions": 7,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44498",
+    "created_at": "2026-03-06T16:23:18Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44454/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44454",
+    "files_url": "https://github.com/huggingface/transformers/pull/44498/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44498",
     "labels": [],
-    "merged": false,
-    "number": 44454,
+    "merged": true,
+    "number": 44498,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Pegasus sinusoidal position embedding init regression in v5",
-    "updated_at": "2026-03-09T02:17:41Z"
+    "title": "feat: add neuron in tensor parallelism initialization",
+    "updated_at": "2026-03-12T18:07:52Z"
   },
   {
-    "additions": 1,
-    "author": "weiguangli-io",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Fix KeyError in `convert_to_native_format` for dict vocab Fixes #44451 ### Problem `AutoTokenizer.from_pretrained(\"vesteinn/ScandiBERT\")` raises `KeyError: 0` in `convert_to_native_format`. ScandiBERT's `tokenizer_config.json` specifies\u2026",
-    "changed_files": 1,
+    "additions": 43,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title. Fixes https://github.com/huggingface/transformers/issues/44466 and avoid issues with torch `.bin` checkpoints which always contain both keys!",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44452",
-    "created_at": "2026-03-05T03:34:02Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44497",
+    "created_at": "2026-03-06T16:21:14Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44452/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44452",
+    "files_url": "https://github.com/huggingface/transformers/pull/44497/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44497",
     "labels": [],
     "merged": true,
-    "number": 44452,
+    "number": 44497,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix KeyError in convert_to_native_format for dict vocab",
-    "updated_at": "2026-03-19T13:59:23Z"
-  },
-  {
-    "additions": 297,
-    "author": "sandesh-bhandari-dev",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 32,
+    "title": "[tie weights] \ud83d\udea8 If both weights are present with same weights, still tie them",
+    "updated_at": "2026-03-09T15:00:25Z"
+  },
+  {
+    "additions": 69,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "As per title, WIP",
+    "changed_files": 354,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44449",
-    "created_at": "2026-03-05T01:40:47Z",
-    "deletions": 319,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44449/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44449",
-    "labels": [
-      "Code agent slop"
-    ],
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44495",
+    "created_at": "2026-03-06T13:57:04Z",
+    "deletions": 521,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44495/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44495",
+    "labels": [],
     "merged": false,
-    "number": 44449,
+    "number": 44495,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: 3 bugs : MoE aux loss, ANSI TTY leak, pipeline removed and also task error",
-    "updated_at": "2026-03-05T13:22:40Z"
+    "state": "open",
+    "title": "[`Gradient Ckpting`] Remove unnecessary attribute definitions",
+    "updated_at": "2026-03-06T13:58:22Z"
   },
   {
-    "additions": 8,
-    "author": "yonigozlan",
+    "additions": 13,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes MiniCPM-o-2_6 related tests failures in vLLM, and improve backward compatibility with remote code in general. Cc @hmellor @zucchini-nlp",
-    "changed_files": 1,
+    "body_excerpt": "- updates `ty` to `0.2.0` - pinned regex package (older versions did not have typing stubs) - fixed a couple of typing failures that went through via other parallel branches",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44447",
-    "created_at": "2026-03-04T21:55:16Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44494",
+    "created_at": "2026-03-06T12:57:25Z",
+    "deletions": 11,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44447/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44447",
+    "files_url": "https://github.com/huggingface/transformers/pull/44494/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44494",
     "labels": [],
     "merged": true,
-    "number": 44447,
-    "review_comments_count": 0,
+    "number": 44494,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "[vLLM] Fix backward compatibility with hardcoded subprocessors classes in processors",
-    "updated_at": "2026-03-05T16:07:28Z"
+    "title": "Update `ty` to 0.0.20",
+    "updated_at": "2026-03-06T13:30:25Z"
   },
   {
-    "additions": 4,
+    "additions": 439,
     "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR fixes fsdp loading for rank!=0 as they should stay on meta device. This was reverted in a PR that I can find anymore.",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Since I removed some folders (fsdp, deepspeed) related to training, I need to modify the workflows !",
+    "changed_files": 18,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44446",
-    "created_at": "2026-03-04T21:38:00Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44491",
+    "created_at": "2026-03-06T11:15:42Z",
+    "deletions": 647,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44446/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44446",
+    "files_url": "https://github.com/huggingface/transformers/pull/44491/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44491",
     "labels": [],
-    "merged": false,
-    "number": 44446,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44491,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Fix fdsp loading meta device",
-    "updated_at": "2026-03-05T15:07:57Z"
+    "title": "Fix training ci and clean some tests",
+    "updated_at": "2026-03-11T16:27:57Z"
   },
   {
-    "additions": 2282,
-    "author": "gabe-l-hart",
+    "additions": 4,
+    "author": "kaixuanliu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds support for the forthcoming Granite Docling model based on the Granite 4 LLM architecture (`GraniteMoeHybrid`). ## Draft Status This PR is in draft pending the possibility of some additional changes: -\u2026",
-    "changed_files": 8,
+    "body_excerpt": "@ArthurZucker @Cyrilvallez pls help review, thx! This PR fixes failed test case: `pytest -rA tests/models/eurobert/test_modeling_eurobert.py::EuroBertModelTest::test_model_parallelism`",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44445",
-    "created_at": "2026-03-04T20:54:17Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44490",
+    "created_at": "2026-03-06T10:56:48Z",
     "deletions": 0,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44445/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44445",
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44490/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44490",
     "labels": [],
-    "merged": false,
-    "number": 44445,
-    "review_comments_count": 44,
-    "state": "open",
-    "title": "Adding support for GraniteDoclingHybrid",
-    "updated_at": "2026-03-11T19:19:44Z"
+    "merged": true,
+    "number": 44490,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix model parallelism bug for eurobert model",
+    "updated_at": "2026-04-01T08:25:28Z"
   },
   {
-    "additions": 7,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following failing use case was identified and fixed in this PR: \u2192 [TOKENIZER_MAPPING_NAMES](https://github.com/huggingface/transformers/blob/main/src/transformers/models/auto/tokenization_auto.py#L63-L338) doe\u2026",
-    "changed_files": 1,
+    "additions": 310,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "This PR makes `.ai` the single source of truth for agent templates and skills, and adds explicit `Makefile` targets to generate `Codex` and `Claude Code` specific artifacts. It contains a first skill aimed at properly dealing with typing e\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44443",
-    "created_at": "2026-03-04T20:01:55Z",
-    "deletions": 7,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44489",
+    "created_at": "2026-03-06T08:42:12Z",
+    "deletions": 62,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44443/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44443",
+    "files_url": "https://github.com/huggingface/transformers/pull/44489/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44489",
     "labels": [],
     "merged": true,
-    "number": 44443,
-    "review_comments_count": 0,
+    "number": 44489,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "fix(tokenizer): Only strip Fast from class names in AutoTokenizer if used as a suffix",
-    "updated_at": "2026-03-09T15:03:49Z"
+    "title": "Centralize AI agent templates in `.ai`",
+    "updated_at": "2026-03-18T14:17:22Z"
   },
   {
-    "additions": 37,
-    "author": "NielsRogge",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR fixes the training of LW-DETR. It turned out that the model was not able to overfit a single batch. Hence I asked Codex to investigate this. It turns out there were 3 bugs: 1. A logits calibration gap, cause\u2026",
-    "changed_files": 3,
+    "additions": 482,
+    "author": "abhijeet-dhumal",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #44486 Adds `KubeflowCallback` to enable automatic progress and metrics reporting for training jobs running on [Kubeflow Trainer](https://github.com/kubeflow/trainer). When training runs inside a Kubeflow Trai\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44441",
-    "created_at": "2026-03-04T19:54:20Z",
-    "deletions": 48,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44487",
+    "created_at": "2026-03-06T08:31:30Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44441/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44441",
+    "files_url": "https://github.com/huggingface/transformers/pull/44487/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44487",
     "labels": [],
     "merged": true,
-    "number": 44441,
-    "review_comments_count": 4,
+    "number": 44487,
+    "review_comments_count": 8,
     "state": "closed",
-    "title": "[LW-DETR] Fix training",
-    "updated_at": "2026-03-09T15:36:02Z"
+    "title": "feat(integration): Add KubeflowCallback to enable automatic progress \u2026",
+    "updated_at": "2026-03-18T14:58:23Z"
   },
   {
-    "additions": 11,
-    "author": "weiguangli-io",
+    "additions": 691,
+    "author": "kaixuanliu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes the noisy `HfHubHTTPError` exception output that appears when loading a transformer model from a repository that has discussions disabled. ### Root cause The `previous_pr()` function in `safetensors_conversio\u2026",
-    "changed_files": 2,
+    "body_excerpt": "@IlyasMoutawwakil pls help review, thx!",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44440",
-    "created_at": "2026-03-04T18:31:13Z",
-    "deletions": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44482",
+    "created_at": "2026-03-06T02:39:41Z",
+    "deletions": 332,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44440/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44440",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44440,
+    "files_url": "https://github.com/huggingface/transformers/pull/44482/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44482",
+    "labels": [],
+    "merged": true,
+    "number": 44482,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: catch HfHubHTTPError in safetensors auto_conversion thread",
-    "updated_at": "2026-03-05T15:39:11Z"
+    "title": "add XPU Expectations for higgs_audio_v2 tests",
+    "updated_at": "2026-04-09T02:32:48Z"
   },
   {
-    "additions": 5,
-    "author": "Sai-Suraj-27",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes this failing [ProphetNetModelIntegrationTest](https://github.com/huggingface/transformers/actions/runs/22606636929/job/65500453490#step:14:2331). <img width=\"2303\" height=\"165\" alt=\"image\" src=\"https://github.\u2026",
-    "changed_files": 1,
+    "additions": 2353,
+    "author": "XingyuHu109",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary This PR adds native Transformers support for DeepSeek-V3.2. It introduces a new `deepseek_v32` model family so the official checkpoints resolve through the standard auto classes without `trust_remote_code`. The implementation ke\u2026",
+    "changed_files": 19,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44439",
-    "created_at": "2026-03-04T16:55:50Z",
-    "deletions": 5,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44481",
+    "created_at": "2026-03-05T21:14:38Z",
+    "deletions": 30,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44439/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44439",
+    "files_url": "https://github.com/huggingface/transformers/pull/44481/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44481",
     "labels": [],
-    "merged": true,
-    "number": 44439,
-    "review_comments_count": 0,
+    "merged": false,
+    "number": 44481,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Fix failing `ProphetNetModelIntegrationTest`",
-    "updated_at": "2026-03-05T15:43:59Z"
+    "title": "Add native DeepSeek-V3.2 support",
+    "updated_at": "2026-04-14T21:34:03Z"
   },
   {
-    "additions": 135,
-    "author": "SunMarc",
+    "additions": 3,
+    "author": "ydshieh",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds flashoptim from databricks team into Trainer ! cc @tomaarsen ### Results ``` Optimizer Loss Time Speed Memory \u0394 Mem \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 AdamW Fused 1.4\u2026",
-    "changed_files": 6,
+    "body_excerpt": "# What does this PR do? add `diffusers` to docker file for `VibeVoice` (added in PR #40546).",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44438",
-    "created_at": "2026-03-04T16:31:35Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44480",
+    "created_at": "2026-03-05T20:54:07Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44438/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44438",
+    "files_url": "https://github.com/huggingface/transformers/pull/44480/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44480",
     "labels": [],
-    "merged": false,
-    "number": 44438,
+    "merged": true,
+    "number": 44480,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add flashoptim",
-    "updated_at": "2026-04-02T13:06:17Z"
+    "state": "closed",
+    "title": "Add `diffusers` to CI docker file",
+    "updated_at": "2026-03-05T21:11:17Z"
   },
   {
-    "additions": 150,
-    "author": "ArthurZucker",
+    "additions": 116,
+    "author": "BenjaminBossan",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 5,
+    "body_excerpt": "Required fixes: - some code was using unordered data structures, making weight order random - adjust alpha to offset increased rank from fusion - import functions from PEFT if available See https://github.com/huggingface/peft/pull/3083.",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44437",
-    "created_at": "2026-03-04T15:34:34Z",
-    "deletions": 134,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44478",
+    "created_at": "2026-03-05T17:19:31Z",
+    "deletions": 26,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44437/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44437",
+    "files_url": "https://github.com/huggingface/transformers/pull/44478/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44478",
     "labels": [],
     "merged": true,
-    "number": 44437,
-    "review_comments_count": 5,
+    "number": 44478,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "protect imports",
-    "updated_at": "2026-03-04T16:19:49Z"
+    "title": "[WIP] FIX Make Mixtral LoRA loading work",
+    "updated_at": "2026-03-11T17:44:20Z"
   },
   {
-    "additions": 8,
-    "author": "jw9603",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes #44423 `continuous_batching_chat_completion` was missing input preprocessing and `tokenize=True` in `apply_chat_template`, causing `'str' object has no attribute 'to'` for multimodal models. Added the same `get_model_modality` + `get\u2026",
+    "additions": 1,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title. It's quite a random rule to fix https://huggingface.co/fixie-ai/ultravox-v0_5-llama-3_2-1b to be honest",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44436",
-    "created_at": "2026-03-04T15:26:48Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44477",
+    "created_at": "2026-03-05T16:58:29Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44436/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44436",
+    "files_url": "https://github.com/huggingface/transformers/pull/44477/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44477",
     "labels": [],
-    "merged": true,
-    "number": 44436,
-    "review_comments_count": 4,
+    "merged": false,
+    "number": 44477,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix continuous batching for multimodal models",
-    "updated_at": "2026-03-09T13:58:37Z"
+    "title": "[vllm compat] Fix remote code inits",
+    "updated_at": "2026-03-11T10:34:06Z"
   },
   {
-    "additions": 138,
-    "author": "remi-or",
+    "additions": 4,
+    "author": "Rocketknight1",
     "author_association": "MEMBER",
-    "body_excerpt": "This PR adds the option to have a ContinuousBatchingManager not be destroyed after generation is over. This allows the user to re-use the manager without requiring him to know any other entry point for CB apart from `generate_batch` or the\u2026",
-    "changed_files": 6,
+    "body_excerpt": "I made an oversight in the fix at #43981 - I didn't realize the dim order changed for torch, so the test was still flaky for torch tensors. The fix reduced the flaky frequency a lot so I thought it had been fixed, but actually it's still t\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44435",
-    "created_at": "2026-03-04T14:17:08Z",
-    "deletions": 54,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44476",
+    "created_at": "2026-03-05T16:39:44Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44435/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44435",
+    "files_url": "https://github.com/huggingface/transformers/pull/44476/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44476",
     "labels": [],
     "merged": true,
-    "number": 44435,
-    "review_comments_count": 2,
+    "number": 44476,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[CB] Persistent manager",
-    "updated_at": "2026-03-26T22:02:28Z"
+    "title": "Fix Llava tests for torch too!",
+    "updated_at": "2026-03-11T16:47:05Z"
   },
   {
-    "additions": 413,
-    "author": "remi-or",
+    "additions": 1,
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": "This PR adds a dedicated config for continuous batching, which is starting to have a lot parameters. This will give the user a clear view of what is possible and make adding new parameters easier. No breaking changes through `account_for_c\u2026",
-    "changed_files": 9,
+    "body_excerpt": "chameleon added to MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44434",
-    "created_at": "2026-03-04T13:49:05Z",
-    "deletions": 303,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44475",
+    "created_at": "2026-03-05T16:29:18Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44434/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44434",
+    "files_url": "https://github.com/huggingface/transformers/pull/44475/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44475",
     "labels": [],
     "merged": true,
-    "number": 44434,
-    "review_comments_count": 12,
+    "number": 44475,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[CB] Add dedicated config",
-    "updated_at": "2026-03-13T13:56:40Z"
+    "title": "chameleon added to MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS",
+    "updated_at": "2026-03-09T22:33:20Z"
   },
   {
-    "additions": 177,
-    "author": "leopold-tzafon",
+    "additions": 875,
+    "author": "JJJYmmm",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Instead of silently failing when mm_token_type_ids is not passed, derives it in Qwen3 and Qwen3.5. Same as it was before: https://github.com/huggingface/transformers/commit/c281a2de8998e66e93fac30a236225528531df9b P\u2026",
-    "changed_files": 18,
+    "body_excerpt": "# What does this PR do? Fix https://github.com/QwenLM/Qwen3.5/issues/58. In the latest code, Qwen3VL and Qwen3.5 use the same `get_rope_index` func of Qwen2VL. But they should be different since Qwen3VL/Qwen3.5 introduce text timestamps. T\u2026",
+    "changed_files": 9,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44433",
-    "created_at": "2026-03-04T13:46:14Z",
-    "deletions": 61,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44474",
+    "created_at": "2026-03-05T15:46:09Z",
+    "deletions": 107,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44433/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44433",
+    "files_url": "https://github.com/huggingface/transformers/pull/44474/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44474",
     "labels": [],
     "merged": true,
-    "number": 44433,
-    "review_comments_count": 0,
+    "number": 44474,
+    "review_comments_count": 10,
     "state": "closed",
-    "title": "fix: raise error if mm_token_type_ids not supplied ",
-    "updated_at": "2026-03-12T17:12:47Z"
+    "title": "[Bugfix] fix video inference of qwen3vl and qwen3.5 series",
+    "updated_at": "2026-03-10T09:52:44Z"
   },
   {
-    "additions": 85,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per title, instead of having to divide image token by norm scale, we can do it same way as in other model (eg. gemma3) and add a custom embed layer. It should be 100% BC because users usually call `self.embed_tok\u2026",
-    "changed_files": 8,
+    "additions": 137,
+    "author": "winglian",
+    "author_association": "COLLABORATOR",
+    "body_excerpt": "# What does this PR do? supersedes #44446 on `main`, when loading to cpu and using meta devices for non-rank0 processes, it now re-initializes weights on those processes as well as uses more CPU memory. In testing with loading llama3-8b. m\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44432",
-    "created_at": "2026-03-04T10:04:40Z",
-    "deletions": 38,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44473",
+    "created_at": "2026-03-05T14:52:15Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44432/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44432",
+    "files_url": "https://github.com/huggingface/transformers/pull/44473/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44473",
     "labels": [],
     "merged": true,
-    "number": 44432,
-    "review_comments_count": 0,
+    "number": 44473,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Make paligemma embed tokens standard",
-    "updated_at": "2026-03-11T08:38:41Z"
+    "title": "fix FSDP loading with meta devices",
+    "updated_at": "2026-03-09T15:46:22Z"
   },
   {
-    "additions": 4094,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Re-opening back a PR on cleaning up clip-like model's backbones. Let's merge it now, I've been seeing quite a lot of ppl reporting it and I am not sure when it will be resolved by the big vision refactor Basically,\u2026",
-    "changed_files": 42,
+    "additions": 13,
+    "author": "jblox26",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this fix? Running video inference with any `Qwen3VL` model raises `StopIteration` during `model.generate()`: ``` File \".../transformers/models/qwen3_vl/modeling_qwen3_vl.py\", line 1126, in get_rope_index grid_thw = next(grid_i\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 25,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44431",
-    "created_at": "2026-03-04T10:02:13Z",
-    "deletions": 2220,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44472",
+    "created_at": "2026-03-05T14:50:06Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44431/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44431",
+    "files_url": "https://github.com/huggingface/transformers/pull/44472/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44472",
     "labels": [],
-    "merged": true,
-    "number": 44431,
-    "review_comments_count": 92,
+    "merged": false,
+    "number": 44472,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor CLIP-like models",
-    "updated_at": "2026-04-10T13:40:31Z"
+    "title": "Fix Qwen3VL get_rope_index StopIteration with per-frame video tokens",
+    "updated_at": "2026-03-06T15:15:58Z"
   },
   {
-    "additions": 0,
-    "author": "Rohang2005",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? This PR fixes an inconsistency in the AFMoE module where `past_key_values` was passed to a function argument expecting `past_key_value`. The function signature expects a singular cache object (`past_key_value`), bu\u2026",
-    "changed_files": 0,
+    "additions": 50,
+    "author": "weiguangli-io",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #44466 After `.to(device)`, PyTorch's `Module._apply` may create new `Parameter` objects that no longer share storage with tied weights. This caused `remove_tied_weights_from_state_dict` to fail to detect and\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44430",
-    "created_at": "2026-03-04T08:13:38Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44471",
+    "created_at": "2026-03-05T14:30:17Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44430/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44430",
+    "files_url": "https://github.com/huggingface/transformers/pull/44471/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44471",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44430,
+    "number": 44471,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix inconsistent past_key_value/past_key_values usage in AFMoE modeling",
-    "updated_at": "2026-03-04T14:07:32Z"
+    "title": "Fix tied weights serialization being device-dependent",
+    "updated_at": "2026-03-06T14:03:18Z"
   },
   {
-    "additions": 14,
-    "author": "thakoreh",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44336 The `loading_report` module was using `PALETTE['italic']` and `PALETTE['bold']` directly in string formatting, which caused ANSI escape codes to be emitted even when stdout is not connected to a terminal (e.g., when\u2026",
-    "changed_files": 1,
+    "additions": 8,
+    "author": "weiguangli-io",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fixes #44360 The reference `fp8_index` kernel clamps per-head q\u00b7k scores with `T.max(logits, 0)` before the weighted sum across heads ([kernel.py#L241](https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp/blob/main/inference/kernel.py#L241\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44429",
-    "created_at": "2026-03-04T07:47:02Z",
-    "deletions": 6,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44470",
+    "created_at": "2026-03-05T14:02:05Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44429/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44429",
+    "files_url": "https://github.com/huggingface/transformers/pull/44470/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44470",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44429,
+    "number": 44470,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix ANSI codes emitted in loading_report when stdout is not a TTY",
-    "updated_at": "2026-03-04T13:58:46Z"
+    "title": "Add missing ReLU in GlmMoeDsaIndexer",
+    "updated_at": "2026-03-05T15:39:38Z"
   },
   {
-    "additions": 10,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@IlyasMoutawwakil pls help review, thx!",
-    "changed_files": 2,
+    "additions": 4,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? For remote code that behave correctly with tied weights, we need to keep the same behavior as for the main lib, i.e. not remove them from tied weights (as tied weights are marked as missing to avoid inits!!)",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44428",
-    "created_at": "2026-03-04T07:41:20Z",
-    "deletions": 3,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44469",
+    "created_at": "2026-03-05T13:51:55Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44428/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44428",
+    "files_url": "https://github.com/huggingface/transformers/pull/44469/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44469",
     "labels": [],
     "merged": true,
-    "number": 44428,
-    "review_comments_count": 1,
+    "number": 44469,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add XPU Expectations for vibe voice acoustic tokenizer tests",
-    "updated_at": "2026-04-02T03:21:38Z"
+    "title": "[remote code/vllm] Fix incorrect tied weights",
+    "updated_at": "2026-03-05T15:07:56Z"
   },
   {
-    "additions": 43,
-    "author": "Jaredw2289-svg",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #44297 ## Problem `tokenizer.save_pretrained()` overwrites `tokenizer_class` in `tokenizer_config.json` with the current wrapper class (e.g. `PreTrainedTokenizerFast`) instead of preserving the original class from the loaded config (\u2026",
-    "changed_files": 2,
+    "additions": 13,
+    "author": "itazap",
+    "author_association": "MEMBER",
+    "body_excerpt": "Replace placeholder tokens as specified in added_tokens_decoder if we have added_tokens_decoder with specific token_ids, we need to overwrite them in spm model ! example: [UNUSED_TOKEN_146] -> <|im_start|> see internlm2: https://huggingfac\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44427",
-    "created_at": "2026-03-04T06:03:56Z",
-    "deletions": 6,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44468",
+    "created_at": "2026-03-05T13:48:56Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44427/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44427",
+    "files_url": "https://github.com/huggingface/transformers/pull/44468/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44468",
     "labels": [],
-    "merged": false,
-    "number": 44427,
+    "merged": true,
+    "number": 44468,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(tokenization): preserve original tokenizer_class in save_pretrained",
-    "updated_at": "2026-03-11T02:59:12Z"
+    "title": "Replace placeholder tokens as specified in added_tokens_decoder",
+    "updated_at": "2026-03-05T16:29:13Z"
   },
   {
-    "additions": 29,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@IlyasMoutawwakil Can you help review? Thx!",
-    "changed_files": 1,
+    "additions": 346,
+    "author": "itazap",
+    "author_association": "MEMBER",
+    "body_excerpt": "Replace placeholder tokens as specified in added_tokens_decoder if we have `added_tokens_decoder` with specific token_ids, we need to overwrite them in spm model ! `example: [UNUSED_TOKEN_146] -> <|im_start|>` see internlm2: https://huggin\u2026",
+    "changed_files": 24,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44426",
-    "created_at": "2026-03-04T05:57:34Z",
-    "deletions": 10,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44467",
+    "created_at": "2026-03-05T13:44:54Z",
+    "deletions": 204,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44426/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44426",
+    "files_url": "https://github.com/huggingface/transformers/pull/44467/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44467",
     "labels": [],
-    "merged": true,
-    "number": 44426,
-    "review_comments_count": 2,
-    "state": "closed",
-    "title": "update the expected output for qwen2_5_vl w/ pytorch 2.10 XPU",
-    "updated_at": "2026-04-09T02:32:41Z"
+    "merged": false,
+    "number": 44467,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Placeholder tokens update",
+    "updated_at": "2026-03-05T13:47:28Z"
   },
   {
-    "additions": 1,
-    "author": "qgallouedec",
+    "additions": 20,
+    "author": "kashif",
     "author_association": "MEMBER",
-    "body_excerpt": "I believe the second `if` should be `elif` so the else branch only triggers when neither the string-truncation NOR the float-formatting conditions apply. Otherwise it overwrites the truncation message with the original long string.",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Fix the loss calculation; we should calculate it on scaled targets. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44425",
-    "created_at": "2026-03-04T02:48:00Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44465",
+    "created_at": "2026-03-05T12:59:23Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44425/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44425",
-    "labels": [],
-    "merged": false,
-    "number": 44425,
+    "files_url": "https://github.com/huggingface/transformers/pull/44465/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44465",
+    "labels": [
+      "bug"
+    ],
+    "merged": true,
+    "number": 44465,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Fix conditional check for float formatting",
-    "updated_at": "2026-03-04T02:48:41Z"
+    "state": "closed",
+    "title": "[timesfm2_5] fix loss scaling",
+    "updated_at": "2026-03-05T14:50:26Z"
   },
   {
-    "additions": 6,
-    "author": "jw9603",
+    "additions": 16,
+    "author": "weiguangli-io",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes `AttributeError: 'str' object has no attribute 'to'` when using `transformers serve --continuous-batching` with multimodal models like Qwen3.5-9B. `processor.apply_chat_template()` returns a plain string (not\u2026",
+    "body_excerpt": "## What does this PR do? Fixes #44462 When a model's `model_type` (e.g. `\"llama\"`) has no entry in `TOKENIZER_MAPPING_NAMES`, `AutoTokenizer.from_pretrained` falls through to loading the tokenizer class declared in `tokenizer_config.json`\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44424",
-    "created_at": "2026-03-04T00:56:08Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44463",
+    "created_at": "2026-03-05T12:45:57Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44424/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44424",
+    "files_url": "https://github.com/huggingface/transformers/pull/44463/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44463",
     "labels": [],
     "merged": false,
-    "number": 44424,
+    "number": 44463,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix `transformers serve --continuous-batching` for multimodal models",
-    "updated_at": "2026-03-05T09:16:25Z"
+    "title": "Fix AutoTokenizer ignoring tokenizer.json for unregistered model types",
+    "updated_at": "2026-03-07T13:50:44Z"
   },
   {
-    "additions": 117,
-    "author": "mitre88",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Adds a Spanish (es) translation of the `conversations.md` guide, which covers the fundamentals of using chat models in Transformers. ### Translated sections: - Chat CLI usage - TextGenerationPipeline in chat mode -\u2026",
-    "changed_files": 2,
+    "additions": 12,
+    "author": "kashif",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? for SP loss we do not have torch device mesh but rather a deepspeed only. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44422",
-    "created_at": "2026-03-04T00:42:43Z",
-    "deletions": 0,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44461",
+    "created_at": "2026-03-05T11:39:02Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44422/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44422",
+    "files_url": "https://github.com/huggingface/transformers/pull/44461/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44461",
     "labels": [],
     "merged": true,
-    "number": 44422,
-    "review_comments_count": 0,
+    "number": 44461,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "docs: add Spanish translation for conversations.md (chat basics)",
-    "updated_at": "2026-03-04T16:45:24Z"
+    "title": "[Trainer] fix SP loss",
+    "updated_at": "2026-03-05T13:00:40Z"
   },
   {
-    "additions": 309,
-    "author": "michaelbenayoun",
+    "additions": 1,
+    "author": "3outeille",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? When we shard weights according to a TP plan, we do not update the corresponding parent module attributes. For instance if we shard the weight of a `torch.nn.Linear`, we should also update its `in_features` or `out_\u2026",
-    "changed_files": 4,
+    "body_excerpt": null,
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44421",
-    "created_at": "2026-03-03T22:51:47Z",
-    "deletions": 5,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44460",
+    "created_at": "2026-03-05T10:53:07Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44421/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44421",
+    "files_url": "https://github.com/huggingface/transformers/pull/44460/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44460",
     "labels": [],
     "merged": true,
-    "number": 44421,
+    "number": 44460,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Update parent module attributes when sharding with TP",
-    "updated_at": "2026-03-05T23:32:06Z"
+    "title": "trigger tensor parallel utils test in the CI",
+    "updated_at": "2026-03-05T11:25:51Z"
   },
   {
-    "additions": 249,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "- removes \"Number of accelerators\" section from \"Accelerator selection\" guide since this is probably pretty commonly known - add a new \"DDP\" guide - refactored \"Accelerate\" guide with a more focused overview of what it is and how to config\u2026",
-    "changed_files": 5,
+    "additions": 95,
+    "author": "weiguangli-io",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes #44458 PR #42848 introduced a regression where `torch.compile` on `MllamaForConditionalGeneration` fails with a C++ compile error from the torch inductor backend (`'tmp2' was not declared in this scope`). The root cause is\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44420",
-    "created_at": "2026-03-03T22:41:59Z",
-    "deletions": 250,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44459",
+    "created_at": "2026-03-05T07:58:28Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44420/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44420",
+    "files_url": "https://github.com/huggingface/transformers/pull/44459/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44459",
     "labels": [],
     "merged": false,
-    "number": 44420,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "[docs] distributed training",
-    "updated_at": "2026-03-11T17:36:12Z"
+    "number": 44459,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "fix: make Mllama cross attention mask compatible with torch.compile",
+    "updated_at": "2026-03-07T13:50:40Z"
   },
   {
-    "additions": 6,
-    "author": "michaelbenayoun",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? To be merged after #44302 and https://github.com/huggingface/kernels/pull/285. It adds the `neuron` device in checks for custom kernels, enabling to load kernels for Neuron devices.",
+    "additions": 1,
+    "author": "Sai-Suraj-27",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes this failing [DepthProModelIntegrationTest](https://github.com/huggingface/transformers/actions/runs/22606636929/job/65500453624#step:14:4893). <img width=\"2231\" height=\"99\" alt=\"image\" src=\"https://github.com\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44417",
-    "created_at": "2026-03-03T20:15:26Z",
-    "deletions": 6,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44456",
+    "created_at": "2026-03-05T06:01:06Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44417/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44417",
+    "files_url": "https://github.com/huggingface/transformers/pull/44456/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44456",
     "labels": [],
     "merged": true,
-    "number": 44417,
+    "number": 44456,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Neuron kernels integration",
-    "updated_at": "2026-03-05T17:09:39Z"
+    "title": "Fix failing `DepthProModelIntegrationTest`",
+    "updated_at": "2026-03-05T14:52:40Z"
   },
   {
-    "additions": 1,
-    "author": "tyler-romero",
+    "additions": 3,
+    "author": "jiqing-feng",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Register `olmo_hybrid` in `TOKENIZER_MAPPING_NAMES` so auto-tokenizer resolution works, matching the other auto-registrations already in place for this model.",
+    "body_excerpt": "Check if accelerator exists before using `pin_memory`. reproduce it on a CPU only node: `python examples/pytorch/continuous_batching_simple.py` output: ``` File \"/home/jiqingfe/transformers/src/transformers/generation/continuous_batching/i\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44416",
-    "created_at": "2026-03-03T19:30:56Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44455",
+    "created_at": "2026-03-05T05:20:13Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44416/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44416",
+    "files_url": "https://github.com/huggingface/transformers/pull/44455/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44455",
     "labels": [],
     "merged": true,
-    "number": 44416,
-    "review_comments_count": 2,
+    "number": 44455,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "[tiny] Add olmo_hybrid to tokenizer auto-mapping",
-    "updated_at": "2026-03-04T19:26:10Z"
+    "title": "fix pin_memory for contiguous batching",
+    "updated_at": "2026-03-09T13:49:30Z"
   },
   {
-    "additions": 2,
-    "author": "SunMarc",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR removes @MekkCyber from the PR template. cc @Rocketknight1 you only need to ping me now ;)",
-    "changed_files": 1,
+    "additions": 17,
+    "author": "weiguangli-io",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary - Fixes the `_init_weights` method in `PegasusPreTrainedModel` and `MarianPreTrainedModel` to handle sinusoidal position embeddings before calling `super()._init_weights(module)`, preventing the generic `nn.Embedding` branch fro\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44415",
-    "created_at": "2026-03-03T16:59:08Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44454",
+    "created_at": "2026-03-05T03:51:38Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44415/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44415",
+    "files_url": "https://github.com/huggingface/transformers/pull/44454/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44454",
     "labels": [],
-    "merged": true,
-    "number": 44415,
+    "merged": false,
+    "number": 44454,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Update PR template",
-    "updated_at": "2026-03-04T14:13:04Z"
+    "title": "Fix Pegasus sinusoidal position embedding init regression in v5",
+    "updated_at": "2026-03-09T02:17:41Z"
   },
   {
-    "additions": 35,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/44303 - see also comments here https://github.com/huggingface/transformers/pull/44316#issuecomment-3984362089. Supersedes https://github.com/huggingface/trans\u2026",
+    "additions": 1,
+    "author": "weiguangli-io",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Fix KeyError in `convert_to_native_format` for dict vocab Fixes #44451 ### Problem `AutoTokenizer.from_pretrained(\"vesteinn/ScandiBERT\")` raises `KeyError: 0` in `convert_to_native_format`. ScandiBERT's `tokenizer_config.json` specifies\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44414",
-    "created_at": "2026-03-03T16:47:47Z",
-    "deletions": 39,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44452",
+    "created_at": "2026-03-05T03:34:02Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44414/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44414",
+    "files_url": "https://github.com/huggingface/transformers/pull/44452/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44452",
     "labels": [],
     "merged": true,
-    "number": 44414,
+    "number": 44452,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Reduce tqdm verbosity during model loading",
-    "updated_at": "2026-03-03T16:57:56Z"
+    "title": "Fix KeyError in convert_to_native_format for dict vocab",
+    "updated_at": "2026-03-19T13:59:23Z"
   },
   {
-    "additions": 4,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title.",
-    "changed_files": 1,
+    "additions": 297,
+    "author": "sandesh-bhandari-dev",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 32,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44413",
-    "created_at": "2026-03-03T16:24:43Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44449",
+    "created_at": "2026-03-05T01:40:47Z",
+    "deletions": 319,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44413/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44413",
-    "labels": [],
-    "merged": true,
-    "number": 44413,
+    "files_url": "https://github.com/huggingface/transformers/pull/44449/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44449",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44449,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix peft conversion mappings",
-    "updated_at": "2026-03-03T17:08:39Z"
+    "title": "fix: 3 bugs : MoE aux loss, ANSI TTY leak, pipeline removed and also task error",
+    "updated_at": "2026-03-05T13:22:40Z"
   },
   {
-    "additions": 138,
-    "author": "tarekziade",
+    "additions": 8,
+    "author": "yonigozlan",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Extends type checking to `src/transformers/quantizers`",
-    "changed_files": 28,
+    "body_excerpt": "# What does this PR do? Fixes MiniCPM-o-2_6 related tests failures in vLLM, and improve backward compatibility with remote code in general. Cc @hmellor @zucchini-nlp",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 25,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44412",
-    "created_at": "2026-03-03T14:53:31Z",
-    "deletions": 74,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44447",
+    "created_at": "2026-03-04T21:55:16Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44412/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44412",
+    "files_url": "https://github.com/huggingface/transformers/pull/44447/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44447",
     "labels": [],
     "merged": true,
-    "number": 44412,
-    "review_comments_count": 33,
+    "number": 44447,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "chore(typing): Add type checking to `src/transformers/quantizers`",
-    "updated_at": "2026-03-11T11:24:11Z"
+    "title": "[vLLM] Fix backward compatibility with hardcoded subprocessors classes in processors",
+    "updated_at": "2026-03-05T16:07:28Z"
   },
   {
-    "additions": 59,
-    "author": "burtenshaw",
+    "additions": 4,
+    "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR updates `AGENTS.md` to discourage duplicated and trivial work by agents. - CLAUDE.md-> AGENTS.md - ssue-thread coordination before PRs - mandatory duplicate-PR checks with gh commands - no one-off busywork P\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? This PR fixes fsdp loading for rank!=0 as they should stay on meta device. This was reverted in a PR that I can find anymore.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44411",
-    "created_at": "2026-03-03T11:35:13Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44446",
+    "created_at": "2026-03-04T21:38:00Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44411/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44411",
+    "files_url": "https://github.com/huggingface/transformers/pull/44446/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44446",
     "labels": [],
-    "merged": true,
-    "number": 44411,
-    "review_comments_count": 4,
+    "merged": false,
+    "number": 44446,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Update agentic contributions guidelines in AGENTS.md to force yielding.",
-    "updated_at": "2026-03-12T09:28:45Z"
+    "title": "Fix fdsp loading meta device",
+    "updated_at": "2026-03-05T15:07:57Z"
   },
   {
-    "additions": 38,
-    "author": "zvik",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "# What does this PR do? This PR allows the Granite-speech model to use hidden states from the encoder hidden layers. This is an internal model option that is required for the next generation of Granite-speech models. ## Changes: - New conf\u2026",
-    "changed_files": 2,
+    "additions": 2282,
+    "author": "gabe-l-hart",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds support for the forthcoming Granite Docling model based on the Granite 4 LLM architecture (`GraniteMoeHybrid`). ## Draft Status This PR is in draft pending the possibility of some additional changes: -\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44408",
-    "created_at": "2026-03-03T07:50:39Z",
-    "deletions": 4,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44408/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44408",
-    "labels": [],
-    "merged": false,
-    "number": 44408,
-    "review_comments_count": 9,
-    "state": "open",
-    "title": "Add option to export  encoder hidden states for Granite-speech",
-    "updated_at": "2026-04-14T13:48:55Z"
-  },
-  {
-    "additions": 23,
-    "author": "hongping-zh",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Adds an \"Energy Efficiency Considerations\" section to the bitsandbytes quantization documentation, providing practical guidance on the energy implications of different quantization configurations. ## Motivation This addresses th\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44407",
-    "created_at": "2026-03-03T04:42:57Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44445",
+    "created_at": "2026-03-04T20:54:17Z",
     "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44407/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44407",
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44445/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44445",
     "labels": [],
     "merged": false,
-    "number": 44407,
-    "review_comments_count": 0,
+    "number": 44445,
+    "review_comments_count": 44,
     "state": "open",
-    "title": "docs: add energy efficiency considerations to bitsandbytes quantization guide",
-    "updated_at": "2026-03-25T11:53:49Z"
+    "title": "Adding support for GraniteDoclingHybrid",
+    "updated_at": "2026-03-11T19:19:44Z"
   },
   {
-    "additions": 21,
-    "author": "medhakimbedhief",
+    "additions": 7,
+    "author": "harshaljanjani",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Adds sequence-classification support for Qwen3.5 in AutoModelForSequenceClassification. **What does this PR do?** This PR enables loading Qwen3.5 checkpoints with `AutoModelForSequenceClassification`, which previously failed with: `ValueEr\u2026",
-    "changed_files": 5,
+    "body_excerpt": "### What does this PR do? The following failing use case was identified and fixed in this PR: \u2192 [TOKENIZER_MAPPING_NAMES](https://github.com/huggingface/transformers/blob/main/src/transformers/models/auto/tokenization_auto.py#L63-L338) doe\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44406",
-    "created_at": "2026-03-03T03:44:37Z",
-    "deletions": 2,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44443",
+    "created_at": "2026-03-04T20:01:55Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44406/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44406",
+    "files_url": "https://github.com/huggingface/transformers/pull/44443/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44443",
     "labels": [],
     "merged": true,
-    "number": 44406,
+    "number": 44443,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add Qwen3.5 support for sequence classification",
-    "updated_at": "2026-04-08T11:33:19Z"
+    "title": "fix(tokenizer): Only strip Fast from class names in AutoTokenizer if used as a suffix",
+    "updated_at": "2026-03-09T15:03:49Z"
   },
   {
-    "additions": 0,
-    "author": "Rocketknight1",
+    "additions": 37,
+    "author": "NielsRogge",
     "author_association": "MEMBER",
-    "body_excerpt": "Some generate tests have a ~1% chance of generating short outputs because they hit an EOS token early, which causes the test to flake because it asserts the output shape. This PR enforces `min_length` so that doesn't happen!",
+    "body_excerpt": "# What does this PR do? This PR fixes the training of LW-DETR. It turned out that the model was not able to overfit a single batch. Hence I asked Codex to investigate this. It turns out there were 3 bugs: 1. A logits calibration gap, cause\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44401",
-    "created_at": "2026-03-02T18:26:56Z",
-    "deletions": 77,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44441",
+    "created_at": "2026-03-04T19:54:20Z",
+    "deletions": 48,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44401/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44401",
+    "files_url": "https://github.com/huggingface/transformers/pull/44441/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44441",
     "labels": [],
     "merged": true,
-    "number": 44401,
+    "number": 44441,
     "review_comments_count": 4,
     "state": "closed",
-    "title": "Enforce min length in some generate tests",
-    "updated_at": "2026-03-04T14:05:50Z"
+    "title": "[LW-DETR] Fix training",
+    "updated_at": "2026-03-09T15:36:02Z"
   },
   {
-    "additions": 14,
-    "author": "Kokonico",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? As per title, confirms interactive terminal before adding formatting to loading_report output. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear\u2026",
-    "changed_files": 1,
+    "additions": 11,
+    "author": "weiguangli-io",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes the noisy `HfHubHTTPError` exception output that appears when loading a transformer model from a repository that has discussions disabled. ### Root cause The `previous_pr()` function in `safetensors_conversio\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44400",
-    "created_at": "2026-03-02T18:21:48Z",
-    "deletions": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44440",
+    "created_at": "2026-03-04T18:31:13Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44400/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44400",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44440/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44440",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44400,
+    "number": 44440,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: interactive terminal checks for formatting in loading_report.py",
-    "updated_at": "2026-03-09T14:46:29Z"
+    "title": "fix: catch HfHubHTTPError in safetensors auto_conversion thread",
+    "updated_at": "2026-03-05T15:39:11Z"
   },
   {
-    "additions": 3,
-    "author": "weiguangli-io",
+    "additions": 5,
+    "author": "Sai-Suraj-27",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #44384 In `Qwen3_5TextModel.forward`, after splitting `position_ids` into `text_position_ids` (index 0, for text) and `position_ids` (indices 1:, for temporal/height/width), the decoder layer call incorrectly passed `posit\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Fixes this failing [ProphetNetModelIntegrationTest](https://github.com/huggingface/transformers/actions/runs/22606636929/job/65500453490#step:14:2331). <img width=\"2303\" height=\"165\" alt=\"image\" src=\"https://github.\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44399",
-    "created_at": "2026-03-02T17:28:59Z",
-    "deletions": 3,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44439",
+    "created_at": "2026-03-04T16:55:50Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44399/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44399",
+    "files_url": "https://github.com/huggingface/transformers/pull/44439/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44439",
     "labels": [],
     "merged": true,
-    "number": 44399,
+    "number": 44439,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix position_ids typo in Qwen3_5TextModel forward pass",
-    "updated_at": "2026-03-06T01:48:22Z"
+    "title": "Fix failing `ProphetNetModelIntegrationTest`",
+    "updated_at": "2026-03-05T15:43:59Z"
   },
   {
-    "additions": 4,
-    "author": "weiguangli-io",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Adds the missing ReLU activation in `GlmMoeDsaIndexer.forward()` on per-head q\u00b7k scores before the weighted sum across heads. The reference DeepSeek-V3.2 `fp8_index` kernel applies `T.max(logits, 0)` (i.e., ReLU) a\u2026",
-    "changed_files": 2,
+    "additions": 135,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR adds flashoptim from databricks team into Trainer ! cc @tomaarsen ### Results ``` Optimizer Loss Time Speed Memory \u0394 Mem \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 AdamW Fused 1.4\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44398",
-    "created_at": "2026-03-02T16:41:01Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44438",
+    "created_at": "2026-03-04T16:31:35Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44398/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44398",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44438/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44438",
+    "labels": [],
     "merged": false,
-    "number": 44398,
+    "number": 44438,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add missing ReLU in GlmMoeDsaIndexer scoring",
-    "updated_at": "2026-03-04T13:54:20Z"
+    "state": "open",
+    "title": "Add flashoptim",
+    "updated_at": "2026-04-02T13:06:17Z"
   },
   {
-    "additions": 86,
-    "author": "weiguangli-io",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes #44380 `GPT2Attention.forward()` did not pass the `scaling` parameter to `attention_interface`, causing `scale_attn_weights` and `scale_attn_by_inverse_layer_idx` config options to be silently ignored when usi\u2026",
-    "changed_files": 3,
+    "additions": 150,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44397",
-    "created_at": "2026-03-02T16:14:37Z",
-    "deletions": 50,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44437",
+    "created_at": "2026-03-04T15:34:34Z",
+    "deletions": 134,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44397/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44397",
+    "files_url": "https://github.com/huggingface/transformers/pull/44437/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44437",
     "labels": [],
     "merged": true,
-    "number": 44397,
-    "review_comments_count": 17,
+    "number": 44437,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "Fix GPT2 attention scaling ignored in SDPA/FlashAttention",
-    "updated_at": "2026-03-04T16:47:42Z"
+    "title": "protect imports",
+    "updated_at": "2026-03-04T16:19:49Z"
   },
   {
-    "additions": 3,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Some renaming should just never be applied when the weight format already matches. (this is actually regardless of remote code). This allows us to remove 1 test added in https://github.com/huggingface/transformers/c\u2026",
-    "changed_files": 2,
+    "additions": 8,
+    "author": "jw9603",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fixes #44423 `continuous_batching_chat_completion` was missing input preprocessing and `tokenize=True` in `apply_chat_template`, causing `'str' object has no attribute 'to'` for multimodal models. Added the same `get_model_modality` + `get\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44396",
-    "created_at": "2026-03-02T15:50:27Z",
-    "deletions": 16,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44436",
+    "created_at": "2026-03-04T15:26:48Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44396/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44396",
+    "files_url": "https://github.com/huggingface/transformers/pull/44436/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44436",
     "labels": [],
     "merged": true,
-    "number": 44396,
-    "review_comments_count": 2,
+    "number": 44436,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "[ Dynamic weight loader] fix remote code when format matches",
-    "updated_at": "2026-03-03T17:53:39Z"
+    "title": "Fix continuous batching for multimodal models",
+    "updated_at": "2026-03-09T13:58:37Z"
   },
   {
-    "additions": 153,
-    "author": "Cyrilvallez",
+    "additions": 138,
+    "author": "remi-or",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title!",
+    "body_excerpt": "This PR adds the option to have a ContinuousBatchingManager not be destroyed after generation is over. This allows the user to re-use the manager without requiring him to know any other entry point for CB apart from `generate_batch` or the\u2026",
     "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44395",
-    "created_at": "2026-03-02T14:52:12Z",
-    "deletions": 31,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44435",
+    "created_at": "2026-03-04T14:17:08Z",
+    "deletions": 54,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44395/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44395",
+    "files_url": "https://github.com/huggingface/transformers/pull/44435/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44435",
     "labels": [],
     "merged": true,
-    "number": 44395,
-    "review_comments_count": 22,
+    "number": 44435,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix kernels security issue",
-    "updated_at": "2026-03-03T15:28:20Z"
+    "title": "[CB] Persistent manager",
+    "updated_at": "2026-03-26T22:02:28Z"
   },
   {
-    "additions": 4742,
-    "author": "eustlb",
+    "additions": 413,
+    "author": "remi-or",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? #43514 precedes this PR",
-    "changed_files": 61,
+    "body_excerpt": "This PR adds a dedicated config for continuous batching, which is starting to have a lot parameters. This will give the user a clear view of what is possible and make adding new parameters easier. No breaking changes through `account_for_c\u2026",
+    "changed_files": 9,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44394",
-    "created_at": "2026-03-02T14:49:05Z",
-    "deletions": 7232,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44394/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44394",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44434",
+    "created_at": "2026-03-04T13:49:05Z",
+    "deletions": 303,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44434/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44434",
     "labels": [],
-    "merged": false,
-    "number": 44394,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "\ud83d\udea8\ud83d\udea7 FeatureExtractor \u2192 AudioProcessor",
-    "updated_at": "2026-04-03T22:16:41Z"
+    "merged": true,
+    "number": 44434,
+    "review_comments_count": 12,
+    "state": "closed",
+    "title": "[CB] Add dedicated config",
+    "updated_at": "2026-03-13T13:56:40Z"
   },
   {
-    "additions": 61,
-    "author": "ebezzam",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The acoustic tokenizer was updated to use `VoxtralRealtimeConv1dPaddingCache` in #43625 but the ASR model wasn't updated.",
-    "changed_files": 2,
+    "additions": 177,
+    "author": "leopold-tzafon",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Instead of silently failing when mm_token_type_ids is not passed, derives it in Qwen3 and Qwen3.5. Same as it was before: https://github.com/huggingface/transformers/commit/c281a2de8998e66e93fac30a236225528531df9b P\u2026",
+    "changed_files": 18,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44392",
-    "created_at": "2026-03-02T13:41:19Z",
-    "deletions": 100,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44433",
+    "created_at": "2026-03-04T13:46:14Z",
+    "deletions": 61,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44392/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44392",
+    "files_url": "https://github.com/huggingface/transformers/pull/44433/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44433",
     "labels": [],
     "merged": true,
-    "number": 44392,
+    "number": 44433,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[VibeVoice ASR] Use updated padding cache for ASR model.",
-    "updated_at": "2026-03-02T13:51:02Z"
+    "title": "fix: raise error if mm_token_type_ids not supplied ",
+    "updated_at": "2026-03-12T17:12:47Z"
   },
   {
-    "additions": 3040,
-    "author": "ebezzam",
+    "additions": 85,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Tokenizer was updated to Voxtral cache object in #43625, but forgot to update that of the ASR model",
-    "changed_files": 30,
+    "body_excerpt": "# What does this PR do? As per title, instead of having to divide image token by norm scale, we can do it same way as in other model (eg. gemma3) and add a custom embed layer. It should be 100% BC because users usually call `self.embed_tok\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44391",
-    "created_at": "2026-03-02T13:29:59Z",
-    "deletions": 311,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44432",
+    "created_at": "2026-03-04T10:04:40Z",
+    "deletions": 38,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44391/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44391",
+    "files_url": "https://github.com/huggingface/transformers/pull/44432/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44432",
     "labels": [],
-    "merged": false,
-    "number": 44391,
+    "merged": true,
+    "number": 44432,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[VibeVoice ASR] Use newer cache object for modular",
-    "updated_at": "2026-03-02T13:34:23Z"
+    "title": "Make paligemma embed tokens standard",
+    "updated_at": "2026-03-11T08:38:41Z"
   },
   {
-    "additions": 3330,
-    "author": "liding-nv",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 13,
+    "additions": 4094,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Re-opening back a PR on cleaning up clip-like model's backbones. Let's merge it now, I've been seeing quite a lot of ppl reporting it and I am not sure when it will be resolved by the big vision refactor Basically,\u2026",
+    "changed_files": 42,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44390",
-    "created_at": "2026-03-02T13:22:21Z",
-    "deletions": 0,
+    "comments_count": 25,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44431",
+    "created_at": "2026-03-04T10:02:13Z",
+    "deletions": 2220,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44390/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44390",
+    "files_url": "https://github.com/huggingface/transformers/pull/44431/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44431",
     "labels": [],
     "merged": true,
-    "number": 44390,
-    "review_comments_count": 20,
+    "number": 44431,
+    "review_comments_count": 92,
     "state": "closed",
-    "title": "add support for nemotron_3",
-    "updated_at": "2026-03-03T18:18:50Z"
+    "title": "Refactor CLIP-like models",
+    "updated_at": "2026-04-10T13:40:31Z"
   },
   {
-    "additions": 5,
-    "author": "Abdennacer-Badaoui",
-    "author_association": "MEMBER",
-    "body_excerpt": "Adds explicit `timm` installation to the AMD ROCm Docker image. This causes ~200 test failures in AMD CI (e.g., [gemma3n vision tests](https://github.com/huggingface/transformers/actions/runs/22474359922/job/65104428291)). This mirrors wha\u2026",
-    "changed_files": 2,
+    "additions": 0,
+    "author": "Rohang2005",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? This PR fixes an inconsistency in the AFMoE module where `past_key_values` was passed to a function argument expecting `past_key_value`. The function signature expects a singular cache object (`past_key_value`), bu\u2026",
+    "changed_files": 0,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44389",
-    "created_at": "2026-03-02T13:09:55Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44430",
+    "created_at": "2026-03-04T08:13:38Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44389/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44389",
-    "labels": [],
-    "merged": true,
-    "number": 44389,
+    "files_url": "https://github.com/huggingface/transformers/pull/44430/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44430",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44430,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[AMD CI] Add missing timm dependency to ROCm Docker image",
-    "updated_at": "2026-03-03T12:00:19Z"
+    "title": "Fix inconsistent past_key_value/past_key_values usage in AFMoE modeling",
+    "updated_at": "2026-03-04T14:07:32Z"
   },
   {
-    "additions": 0,
-    "author": "sahilmaniyar888",
+    "additions": 14,
+    "author": "thakoreh",
     "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Fixes #44336 ### Summary This PR prevents ANSI style escape sequences from being emitted by `loading_report` when stdout is non-interactive (for example, redirected logs/files). ### Changes - Added a small helper `_\u2026",
-    "changed_files": 0,
+    "body_excerpt": "## Summary Fixes #44336 The `loading_report` module was using `PALETTE['italic']` and `PALETTE['bold']` directly in string formatting, which caused ANSI escape codes to be emitted even when stdout is not connected to a terminal (e.g., when\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44388",
-    "created_at": "2026-03-02T11:40:49Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44429",
+    "created_at": "2026-03-04T07:47:02Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44388/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44388",
+    "files_url": "https://github.com/huggingface/transformers/pull/44429/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44429",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44388,
+    "number": 44429,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix loading report ANSI styles for non-TTY output",
-    "updated_at": "2026-03-11T06:29:31Z"
+    "title": "Fix ANSI codes emitted in loading_report when stdout is not a TTY",
+    "updated_at": "2026-03-04T13:58:46Z"
   },
   {
-    "additions": 62,
-    "author": "eustlb",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per title",
+    "additions": 10,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@IlyasMoutawwakil pls help review, thx!",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44428",
+    "created_at": "2026-03-04T07:41:20Z",
+    "deletions": 3,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44428/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44428",
+    "labels": [],
+    "merged": true,
+    "number": 44428,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "Add XPU Expectations for vibe voice acoustic tokenizer tests",
+    "updated_at": "2026-04-02T03:21:38Z"
+  },
+  {
+    "additions": 43,
+    "author": "Jaredw2289-svg",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #44297 ## Problem `tokenizer.save_pretrained()` overwrites `tokenizer_class` in `tokenizer_config.json` with the current wrapper class (e.g. `PreTrainedTokenizerFast`) instead of preserving the original class from the loaded config (\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44386",
-    "created_at": "2026-03-02T10:43:13Z",
-    "deletions": 16,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44427",
+    "created_at": "2026-03-04T06:03:56Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44386/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44386",
+    "files_url": "https://github.com/huggingface/transformers/pull/44427/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44427",
+    "labels": [],
+    "merged": false,
+    "number": 44427,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "fix(tokenization): preserve original tokenizer_class in save_pretrained",
+    "updated_at": "2026-03-11T02:59:12Z"
+  },
+  {
+    "additions": 29,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@IlyasMoutawwakil Can you help review? Thx!",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44426",
+    "created_at": "2026-03-04T05:57:34Z",
+    "deletions": 10,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44426/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44426",
     "labels": [],
     "merged": true,
-    "number": 44386,
+    "number": 44426,
     "review_comments_count": 2,
     "state": "closed",
-    "title": "[higgs-audio-v2] fix sampling",
-    "updated_at": "2026-03-02T13:06:23Z"
+    "title": "update the expected output for qwen2_5_vl w/ pytorch 2.10 XPU",
+    "updated_at": "2026-04-09T02:32:41Z"
   },
   {
-    "additions": 8,
-    "author": "NielsRogge",
+    "additions": 1,
+    "author": "qgallouedec",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? I wasn't able to run make check-repo locally successfully, unless the following 3 fixes were applied.",
-    "changed_files": 3,
+    "body_excerpt": "I believe the second `if` should be `elif` so the else branch only triggers when neither the string-truncation NOR the float-formatting conditions apply. Otherwise it overwrites the truncation message with the original long string.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44385",
-    "created_at": "2026-03-02T09:45:15Z",
-    "deletions": 5,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44425",
+    "created_at": "2026-03-04T02:48:00Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44385/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44385",
+    "files_url": "https://github.com/huggingface/transformers/pull/44425/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44425",
     "labels": [],
     "merged": false,
-    "number": 44385,
+    "number": 44425,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Fix make check-repo",
-    "updated_at": "2026-03-02T09:54:23Z"
+    "title": "Fix conditional check for float formatting",
+    "updated_at": "2026-03-04T02:48:41Z"
   },
   {
-    "additions": 4,
-    "author": "JJJYmmm",
+    "additions": 6,
+    "author": "jw9603",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fix the attr `_no_split_modules` of `Qwen3_5Model` and `Qwen3_5MoeModel`, which affect the FSDP init of hf Trainer.",
-    "changed_files": 3,
+    "body_excerpt": "## What does this PR do? Fixes `AttributeError: 'str' object has no attribute 'to'` when using `transformers serve --continuous-batching` with multimodal models like Qwen3.5-9B. `processor.apply_chat_template()` returns a plain string (not\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44382",
-    "created_at": "2026-03-02T05:42:48Z",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44424",
+    "created_at": "2026-03-04T00:56:08Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44382/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44382",
+    "files_url": "https://github.com/huggingface/transformers/pull/44424/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44424",
     "labels": [],
-    "merged": true,
-    "number": 44382,
+    "merged": false,
+    "number": 44424,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Bugfix] fix qwen3.5 no split module",
-    "updated_at": "2026-03-02T16:17:22Z"
+    "title": "Fix `transformers serve --continuous-batching` for multimodal models",
+    "updated_at": "2026-03-05T09:16:25Z"
   },
   {
-    "additions": 2,
-    "author": "carcel-yu",
+    "additions": 117,
+    "author": "mitre88",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? This PR adds MLU support to `is_torch_bf16_gpu_available()` by checking `torch.mlu.is_bf16_supported()` when an MLU device is available. ### Why is this needed? MLU devices support bf16 training, but they are curr\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## What does this PR do? Adds a Spanish (es) translation of the `conversations.md` guide, which covers the fundamentals of using chat models in Transformers. ### Translated sections: - Chat CLI usage - TextGenerationPipeline in chat mode -\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44381",
-    "created_at": "2026-03-02T05:34:49Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44422",
+    "created_at": "2026-03-04T00:42:43Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44381/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44381",
+    "files_url": "https://github.com/huggingface/transformers/pull/44422/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44422",
     "labels": [],
     "merged": true,
-    "number": 44381,
+    "number": 44422,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add MLU bf16 support to is_torch_bf16_gpu_available",
-    "updated_at": "2026-03-06T14:34:30Z"
+    "title": "docs: add Spanish translation for conversations.md (chat basics)",
+    "updated_at": "2026-03-04T16:45:24Z"
   },
   {
-    "additions": 579,
-    "author": "remi-or",
+    "additions": 309,
+    "author": "michaelbenayoun",
     "author_association": "MEMBER",
-    "body_excerpt": "# Summary ! This PR is in draft, waiting for https://github.com/huggingface/transformers/pull/44227 to be merged This PR adds support for the `flash_attention_with_kvcache` kernel in continuoys batching. This is very efficient for decode-o\u2026",
-    "changed_files": 14,
+    "body_excerpt": "# What does this PR do? When we shard weights according to a TP plan, we do not update the corresponding parent module attributes. For instance if we shard the weight of a `torch.nn.Linear`, we should also update its `in_features` or `out_\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44379",
-    "created_at": "2026-03-01T23:13:17Z",
-    "deletions": 235,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44421",
+    "created_at": "2026-03-03T22:51:47Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44379/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44379",
+    "files_url": "https://github.com/huggingface/transformers/pull/44421/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44421",
     "labels": [],
     "merged": true,
-    "number": 44379,
-    "review_comments_count": 19,
+    "number": 44421,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[CB] Add paged_attention kernel",
-    "updated_at": "2026-03-09T22:16:31Z"
+    "title": "Update parent module attributes when sharding with TP",
+    "updated_at": "2026-03-05T23:32:06Z"
   },
   {
-    "additions": 1,
-    "author": "redpanda1995",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
+    "additions": 249,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "- removes \"Number of accelerators\" section from \"Accelerator selection\" guide since this is probably pretty commonly known - add a new \"DDP\" guide - refactored \"Accelerate\" guide with a more focused overview of what it is and how to config\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44378",
-    "created_at": "2026-03-01T22:57:50Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44420",
+    "created_at": "2026-03-03T22:41:59Z",
+    "deletions": 250,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44378/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44378",
+    "files_url": "https://github.com/huggingface/transformers/pull/44420/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44420",
     "labels": [],
     "merged": false,
-    "number": 44378,
+    "number": 44420,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix logging with each layer with ms-swift lora fine-tuning ",
-    "updated_at": "2026-03-02T14:18:22Z"
+    "state": "open",
+    "title": "[docs] distributed training",
+    "updated_at": "2026-03-11T17:36:12Z"
   },
   {
-    "additions": 175,
-    "author": "redpanda1995",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes TODO: Implement proper TP support for compressed tensors quantization <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release not\u2026",
+    "additions": 6,
+    "author": "michaelbenayoun",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? To be merged after #44302 and https://github.com/huggingface/kernels/pull/285. It adds the `neuron` device in checks for custom kernels, enabling to load kernels for Neuron devices.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44377",
-    "created_at": "2026-03-01T22:46:07Z",
-    "deletions": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44417",
+    "created_at": "2026-03-03T20:15:26Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44377/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44377",
+    "files_url": "https://github.com/huggingface/transformers/pull/44417/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44417",
     "labels": [],
-    "merged": false,
-    "number": 44377,
+    "merged": true,
+    "number": 44417,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Implement Tensor Parallelism (TP) support for compressed tensors quantization",
-    "updated_at": "2026-03-02T14:15:47Z"
+    "title": "Neuron kernels integration",
+    "updated_at": "2026-03-05T17:09:39Z"
   },
   {
-    "additions": 4,
-    "author": "stuckvgn",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary The quick-start code examples in `flan-t5.md` and `flan-ul2.md` use `\"A step by step recipe to make bolognese pasta:\"` as the demo prompt, with output that includes `ground beef`. This PR replaces it with `\"A step by step recipe\u2026",
-    "changed_files": 2,
+    "additions": 1,
+    "author": "tyler-romero",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Register `olmo_hybrid` in `TOKENIZER_MAPPING_NAMES` so auto-tokenizer resolution works, matching the other auto-registrations already in place for this model.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44376",
-    "created_at": "2026-03-01T17:41:48Z",
-    "deletions": 4,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44416",
+    "created_at": "2026-03-03T19:30:56Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44376/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44376",
+    "files_url": "https://github.com/huggingface/transformers/pull/44416/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44416",
     "labels": [],
-    "merged": false,
-    "number": 44376,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44416,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "docs: update Flan-T5 and Flan-UL2 example to use plant-based recipe prompt",
-    "updated_at": "2026-03-14T06:47:36Z"
+    "title": "[tiny] Add olmo_hybrid to tokenizer auto-mapping",
+    "updated_at": "2026-03-04T19:26:10Z"
   },
   {
-    "additions": 6829,
-    "author": "NielsRogge",
+    "additions": 2,
+    "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds RF-DETR using Codex 5.3. It did everything: modular (in 600 lines of code), fast and slow image processors, conversion script with bells and whistles (setting `id2label` etc.) To do: - [x] verify loss c\u2026",
-    "changed_files": 21,
+    "body_excerpt": "# What does this PR do? This PR removes @MekkCyber from the PR template. cc @Rocketknight1 you only need to ping me now ;)",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44375",
-    "created_at": "2026-03-01T17:32:17Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44415",
+    "created_at": "2026-03-03T16:59:08Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44375/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44375",
+    "files_url": "https://github.com/huggingface/transformers/pull/44415/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44415",
     "labels": [],
-    "merged": false,
-    "number": 44375,
+    "merged": true,
+    "number": 44415,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add RF-DETR",
-    "updated_at": "2026-03-05T16:00:53Z"
+    "state": "closed",
+    "title": "Update PR template",
+    "updated_at": "2026-03-04T14:13:04Z"
   },
   {
-    "additions": 4,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes #44373 The `position_ids` parameter docstrings in `_get_unpad_data()` (line 360) and `_upad_input()` (line 413) in `src/transformers/modeling_flash_attention_utils.py` were incorrectly describing `attention_m\u2026",
+    "additions": 35,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/44303 - see also comments here https://github.com/huggingface/transformers/pull/44316#issuecomment-3984362089. Supersedes https://github.com/huggingface/trans\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44374",
-    "created_at": "2026-03-01T17:07:35Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44414",
+    "created_at": "2026-03-03T16:47:47Z",
+    "deletions": 39,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44374/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44374",
+    "files_url": "https://github.com/huggingface/transformers/pull/44414/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44414",
     "labels": [],
-    "merged": false,
-    "number": 44374,
+    "merged": true,
+    "number": 44414,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix incorrect position_ids docstring in modeling_flash_attention_utils.py",
-    "updated_at": "2026-03-02T14:06:30Z"
+    "title": "Reduce tqdm verbosity during model loading",
+    "updated_at": "2026-03-03T16:57:56Z"
   },
   {
-    "additions": 5,
-    "author": "N3u0ns",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44355 `inspect.getsource()` fails with `TypeError` when called on Cython-compiled functions or built-in functions that don't have Python source code. This adds a try/except block to gracefully handle this case by returnin\u2026",
+    "additions": 4,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44372",
-    "created_at": "2026-03-01T13:53:58Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44413",
+    "created_at": "2026-03-03T16:24:43Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44372/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44372",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44372,
+    "files_url": "https://github.com/huggingface/transformers/pull/44413/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44413",
+    "labels": [],
+    "merged": true,
+    "number": 44413,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: handle Cython-compiled functions in get_docstring_indentation_level",
-    "updated_at": "2026-03-02T13:39:50Z"
+    "title": "Fix peft conversion mappings",
+    "updated_at": "2026-03-03T17:08:39Z"
   },
   {
-    "additions": 12,
-    "author": "leaderofARS",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# Fix documentation inconsistencies in integrations folder ## Description This PR addresses documentation errors and inconsistencies across the integrations module, specifically clarifying terminology and deprecation status in two key inte\u2026",
-    "changed_files": 3,
+    "additions": 138,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Extends type checking to `src/transformers/quantizers`",
+    "changed_files": 28,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44369",
-    "created_at": "2026-03-01T07:34:43Z",
-    "deletions": 9,
+    "comments_count": 25,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44412",
+    "created_at": "2026-03-03T14:53:31Z",
+    "deletions": 74,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44369/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44369",
+    "files_url": "https://github.com/huggingface/transformers/pull/44412/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44412",
     "labels": [],
-    "merged": false,
-    "number": 44369,
-    "review_comments_count": 7,
-    "state": "open",
-    "title": "Feature/integrations docs fix",
-    "updated_at": "2026-03-06T19:47:39Z"
+    "merged": true,
+    "number": 44412,
+    "review_comments_count": 33,
+    "state": "closed",
+    "title": "chore(typing): Add type checking to `src/transformers/quantizers`",
+    "updated_at": "2026-03-11T11:24:11Z"
   },
-  {
-    "additions": 171,
-    "author": "jayakumarpujar",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #43701 Models with `_checkpoint_conversion_mapping` (e.g. VLMs like Qwen2.5VL, LLaVA, ColPali, etc.) use a key renaming system: - **Loading** (`from_pretrained`): Checkpoint keys are renamed from original format \u2192 model fo\u2026",
+  {
+    "additions": 59,
+    "author": "burtenshaw",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR updates `AGENTS.md` to discourage duplicated and trivial work by agents. - CLAUDE.md-> AGENTS.md - ssue-thread coordination before PRs - mandatory duplicate-PR checks with gh commands - no one-off busywork P\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44366",
-    "created_at": "2026-03-01T03:43:16Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44411",
+    "created_at": "2026-03-03T11:35:13Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44366/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44366",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44366,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44411/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44411",
+    "labels": [],
+    "merged": true,
+    "number": 44411,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Fix resume_from_checkpoint key mismatch for models with _checkpoint_conversion_mapping",
-    "updated_at": "2026-03-02T14:02:22Z"
+    "title": "Update agentic contributions guidelines in AGENTS.md to force yielding.",
+    "updated_at": "2026-03-12T09:28:45Z"
   },
   {
-    "additions": 8,
-    "author": "jayakumarpujar",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary - Fixes #44360 - The reference DeepSeek-V3.2 `fp8_index` kernel applies **ReLU** to per-head q\u00b7k scores before weighting and summing across heads: ``` logits[i3_n, i_h] = T.max(logits[i3_n, i_h], 0) * q_s_frag[i_h] ``` [Referenc\u2026",
+    "additions": 38,
+    "author": "zvik",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "# What does this PR do? This PR allows the Granite-speech model to use hidden states from the encoder hidden layers. This is an internal model option that is required for the next generation of Granite-speech models. ## Changes: - New conf\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44364",
-    "created_at": "2026-03-01T02:19:14Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44408",
+    "created_at": "2026-03-03T07:50:39Z",
     "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44364/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44364",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44408/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44408",
+    "labels": [],
     "merged": false,
-    "number": 44364,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add missing ReLU in GlmMoeDsaIndexer scoring",
-    "updated_at": "2026-03-02T13:55:48Z"
+    "number": 44408,
+    "review_comments_count": 10,
+    "state": "open",
+    "title": "Add option to export  encoder hidden states for Granite-speech",
+    "updated_at": "2026-04-16T21:00:13Z"
   },
   {
-    "additions": 57,
-    "author": "jayakumarpujar",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary - Fixes #44327 - `decode_spans()` in the QA pipeline crashes with `ValueError: kth(=N) out of bounds (N)` when `len(scores_flat) == topk` (e.g., `top_k=100` with `seq_len=10`, since `10\u00b2 = 100`) - Root cause: `np.argpartition(ar\u2026",
-    "changed_files": 2,
+    "additions": 23,
+    "author": "hongping-zh",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Adds an \"Energy Efficiency Considerations\" section to the bitsandbytes quantization documentation, providing practical guidance on the energy implications of different quantization configurations. ## Motivation This addresses th\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44363",
-    "created_at": "2026-03-01T01:47:44Z",
-    "deletions": 1,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44407",
+    "created_at": "2026-03-03T04:42:57Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44363/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44363",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44407/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44407",
+    "labels": [],
     "merged": false,
-    "number": 44363,
+    "number": 44407,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix off-by-one in decode_spans causing ValueError with np.argpartition",
-    "updated_at": "2026-03-02T13:03:02Z"
+    "state": "open",
+    "title": "docs: add energy efficiency considerations to bitsandbytes quantization guide",
+    "updated_at": "2026-03-25T11:53:49Z"
   },
   {
-    "additions": 4,
-    "author": "harshaljanjani",
+    "additions": 21,
+    "author": "medhakimbedhief",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following failing Dia use case was identified and fixed in this PR: \u2192 [MIGRATION_GUIDE_V5.md](https://github.com/harshaljanjani/transformers/blob/main/MIGRATION_GUIDE_V5.md) states that v5 renamed `additional_\u2026",
-    "changed_files": 1,
+    "body_excerpt": "Adds sequence-classification support for Qwen3.5 in AutoModelForSequenceClassification. **What does this PR do?** This PR enables loading Qwen3.5 checkpoints with `AutoModelForSequenceClassification`, which previously failed with: `ValueEr\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44362",
-    "created_at": "2026-02-28T20:04:05Z",
-    "deletions": 6,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44406",
+    "created_at": "2026-03-03T03:44:37Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44362/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44362",
+    "files_url": "https://github.com/huggingface/transformers/pull/44406/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44406",
     "labels": [],
     "merged": true,
-    "number": 44362,
+    "number": 44406,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(tokenizer): Fix MLukeTokenizer AttributeError post-v5 refactor",
-    "updated_at": "2026-03-02T14:51:18Z"
+    "title": "Add Qwen3.5 support for sequence classification",
+    "updated_at": "2026-04-08T11:33:19Z"
   },
   {
-    "additions": 341,
-    "author": "sxu75374",
+    "additions": 0,
+    "author": "Rocketknight1",
+    "author_association": "MEMBER",
+    "body_excerpt": "Some generate tests have a ~1% chance of generating short outputs because they hit an EOS token early, which causes the test to flake because it asserts the output shape. This PR enforces `min_length` so that doesn't happen!",
+    "changed_files": 3,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44401",
+    "created_at": "2026-03-02T18:26:56Z",
+    "deletions": 77,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44401/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44401",
+    "labels": [],
+    "merged": true,
+    "number": 44401,
+    "review_comments_count": 4,
+    "state": "closed",
+    "title": "Enforce min length in some generate tests",
+    "updated_at": "2026-03-04T14:05:50Z"
+  },
+  {
+    "additions": 14,
+    "author": "Kokonico",
     "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Adds a workaround for the PyTorch MPS `sdpa_vector_2pass_mps` correctness bug ([pytorch/pytorch#174861](https://github.com/pytorch/pytorch/issues/174861)). **The problem:** On Apple Silicon with MPS backend, `F.scal\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? As per title, confirms interactive terminal before adding formatting to loading_report output. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44359",
-    "created_at": "2026-02-28T17:47:01Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44400",
+    "created_at": "2026-03-02T18:21:48Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44359/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44359",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44400/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44400",
+    "labels": [],
     "merged": false,
-    "number": 44359,
+    "number": 44400,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(sdpa): add workaround for MPS sdpa_vector_2pass_mps correctness bug",
-    "updated_at": "2026-03-02T13:54:58Z"
+    "title": "fix: interactive terminal checks for formatting in loading_report.py",
+    "updated_at": "2026-03-09T14:46:29Z"
   },
   {
-    "additions": 6,
-    "author": "hardikmeisheri",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "## Summary - `ShieldGemma2ForImageClassification` was missing `_tied_weights_keys`, so `model.lm_head.weight` was randomly re-initialized on every `from_pretrained` call instead of being tied to `embed_tokens.weight`. - This caused non-det\u2026",
-    "changed_files": 1,
+    "additions": 3,
+    "author": "weiguangli-io",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes #44384 In `Qwen3_5TextModel.forward`, after splitting `position_ids` into `text_position_ids` (index 0, for text) and `position_ids` (indices 1:, for temporal/height/width), the decoder layer call incorrectly passed `posit\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 14,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44358",
-    "created_at": "2026-02-28T16:49:27Z",
-    "deletions": 2,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44399",
+    "created_at": "2026-03-02T17:28:59Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44358/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44358",
+    "files_url": "https://github.com/huggingface/transformers/pull/44399/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44399",
     "labels": [],
     "merged": true,
-    "number": 44358,
-    "review_comments_count": 1,
+    "number": 44399,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix ShieldGemma2 non-reproducible outputs by adding _tied_weights_keys",
-    "updated_at": "2026-03-16T20:02:09Z"
+    "title": "Fix position_ids typo in Qwen3_5TextModel forward pass",
+    "updated_at": "2026-03-06T01:48:22Z"
   },
   {
-    "additions": 482,
-    "author": "NabilMch",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 127,
+    "additions": 4,
+    "author": "weiguangli-io",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Adds the missing ReLU activation in `GlmMoeDsaIndexer.forward()` on per-head q\u00b7k scores before the weighted sum across heads. The reference DeepSeek-V3.2 `fp8_index` kernel applies `T.max(logits, 0)` (i.e., ReLU) a\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44357",
-    "created_at": "2026-02-28T15:11:37Z",
-    "deletions": 489,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44398",
+    "created_at": "2026-03-02T16:41:01Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44357/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44357",
+    "files_url": "https://github.com/huggingface/transformers/pull/44398/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44398",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44357,
+    "number": 44398,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix RoPE inv_freq default initialization (Issue #39753)",
-    "updated_at": "2026-03-02T13:50:00Z"
+    "title": "Add missing ReLU in GlmMoeDsaIndexer scoring",
+    "updated_at": "2026-03-04T13:54:20Z"
   },
   {
-    "additions": 6,
-    "author": "iamaber",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Checks if model is already in target dtype before casting to avoid redundant copies that cause 25% performance degradation with `--fp16_full_eval`. ## Changes - Added dtype check before casting model to fp16/bf16 in `evaluation_\u2026",
-    "changed_files": 1,
+    "additions": 86,
+    "author": "weiguangli-io",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes #44380 `GPT2Attention.forward()` did not pass the `scaling` parameter to `attention_interface`, causing `scale_attn_weights` and `scale_attn_by_inverse_layer_idx` config options to be silently ignored when usi\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44356",
-    "created_at": "2026-02-28T14:24:32Z",
-    "deletions": 2,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44397",
+    "created_at": "2026-03-02T16:14:37Z",
+    "deletions": 50,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44356/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44356",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44356,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44397/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44397",
+    "labels": [],
+    "merged": true,
+    "number": 44397,
+    "review_comments_count": 17,
     "state": "closed",
-    "title": "fix: avoid redundant fp16/bf16 model casts in evaluation_loop",
-    "updated_at": "2026-03-02T13:59:38Z"
+    "title": "Fix GPT2 attention scaling ignored in SDPA/FlashAttention",
+    "updated_at": "2026-03-04T16:47:42Z"
   },
   {
-    "additions": 73,
-    "author": "sxu75374",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes an off-by-one error in `decode_spans()` where `np.argpartition` is called with `kth == len(arr)` when `topk` equals the number of candidate scores. This raises `ValueError: kth(=N) out of bounds (N)`. **Root\u2026",
+    "additions": 3,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Some renaming should just never be applied when the weight format already matches. (this is actually regardless of remote code). This allows us to remove 1 test added in https://github.com/huggingface/transformers/c\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44354",
-    "created_at": "2026-02-28T08:46:39Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44396",
+    "created_at": "2026-03-02T15:50:27Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44354/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44354",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44354,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44396/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44396",
+    "labels": [],
+    "merged": true,
+    "number": 44396,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "fix: off-by-one in decode_spans causes ValueError when topk == len(scores)",
-    "updated_at": "2026-03-02T13:02:38Z"
+    "title": "[ Dynamic weight loader] fix remote code when format matches",
+    "updated_at": "2026-03-03T17:53:39Z"
   },
   {
-    "additions": 50,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@IlyasMoutawwakil , pls help review, thx!",
-    "changed_files": 1,
+    "additions": 153,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title!",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44353",
-    "created_at": "2026-02-28T07:50:37Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44395",
+    "created_at": "2026-03-02T14:52:12Z",
+    "deletions": 31,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44353/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44353",
+    "files_url": "https://github.com/huggingface/transformers/pull/44395/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44395",
     "labels": [],
     "merged": true,
-    "number": 44353,
-    "review_comments_count": 1,
+    "number": 44395,
+    "review_comments_count": 22,
     "state": "closed",
-    "title": "add expectations for xpu for olmo_hybrid model",
-    "updated_at": "2026-04-02T03:22:06Z"
+    "title": "Fix kernels security issue",
+    "updated_at": "2026-03-03T15:28:20Z"
   },
   {
-    "additions": 13,
-    "author": "giulio-leone",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44336 The `loading_report` module emitted **bold/italic ANSI escape codes** even when `stdout` was not connected to a terminal (e.g. piped or redirected output). While `_color()` already gated color codes behind `sys.stdo\u2026",
-    "changed_files": 1,
+    "additions": 4742,
+    "author": "eustlb",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? #43514 precedes this PR",
+    "changed_files": 61,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44352",
-    "created_at": "2026-02-28T06:22:19Z",
-    "deletions": 6,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44352/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44352",
-    "labels": [
-      "Code agent slop"
-    ],
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44394",
+    "created_at": "2026-03-02T14:49:05Z",
+    "deletions": 7232,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44394/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44394",
+    "labels": [],
     "merged": false,
-    "number": 44352,
+    "number": 44394,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: suppress ANSI escape codes when stdout is not a terminal",
-    "updated_at": "2026-03-02T13:59:15Z"
+    "state": "open",
+    "title": "\ud83d\udea8\ud83d\udea7 FeatureExtractor \u2192 AudioProcessor",
+    "updated_at": "2026-04-03T22:16:41Z"
   },
   {
-    "additions": 7,
-    "author": "kaixuanliu",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": null,
+    "additions": 61,
+    "author": "ebezzam",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? The acoustic tokenizer was updated to use `VoxtralRealtimeConv1dPaddingCache` in #43625 but the ASR model wasn't updated.",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44350",
-    "created_at": "2026-02-28T03:20:47Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44392",
+    "created_at": "2026-03-02T13:41:19Z",
+    "deletions": 100,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44350/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44350",
+    "files_url": "https://github.com/huggingface/transformers/pull/44392/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44392",
     "labels": [],
     "merged": true,
-    "number": 44350,
-    "review_comments_count": 6,
+    "number": 44392,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "skip 1 invalid test case for higgs_audio_v2",
-    "updated_at": "2026-04-09T02:32:42Z"
+    "title": "[VibeVoice ASR] Use updated padding cache for ASR model.",
+    "updated_at": "2026-03-02T13:51:02Z"
   },
   {
-    "additions": 49,
-    "author": "zzc0430",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? > Inspired by https://github.com/huggingface/transformers/pull/44347#issuecomment-3976028358 Fixes `transformers serve` failing with hybrid models like Qwen3.5 that use `linear_attention` layers. Two issues are addr\u2026",
-    "changed_files": 3,
+    "additions": 3040,
+    "author": "ebezzam",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Tokenizer was updated to Voxtral cache object in #43625, but forgot to update that of the ASR model",
+    "changed_files": 30,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44349",
-    "created_at": "2026-02-28T03:09:30Z",
-    "deletions": 3,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44391",
+    "created_at": "2026-03-02T13:29:59Z",
+    "deletions": 311,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44349/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44349",
+    "files_url": "https://github.com/huggingface/transformers/pull/44391/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44391",
     "labels": [],
     "merged": false,
-    "number": 44349,
+    "number": 44391,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: support linear_attention in continuous batching and fix serve ch\u2026",
-    "updated_at": "2026-03-02T13:48:04Z"
+    "title": "[VibeVoice ASR] Use newer cache object for modular",
+    "updated_at": "2026-03-02T13:34:23Z"
   },
   {
-    "additions": 341,
-    "author": "n0kovo",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Most quantized models for Apple Silicon on the Hub are in MLX format. The `MetalConfig` quantization backend supports on-the-fly quantization of standard checkpoints but cannot load pre-quantized MLX models. This PR fixes the fi\u2026",
-    "changed_files": 4,
+    "additions": 3330,
+    "author": "liding-nv",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44348",
-    "created_at": "2026-02-28T00:24:32Z",
-    "deletions": 32,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44390",
+    "created_at": "2026-03-02T13:22:21Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44348/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44348",
+    "files_url": "https://github.com/huggingface/transformers/pull/44390/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44390",
     "labels": [],
-    "merged": false,
-    "number": 44348,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Enable MetalConfig to load pre-quantized MLX models from HuggingFace Hub",
-    "updated_at": "2026-03-02T17:18:46Z"
+    "merged": true,
+    "number": 44390,
+    "review_comments_count": 20,
+    "state": "closed",
+    "title": "add support for nemotron_3",
+    "updated_at": "2026-03-03T18:18:50Z"
   },
   {
-    "additions": 49,
-    "author": "sxu75374",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes `AttributeError` when using continuous batching with composite model configs (e.g. `Qwen3_5Config` for vision-language models). Composite configs store attributes like `num_attention_heads` and `num_key_value\u2026",
+    "additions": 5,
+    "author": "Abdennacer-Badaoui",
+    "author_association": "MEMBER",
+    "body_excerpt": "Adds explicit `timm` installation to the AMD ROCm Docker image. This causes ~200 test failures in AMD CI (e.g., [gemma3n vision tests](https://github.com/huggingface/transformers/actions/runs/22474359922/job/65104428291)). This mirrors wha\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44347",
-    "created_at": "2026-02-27T22:48:49Z",
-    "deletions": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44389",
+    "created_at": "2026-03-02T13:09:55Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44347/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44347",
+    "files_url": "https://github.com/huggingface/transformers/pull/44389/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44389",
+    "labels": [],
+    "merged": true,
+    "number": 44389,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "[AMD CI] Add missing timm dependency to ROCm Docker image",
+    "updated_at": "2026-03-03T12:00:19Z"
+  },
+  {
+    "additions": 0,
+    "author": "sahilmaniyar888",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Fixes #44336 ### Summary This PR prevents ANSI style escape sequences from being emitted by `loading_report` when stdout is non-interactive (for example, redirected logs/files). ### Changes - Added a small helper `_\u2026",
+    "changed_files": 0,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44388",
+    "created_at": "2026-03-02T11:40:49Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44388/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44388",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44347,
+    "number": 44388,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: resolve composite config in PagedAttentionCache and group_layers_by_attn_type",
-    "updated_at": "2026-03-02T13:41:23Z"
+    "title": "Fix loading report ANSI styles for non-TTY output",
+    "updated_at": "2026-03-11T06:29:31Z"
   },
   {
-    "additions": 4,
-    "author": "sxu75374",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Fixes the `q_a_layernorm` and `kv_a_layernorm` in DeepSeek V2/V3 MLA attention to explicitly receive `config.rms_norm_eps` instead of falling back to the RMSNorm class default (`1e-6`). **The problem:** All other RM\u2026",
+    "additions": 62,
+    "author": "eustlb",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per title",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44346",
-    "created_at": "2026-02-27T21:47:45Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44386",
+    "created_at": "2026-03-02T10:43:13Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44346/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44346",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44346,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44386/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44386",
+    "labels": [],
+    "merged": true,
+    "number": 44386,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "fix(deepseek): pass config.rms_norm_eps to MLA q/kv layernorms",
-    "updated_at": "2026-03-02T13:26:21Z"
+    "title": "[higgs-audio-v2] fix sampling",
+    "updated_at": "2026-03-02T13:06:23Z"
   },
   {
-    "additions": 13,
-    "author": "manavshrivastavagit",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44303 When redirecting `from_pretrained` output to a log file (e.g. in CI), the \"Loading weights\" tqdm bar was updating its postfix with `Materializing param=...` on every parameter, producing huge log files. ## Change -\u2026",
-    "changed_files": 2,
+    "additions": 8,
+    "author": "NielsRogge",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? I wasn't able to run make check-repo locally successfully, unless the following 3 fixes were applied.",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44345",
-    "created_at": "2026-02-27T21:05:22Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44385",
+    "created_at": "2026-03-02T09:45:15Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44345/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44345",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44385/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44385",
+    "labels": [],
     "merged": false,
-    "number": 44345,
+    "number": 44385,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Less verbose weight-loading tqdm when stdout is not a TTY (fixes #44303)",
-    "updated_at": "2026-03-02T13:49:11Z"
+    "state": "open",
+    "title": "Fix make check-repo",
+    "updated_at": "2026-03-02T09:54:23Z"
   },
   {
-    "additions": 6,
-    "author": "manavshrivastavagit",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44297 Qwen3.5 models on the Hub (e.g. [Qwen/Qwen3.5-27B](https://huggingface.co/Qwen/Qwen3.5-27B)) use `\"tokenizer_class\": \"Qwen2Tokenizer\"` in `tokenizer_config.json`, but `TOKENIZER_MAPPING_NAMES` had `qwen3_5` \u2192 `\"Qwen\u2026",
-    "changed_files": 1,
+    "additions": 4,
+    "author": "JJJYmmm",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fix the attr `_no_split_modules` of `Qwen3_5Model` and `Qwen3_5MoeModel`, which affect the FSDP init of hf Trainer.",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44344",
-    "created_at": "2026-02-27T21:04:27Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44382",
+    "created_at": "2026-03-02T05:42:48Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44344/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44344",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44344,
+    "files_url": "https://github.com/huggingface/transformers/pull/44382/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44382",
+    "labels": [],
+    "merged": true,
+    "number": 44382,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix tokenizer_class in tokenizer_config.json for Qwen3.5 save_pretrained (fixes #44297)",
-    "updated_at": "2026-03-02T13:17:41Z"
+    "title": "[Bugfix] fix qwen3.5 no split module",
+    "updated_at": "2026-03-02T16:17:22Z"
   },
   {
-    "additions": 16,
-    "author": "manavshrivastavagit",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44336 `utils/loading_report.py` was emitting ANSI codes for **bold** and *italic* via `PALETTE['bold']` and `PALETTE['italic']` without checking if stdout is connected to a terminal. `_color()` already respects `sys.stdou\u2026",
-    "changed_files": 2,
+    "additions": 2,
+    "author": "carcel-yu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? This PR adds MLU support to `is_torch_bf16_gpu_available()` by checking `torch.mlu.is_bf16_supported()` when an MLU device is available. ### Why is this needed? MLU devices support bf16 training, but they are curr\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44343",
-    "created_at": "2026-02-27T20:58:33Z",
-    "deletions": 9,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44381",
+    "created_at": "2026-03-02T05:34:49Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44343/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44343",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44343,
+    "files_url": "https://github.com/huggingface/transformers/pull/44381/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44381",
+    "labels": [],
+    "merged": true,
+    "number": 44381,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix ANSI codes in loading_report when stdout is not a TTY (fixes #44336)",
-    "updated_at": "2026-03-02T13:44:43Z"
+    "title": "Add MLU bf16 support to is_torch_bf16_gpu_available",
+    "updated_at": "2026-03-06T14:34:30Z"
   },
   {
-    "additions": 384,
-    "author": "stevhliu",
+    "additions": 579,
+    "author": "remi-or",
     "author_association": "MEMBER",
-    "body_excerpt": "- created a new performance section divided into memory and speed optimizations - model memory training anatomy [guide](https://huggingface.co/docs/transformers/main/en/model_memory_anatomy) is now the more descriptive and simplified GPU m\u2026",
-    "changed_files": 9,
+    "body_excerpt": "# Summary ! This PR is in draft, waiting for https://github.com/huggingface/transformers/pull/44227 to be merged This PR adds support for the `flash_attention_with_kvcache` kernel in continuoys batching. This is very efficient for decode-o\u2026",
+    "changed_files": 14,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44342",
-    "created_at": "2026-02-27T20:10:49Z",
-    "deletions": 274,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44379",
+    "created_at": "2026-03-01T23:13:17Z",
+    "deletions": 235,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44342/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44342",
+    "files_url": "https://github.com/huggingface/transformers/pull/44379/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44379",
     "labels": [],
     "merged": true,
-    "number": 44342,
-    "review_comments_count": 12,
+    "number": 44379,
+    "review_comments_count": 19,
     "state": "closed",
-    "title": "[docs] training performance",
-    "updated_at": "2026-04-09T20:43:32Z"
+    "title": "[CB] Add paged_attention kernel",
+    "updated_at": "2026-03-09T22:16:31Z"
   },
   {
-    "additions": 12,
-    "author": "Kokonico",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #44336 ## Changes * Added a new `_palette` function to return the ANSI code for a given color or format only if `sys.stdout` is interactive. (`src/transformers/utils/loading_report.py`) * Updated all usages of `PALETTE[<format>]` in\u2026",
+    "additions": 1,
+    "author": "redpanda1995",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44341",
-    "created_at": "2026-02-27T19:30:30Z",
-    "deletions": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44378",
+    "created_at": "2026-03-01T22:57:50Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44341/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44341",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44378/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44378",
+    "labels": [],
     "merged": false,
-    "number": 44341,
-    "review_comments_count": 2,
+    "number": 44378,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix and optimize ANSI color handling in loading report for interactive terminals",
-    "updated_at": "2026-03-02T18:16:00Z"
+    "title": "Fix logging with each layer with ms-swift lora fine-tuning ",
+    "updated_at": "2026-03-02T14:18:22Z"
   },
   {
-    "additions": 33,
-    "author": "Rocketknight1",
-    "author_association": "MEMBER",
-    "body_excerpt": "Some speculative tests seem flaky with SDPA but reliable with `eager` attention. In local testing, `test_speculative_decoding_equals_regular_decoding` fails 5-10% of the time without this change. and I also saw CI failures. Failures are re\u2026",
+    "additions": 175,
+    "author": "redpanda1995",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes TODO: Implement proper TP support for compressed tensors quantization <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release not\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44340",
-    "created_at": "2026-02-27T18:09:09Z",
-    "deletions": 27,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44377",
+    "created_at": "2026-03-01T22:46:07Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44340/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44340",
+    "files_url": "https://github.com/huggingface/transformers/pull/44377/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44377",
     "labels": [],
-    "merged": true,
-    "number": 44340,
-    "review_comments_count": 6,
+    "merged": false,
+    "number": 44377,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix speculative tests that are flaky with SDPA",
-    "updated_at": "2026-03-02T17:18:27Z"
+    "title": "Implement Tensor Parallelism (TP) support for compressed tensors quantization",
+    "updated_at": "2026-03-02T14:15:47Z"
   },
   {
-    "additions": 6221,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? \u2192 This PR adds **DEIMv2** to Transformers! \u2192 **IMP:** I've linked two notebooks: a [Colab notebook here](https://colab.research.google.com/drive/1jCNefxrKiHWdBEIYTcU3jsd9xyWDwIxC?usp=sharing) demonstrating the fun\u2026",
-    "changed_files": 16,
-    "cluster_id": "cluster-41211-3",
-    "cluster_ids": [
-      "cluster-41211-3"
-    ],
-    "cluster_role": "canonical",
-    "comments_count": 24,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44339",
-    "created_at": "2026-02-27T18:08:53Z",
-    "deletions": 0,
+    "additions": 4,
+    "author": "stuckvgn",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary The quick-start code examples in `flan-t5.md` and `flan-ul2.md` use `\"A step by step recipe to make bolognese pasta:\"` as the demo prompt, with output that includes `ground beef`. This PR replaces it with `\"A step by step recipe\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44376",
+    "created_at": "2026-03-01T17:41:48Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44339/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44339",
+    "files_url": "https://github.com/huggingface/transformers/pull/44376/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44376",
     "labels": [],
     "merged": false,
-    "number": 44339,
-    "review_comments_count": 215,
-    "state": "open",
-    "title": "model: Add DEIMv2 to Transformers",
-    "updated_at": "2026-04-15T17:06:41Z"
+    "number": 44376,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "docs: update Flan-T5 and Flan-UL2 example to use plant-based recipe prompt",
+    "updated_at": "2026-03-14T06:47:36Z"
   },
   {
-    "additions": 3641,
-    "author": "SunMarc",
+    "additions": 6829,
+    "author": "NielsRogge",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR rework a bit how distributed tests are tested. I tried to keep some of the existing tests and added new tests also. For each of these distributed methods (ddp, fsdp, deepspeeed), we have some common tests li\u2026",
-    "changed_files": 38,
+    "body_excerpt": "# What does this PR do? This PR adds RF-DETR using Codex 5.3. It did everything: modular (in 600 lines of code), fast and slow image processors, conversion script with bells and whistles (setting `id2label` etc.) To do: - [x] verify loss c\u2026",
+    "changed_files": 21,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44338",
-    "created_at": "2026-02-27T17:50:16Z",
-    "deletions": 3762,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44375",
+    "created_at": "2026-03-01T17:32:17Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44338/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44338",
+    "files_url": "https://github.com/huggingface/transformers/pull/44375/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44375",
     "labels": [],
-    "merged": true,
-    "number": 44338,
+    "merged": false,
+    "number": 44375,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Update distributed tests",
-    "updated_at": "2026-03-05T23:35:36Z"
+    "state": "open",
+    "title": "Add RF-DETR",
+    "updated_at": "2026-03-05T16:00:53Z"
   },
   {
-    "additions": 2,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "- moves `kernels-community/flash-attn2:FlashAttention2` to `from_pretrained(attn_implementation...)` - fix error message for registering a kernel",
-    "changed_files": 2,
+    "additions": 4,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes #44373 The `position_ids` parameter docstrings in `_get_unpad_data()` (line 360) and `_upad_input()` (line 413) in `src/transformers/modeling_flash_attention_utils.py` were incorrectly describing `attention_m\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44337",
-    "created_at": "2026-02-27T17:36:54Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44374",
+    "created_at": "2026-03-01T17:07:35Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44337/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44337",
+    "files_url": "https://github.com/huggingface/transformers/pull/44374/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44374",
     "labels": [],
-    "merged": true,
-    "number": 44337,
+    "merged": false,
+    "number": 44374,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] kernelconfig fix",
-    "updated_at": "2026-02-27T22:46:30Z"
+    "title": "Fix incorrect position_ids docstring in modeling_flash_attention_utils.py",
+    "updated_at": "2026-03-02T14:06:30Z"
   },
   {
-    "additions": 57,
-    "author": "ManasVardhan",
+    "additions": 5,
+    "author": "N3u0ns",
     "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the RoFormer model to use the `@capture_outputs` and `@can_return_tuple` decorators, following the established pattern (similar to #44047 for Bloom, #44151 for BioGPT, etc.). ### Changes: - **`RoFormerMod\u2026",
+    "body_excerpt": "## Summary Fixes #44355 `inspect.getsource()` fails with `TypeError` when called on Cython-compiled functions or built-in functions that don't have Python source code. This adds a try/except block to gracefully handle this case by returnin\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44335",
-    "created_at": "2026-02-27T17:23:01Z",
-    "deletions": 172,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44372",
+    "created_at": "2026-03-01T13:53:58Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44335/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44335",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44372/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44372",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44335,
+    "number": 44372,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor RoFormer output tracing with @capture_outputs and @can_return_tuple",
-    "updated_at": "2026-03-03T00:29:59Z"
+    "title": "fix: handle Cython-compiled functions in get_docstring_indentation_level",
+    "updated_at": "2026-03-02T13:39:50Z"
   },
   {
-    "additions": 1,
-    "author": "NielsRogge",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? I had some issues with running `transformers-cli add-new-model-like`. This PR fixes it. Fixes #44661.",
-    "changed_files": 1,
+    "additions": 12,
+    "author": "leaderofARS",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# Fix documentation inconsistencies in integrations folder ## Description This PR addresses documentation errors and inconsistencies across the integrations module, specifically clarifying terminology and deprecation status in two key inte\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44334",
-    "created_at": "2026-02-27T17:13:44Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44369",
+    "created_at": "2026-03-01T07:34:43Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44334/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44334",
+    "files_url": "https://github.com/huggingface/transformers/pull/44369/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44369",
     "labels": [],
-    "merged": true,
-    "number": 44334,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix CookieCutter",
-    "updated_at": "2026-03-13T17:13:28Z"
+    "merged": false,
+    "number": 44369,
+    "review_comments_count": 7,
+    "state": "open",
+    "title": "Feature/integrations docs fix",
+    "updated_at": "2026-03-06T19:47:39Z"
   },
   {
-    "additions": 13,
-    "author": "ManasVardhan",
+    "additions": 171,
+    "author": "jayakumarpujar",
     "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the ALBERT model to use named attribute access instead of index-based access on model outputs, and removes redundant `return_dict=True` arguments from inner model calls (already handled by `@capture_outpu\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary Fixes #43701 Models with `_checkpoint_conversion_mapping` (e.g. VLMs like Qwen2.5VL, LLaVA, ColPali, etc.) use a key renaming system: - **Loading** (`from_pretrained`): Checkpoint keys are renamed from original format \u2192 model fo\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44333",
-    "created_at": "2026-02-27T17:12:25Z",
-    "deletions": 18,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44366",
+    "created_at": "2026-03-01T03:43:16Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44333/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44333",
+    "files_url": "https://github.com/huggingface/transformers/pull/44366/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44366",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44333,
+    "number": 44366,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor ALBERT to use named attributes and remove redundant return_dict=True",
-    "updated_at": "2026-03-02T13:05:54Z"
+    "title": "Fix resume_from_checkpoint key mismatch for models with _checkpoint_conversion_mapping",
+    "updated_at": "2026-03-02T14:02:22Z"
   },
   {
-    "additions": 3,
-    "author": "tysoncung",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fix minor typos found in comments and docstrings: - `orignal` \u2192 `original` in `src/transformers/integrations/peft.py` (lines 245, 284) - Duplicate word `is is` \u2192 `is` in `src/transformers/models/dia/processing_dia.py` (line 89) Small clean\u2026",
+    "additions": 8,
+    "author": "jayakumarpujar",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Fixes #44360 - The reference DeepSeek-V3.2 `fp8_index` kernel applies **ReLU** to per-head q\u00b7k scores before weighting and summing across heads: ``` logits[i3_n, i_h] = T.max(logits[i3_n, i_h], 0) * q_s_frag[i_h] ``` [Referenc\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44332",
-    "created_at": "2026-02-27T16:11:46Z",
-    "deletions": 3,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44364",
+    "created_at": "2026-03-01T02:19:14Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44332/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44332",
-    "labels": [],
-    "merged": true,
-    "number": 44332,
+    "files_url": "https://github.com/huggingface/transformers/pull/44364/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44364",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44364,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix typos in comments and docstrings",
-    "updated_at": "2026-02-27T18:02:59Z"
+    "title": "Add missing ReLU in GlmMoeDsaIndexer scoring",
+    "updated_at": "2026-03-02T13:55:48Z"
   },
   {
-    "additions": 33,
-    "author": "kashif",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? fixed the bfloat16 dtype mismatch and Loss computation shape mismatch. Also added tests for these. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to ap\u2026",
-    "changed_files": 3,
+    "additions": 57,
+    "author": "jayakumarpujar",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Fixes #44327 - `decode_spans()` in the QA pipeline crashes with `ValueError: kth(=N) out of bounds (N)` when `len(scores_flat) == topk` (e.g., `top_k=100` with `seq_len=10`, since `10\u00b2 = 100`) - Root cause: `np.argpartition(ar\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44331",
-    "created_at": "2026-02-27T15:46:08Z",
-    "deletions": 10,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44363",
+    "created_at": "2026-03-01T01:47:44Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44331/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44331",
-    "labels": [],
-    "merged": true,
-    "number": 44331,
-    "review_comments_count": 9,
+    "files_url": "https://github.com/huggingface/transformers/pull/44363/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44363",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44363,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[timesfm2_5] fix timesfm2.5 loss",
-    "updated_at": "2026-03-03T17:22:56Z"
+    "title": "Fix off-by-one in decode_spans causing ValueError with np.argpartition",
+    "updated_at": "2026-03-02T13:03:02Z"
   },
   {
-    "additions": 289,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title! Follow-up of https://github.com/huggingface/transformers/pull/44181 with more models!",
-    "changed_files": 136,
+    "additions": 4,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? The following failing Dia use case was identified and fixed in this PR: \u2192 [MIGRATION_GUIDE_V5.md](https://github.com/harshaljanjani/transformers/blob/main/MIGRATION_GUIDE_V5.md) states that v5 renamed `additional_\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 12,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44330",
-    "created_at": "2026-02-27T15:33:02Z",
-    "deletions": 1682,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44362",
+    "created_at": "2026-02-28T20:04:05Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44330/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44330",
+    "files_url": "https://github.com/huggingface/transformers/pull/44362/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44362",
     "labels": [],
     "merged": true,
-    "number": 44330,
-    "review_comments_count": 13,
+    "number": 44362,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Remove `cache_position` in more models",
-    "updated_at": "2026-03-11T14:47:50Z"
+    "title": "fix(tokenizer): Fix MLukeTokenizer AttributeError post-v5 refactor",
+    "updated_at": "2026-03-02T14:51:18Z"
   },
   {
-    "additions": 3,
-    "author": "linfeng-du",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
+    "additions": 341,
+    "author": "sxu75374",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Adds a workaround for the PyTorch MPS `sdpa_vector_2pass_mps` correctness bug ([pytorch/pytorch#174861](https://github.com/pytorch/pytorch/issues/174861)). **The problem:** On Apple Silicon with MPS backend, `F.scal\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44329",
-    "created_at": "2026-02-27T15:27:39Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44359",
+    "created_at": "2026-02-28T17:47:01Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44329/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44329",
-    "labels": [],
-    "merged": true,
-    "number": 44329,
-    "review_comments_count": 4,
+    "files_url": "https://github.com/huggingface/transformers/pull/44359/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44359",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44359,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Enable Liger Kernel when doing hyperparameter search.",
-    "updated_at": "2026-03-03T13:44:56Z"
+    "title": "fix(sdpa): add workaround for MPS sdpa_vector_2pass_mps correctness bug",
+    "updated_at": "2026-03-02T13:54:58Z"
   },
   {
-    "additions": 92,
-    "author": "SunMarc",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds guidelines for agents when it comes to add/run trainer tests. This needs to be updated as we modify, refactor the code !",
+    "additions": 6,
+    "author": "hardikmeisheri",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "## Summary - `ShieldGemma2ForImageClassification` was missing `_tied_weights_keys`, so `model.lm_head.weight` was randomly re-initialized on every `from_pretrained` call instead of being tied to `embed_tokens.weight`. - This caused non-det\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44328",
-    "created_at": "2026-02-27T15:17:24Z",
-    "deletions": 0,
+    "comments_count": 14,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44358",
+    "created_at": "2026-02-28T16:49:27Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44328/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44328",
+    "files_url": "https://github.com/huggingface/transformers/pull/44358/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44358",
     "labels": [],
     "merged": true,
-    "number": 44328,
-    "review_comments_count": 2,
+    "number": 44358,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Add testing guide for agents for trainer tests",
-    "updated_at": "2026-02-27T17:32:11Z"
+    "title": "Fix ShieldGemma2 non-reproducible outputs by adding _tied_weights_keys",
+    "updated_at": "2026-03-16T20:02:09Z"
   },
   {
-    "additions": 38,
-    "author": "overcastbulb",
+    "additions": 482,
+    "author": "NabilMch",
     "author_association": "NONE",
-    "body_excerpt": "Adds missing pipeline tutorial example for zero-shot-classification following the existing format of other task examples in pipeline_tutorial.md. Related: #18926",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 127,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44326",
-    "created_at": "2026-02-27T14:37:19Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44357",
+    "created_at": "2026-02-28T15:11:37Z",
+    "deletions": 489,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44326/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44326",
+    "files_url": "https://github.com/huggingface/transformers/pull/44357/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44357",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44326,
+    "number": 44357,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "docs: Add zero-shot-classification example to pipeline tutorial",
-    "updated_at": "2026-02-27T14:46:24Z"
+    "title": "Fix RoPE inv_freq default initialization (Issue #39753)",
+    "updated_at": "2026-03-02T13:50:00Z"
   },
   {
-    "additions": 4,
-    "author": "kashif",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The `CLIPMLP` has the bias set to True but timesFM 2.5 uses `bias=False` in the pretrained model <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appe\u2026",
-    "changed_files": 2,
+    "additions": 6,
+    "author": "iamaber",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Checks if model is already in target dtype before casting to avoid redundant copies that cause 25% performance degradation with `--fp16_full_eval`. ## Changes - Added dtype check before casting model to fp16/bf16 in `evaluation_\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44325",
-    "created_at": "2026-02-27T13:18:40Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44356",
+    "created_at": "2026-02-28T14:24:32Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44325/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44325",
-    "labels": [],
-    "merged": true,
-    "number": 44325,
+    "files_url": "https://github.com/huggingface/transformers/pull/44356/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44356",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44356,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[timesfm2_5] fix timesfm mlp bias",
-    "updated_at": "2026-02-27T13:36:13Z"
+    "title": "fix: avoid redundant fp16/bf16 model casts in evaluation_loop",
+    "updated_at": "2026-03-02T13:59:38Z"
   },
   {
-    "additions": 16,
-    "author": "tonglei19961121",
+    "additions": 73,
+    "author": "sxu75374",
     "author_association": "NONE",
-    "body_excerpt": "Add tutorial example for DocumentQuestionAnswering pipeline following the existing format of other task examples. This PR addresses the first TODO item in #18926. Changes: - Added document-question-answering task example to pipeline_tutori\u2026",
+    "body_excerpt": "## What does this PR do? Fixes an off-by-one error in `decode_spans()` where `np.argpartition` is called with `kth == len(arr)` when `topk` equals the number of candidate scores. This raises `ValueError: kth(=N) out of bounds (N)`. **Root\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44324",
-    "created_at": "2026-02-27T12:29:06Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44354",
+    "created_at": "2026-02-28T08:46:39Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44324/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44324",
+    "files_url": "https://github.com/huggingface/transformers/pull/44354/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44354",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44324,
+    "number": 44354,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "docs: Add document-question-answering example to pipeline tutorial",
-    "updated_at": "2026-02-27T14:35:23Z"
+    "title": "fix: off-by-one in decode_spans causes ValueError when topk == len(scores)",
+    "updated_at": "2026-03-02T13:02:38Z"
   },
   {
-    "additions": 16,
-    "author": "tonglei19961121",
+    "additions": 50,
+    "author": "kaixuanliu",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "@IlyasMoutawwakil , pls help review, thx!",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44353",
+    "created_at": "2026-02-28T07:50:37Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44353/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44353",
+    "labels": [],
+    "merged": true,
+    "number": 44353,
+    "review_comments_count": 1,
+    "state": "closed",
+    "title": "add expectations for xpu for olmo_hybrid model",
+    "updated_at": "2026-04-02T03:22:06Z"
+  },
+  {
+    "additions": 13,
+    "author": "giulio-leone",
     "author_association": "NONE",
-    "body_excerpt": "Add tutorial example for DocumentQuestionAnswering pipeline following the existing format of other task examples. Fixes #18926",
-    "changed_files": 2,
+    "body_excerpt": "## Summary Fixes #44336 The `loading_report` module emitted **bold/italic ANSI escape codes** even when `stdout` was not connected to a terminal (e.g. piped or redirected output). While `_color()` already gated color codes behind `sys.stdo\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44323",
-    "created_at": "2026-02-27T12:26:00Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44352",
+    "created_at": "2026-02-28T06:22:19Z",
+    "deletions": 6,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44323/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44323",
+    "files_url": "https://github.com/huggingface/transformers/pull/44352/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44352",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44323,
+    "number": 44352,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "docs: Add document-question-answering example to pipeline tutorial",
-    "updated_at": "2026-02-27T14:38:43Z"
+    "title": "fix: suppress ANSI escape codes when stdout is not a terminal",
+    "updated_at": "2026-03-02T13:59:15Z"
   },
   {
-    "additions": 12,
+    "additions": 7,
     "author": "kaixuanliu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "@vasqu This PR skipped 2 invalid test cases: ``` tests/models/voxtral_realtime/test_modeling_voxtral_realtime.py::VoxtralRealtimeForConditionalGenerationModelTest::test_generate_with_quant_cache tests/models/voxtral_realtime/test_modeling_\u2026",
-    "changed_files": 1,
+    "body_excerpt": null,
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44321",
-    "created_at": "2026-02-27T09:54:14Z",
-    "deletions": 0,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44350",
+    "created_at": "2026-02-28T03:20:47Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44321/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44321",
+    "files_url": "https://github.com/huggingface/transformers/pull/44350/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44350",
     "labels": [],
     "merged": true,
-    "number": 44321,
+    "number": 44350,
+    "review_comments_count": 6,
+    "state": "closed",
+    "title": "skip 1 invalid test case for higgs_audio_v2",
+    "updated_at": "2026-04-09T02:32:42Z"
+  },
+  {
+    "additions": 49,
+    "author": "zzc0430",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? > Inspired by https://github.com/huggingface/transformers/pull/44347#issuecomment-3976028358 Fixes `transformers serve` failing with hybrid models like Qwen3.5 that use `linear_attention` layers. Two issues are addr\u2026",
+    "changed_files": 3,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44349",
+    "created_at": "2026-02-28T03:09:30Z",
+    "deletions": 3,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44349/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44349",
+    "labels": [],
+    "merged": false,
+    "number": 44349,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "skip 2 invalid test cases for voxtral_realtime model",
-    "updated_at": "2026-04-02T03:22:04Z"
+    "title": "fix: support linear_attention in continuous batching and fix serve ch\u2026",
+    "updated_at": "2026-03-02T13:48:04Z"
   },
   {
-    "additions": 5231,
-    "author": "NielsRogge",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds [SAM3-LiteText: An Anatomical Study of the SAM3 Text Encoder for Efficient Vision-Language Segmentation](https://huggingface.co/papers/2602.12173). Fixes #44205",
-    "changed_files": 22,
+    "additions": 341,
+    "author": "n0kovo",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Most quantized models for Apple Silicon on the Hub are in MLX format. The `MetalConfig` quantization backend supports on-the-fly quantization of standard checkpoints but cannot load pre-quantized MLX models. This PR fixes the fi\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 13,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44320",
-    "created_at": "2026-02-27T08:29:00Z",
-    "deletions": 19,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44348",
+    "created_at": "2026-02-28T00:24:32Z",
+    "deletions": 32,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44320/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44320",
+    "files_url": "https://github.com/huggingface/transformers/pull/44348/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44348",
+    "labels": [],
+    "merged": false,
+    "number": 44348,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Enable MetalConfig to load pre-quantized MLX models from HuggingFace Hub",
+    "updated_at": "2026-03-02T17:18:46Z"
+  },
+  {
+    "additions": 49,
+    "author": "sxu75374",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Fixes `AttributeError` when using continuous batching with composite model configs (e.g. `Qwen3_5Config` for vision-language models). Composite configs store attributes like `num_attention_heads` and `num_key_value\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44347",
+    "created_at": "2026-02-27T22:48:49Z",
+    "deletions": 8,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44347/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44347",
     "labels": [
-      "New model"
+      "Code agent slop"
     ],
-    "merged": true,
-    "number": 44320,
-    "review_comments_count": 63,
+    "merged": false,
+    "number": 44347,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add SAM3-LiteText",
-    "updated_at": "2026-04-13T18:37:07Z"
+    "title": "fix: resolve composite config in PagedAttentionCache and group_layers_by_attn_type",
+    "updated_at": "2026-03-02T13:41:23Z"
   },
   {
-    "additions": 74,
-    "author": "IlyasMoutawwakil",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 3,
+    "additions": 4,
+    "author": "sxu75374",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Fixes the `q_a_layernorm` and `kv_a_layernorm` in DeepSeek V2/V3 MLA attention to explicitly receive `config.rms_norm_eps` instead of falling back to the RMSNorm class default (`1e-6`). **The problem:** All other RM\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44319",
-    "created_at": "2026-02-27T08:20:45Z",
-    "deletions": 56,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44346",
+    "created_at": "2026-02-27T21:47:45Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44319/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44319",
-    "labels": [],
-    "merged": true,
-    "number": 44319,
-    "review_comments_count": 3,
+    "files_url": "https://github.com/huggingface/transformers/pull/44346/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44346",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44346,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Support non-gated experts",
-    "updated_at": "2026-03-02T19:26:38Z"
+    "title": "fix(deepseek): pass config.rms_norm_eps to MLA q/kv layernorms",
+    "updated_at": "2026-03-02T13:26:21Z"
   },
   {
-    "additions": 10,
-    "author": "yoginlangalia",
+    "additions": 13,
+    "author": "manavshrivastavagit",
     "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Adds input validation for bounding box shape in `LayoutLMv3Tokenizer`. When users pass boxes with fewer (or more) than 4 values per box, the tokenizer now raises a clear `ValueError` instead of a confusing generic\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary Fixes #44303 When redirecting `from_pretrained` output to a log file (e.g. in CI), the \"Loading weights\" tqdm bar was updating its postfix with `Materializing param=...` on every parameter, producing huge log files. ## Change -\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44318",
-    "created_at": "2026-02-27T06:40:02Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44345",
+    "created_at": "2026-02-27T21:05:22Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44318/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44318",
+    "files_url": "https://github.com/huggingface/transformers/pull/44345/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44345",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44318,
+    "number": 44345,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Validate bounding box shape in LayoutLMv3Tokenizer",
-    "updated_at": "2026-02-27T14:43:08Z"
+    "title": "Less verbose weight-loading tqdm when stdout is not a TTY (fixes #44303)",
+    "updated_at": "2026-03-02T13:49:11Z"
   },
   {
-    "additions": 4,
-    "author": "sxu75374",
+    "additions": 6,
+    "author": "manavshrivastavagit",
     "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Passes `config.rms_norm_eps` explicitly to `q_a_layernorm` and `kv_a_layernorm` in both DeepSeek V2 and V3 MLA attention. Currently these two norms are constructed without `eps`, falling back to the `RMSNorm` class\u2026",
-    "changed_files": 2,
+    "body_excerpt": "## Summary Fixes #44297 Qwen3.5 models on the Hub (e.g. [Qwen/Qwen3.5-27B](https://huggingface.co/Qwen/Qwen3.5-27B)) use `\"tokenizer_class\": \"Qwen2Tokenizer\"` in `tokenizer_config.json`, but `TOKENIZER_MAPPING_NAMES` had `qwen3_5` \u2192 `\"Qwen\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44317",
-    "created_at": "2026-02-27T04:48:08Z",
-    "deletions": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44344",
+    "created_at": "2026-02-27T21:04:27Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44317/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44317",
+    "files_url": "https://github.com/huggingface/transformers/pull/44344/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44344",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44317,
+    "number": 44344,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(deepseek): pass rms_norm_eps to MLA q/kv layernorms",
-    "updated_at": "2026-02-27T14:30:04Z"
+    "title": "Fix tokenizer_class in tokenizer_config.json for Qwen3.5 save_pretrained (fixes #44297)",
+    "updated_at": "2026-03-02T13:17:41Z"
   },
   {
-    "additions": 2,
-    "author": "jashshah999",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes #44303 The weight loading progress bar called `pbar.refresh()` on every single parameter, bypassing tqdm's built-in rate-limiting. When output is redirected to a log file (e.g. in CI), this produced one line per parameter -- hundreds\u2026",
-    "changed_files": 1,
+    "additions": 16,
+    "author": "manavshrivastavagit",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #44336 `utils/loading_report.py` was emitting ANSI codes for **bold** and *italic* via `PALETTE['bold']` and `PALETTE['italic']` without checking if stdout is connected to a terminal. `_color()` already respects `sys.stdou\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44316",
-    "created_at": "2026-02-27T03:08:28Z",
-    "deletions": 3,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44343",
+    "created_at": "2026-02-27T20:58:33Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44316/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44316",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44343/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44343",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44316,
+    "number": 44343,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Reduce tqdm verbosity during weight loading",
-    "updated_at": "2026-03-03T17:02:34Z"
+    "title": "Fix ANSI codes in loading_report when stdout is not a TTY (fixes #44336)",
+    "updated_at": "2026-03-02T13:44:43Z"
   },
   {
-    "additions": 3484,
-    "author": "jp1924",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Hello, Transformers team! I submitted a PR to add naver-hyperclovax/HyperCLOVAX-SEED-Think-32B (hereafter HCX), developed by the Korean IT company Naver while executing the government's national AI model project. Th\u2026",
-    "changed_files": 24,
+    "additions": 384,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "- created a new performance section divided into memory and speed optimizations - model memory training anatomy [guide](https://huggingface.co/docs/transformers/main/en/model_memory_anatomy) is now the more descriptive and simplified GPU m\u2026",
+    "changed_files": 9,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 22,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44314",
-    "created_at": "2026-02-27T02:01:28Z",
-    "deletions": 18,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44342",
+    "created_at": "2026-02-27T20:10:49Z",
+    "deletions": 274,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44314/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44314",
+    "files_url": "https://github.com/huggingface/transformers/pull/44342/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44342",
     "labels": [],
-    "merged": false,
-    "number": 44314,
-    "review_comments_count": 77,
-    "state": "open",
-    "title": "add HyperClovaX Vision",
-    "updated_at": "2026-04-13T02:23:53Z"
+    "merged": true,
+    "number": 44342,
+    "review_comments_count": 12,
+    "state": "closed",
+    "title": "[docs] training performance",
+    "updated_at": "2026-04-09T20:43:32Z"
   },
   {
-    "additions": 4,
-    "author": "jashshah999",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes the same `TypeError: AddedToken() got multiple values for keyword argument 'special'` that #44281 addressed, but for the `extra_special_tokens` code path which was missed. #44281 (commit 8e663c7) correctly added `value.pop(\"special\",\u2026",
+    "additions": 12,
+    "author": "Kokonico",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #44336 ## Changes * Added a new `_palette` function to return the ANSI code for a given color or format only if `sys.stdout` is interactive. (`src/transformers/utils/loading_report.py`) * Updated all usages of `PALETTE[<format>]` in\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44313",
-    "created_at": "2026-02-27T01:37:45Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44341",
+    "created_at": "2026-02-27T19:30:30Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44313/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44313",
+    "files_url": "https://github.com/huggingface/transformers/pull/44341/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44341",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44313,
-    "review_comments_count": 0,
+    "number": 44341,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix AddedToken duplicate 'special' kwarg for extra_special_tokens",
-    "updated_at": "2026-02-27T14:26:28Z"
+    "title": "Fix and optimize ANSI color handling in loading report for interactive terminals",
+    "updated_at": "2026-03-02T18:16:00Z"
   },
   {
-    "additions": 8,
-    "author": "haosenwang1018",
-    "author_association": "NONE",
-    "body_excerpt": "Replace bare `except:` clauses with `except Exception:` for PEP 8 compliance.",
-    "changed_files": 4,
+    "additions": 33,
+    "author": "Rocketknight1",
+    "author_association": "MEMBER",
+    "body_excerpt": "Some speculative tests seem flaky with SDPA but reliable with `eager` attention. In local testing, `test_speculative_decoding_equals_regular_decoding` fails 5-10% of the time without this change. and I also saw CI failures. Failures are re\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44312",
-    "created_at": "2026-02-27T01:00:33Z",
-    "deletions": 8,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44340",
+    "created_at": "2026-02-27T18:09:09Z",
+    "deletions": 27,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44312/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44312",
+    "files_url": "https://github.com/huggingface/transformers/pull/44340/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44340",
     "labels": [],
-    "merged": false,
-    "number": 44312,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44340,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "fix: replace 8 bare except clauses with except Exception",
-    "updated_at": "2026-02-27T03:27:27Z"
+    "title": "Fix speculative tests that are flaky with SDPA",
+    "updated_at": "2026-03-02T17:18:27Z"
   },
   {
-    "additions": 38,
-    "author": "onel",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Adds species bias documentation across the transformers repository to help model authors and users recognize and address potential biases in language models. The updates include guidance on documenting bias categori\u2026",
-    "changed_files": 5,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44311",
-    "created_at": "2026-02-27T00:02:49Z",
+    "additions": 6220,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? \u2192 This PR adds **DEIMv2** to Transformers! \u2192 **IMP:** I've linked two notebooks: a [Colab notebook here](https://colab.research.google.com/drive/1jCNefxrKiHWdBEIYTcU3jsd9xyWDwIxC?usp=sharing) demonstrating the fun\u2026",
+    "changed_files": 16,
+    "cluster_id": "cluster-41211-3",
+    "cluster_ids": [
+      "cluster-41211-3"
+    ],
+    "cluster_role": "canonical",
+    "comments_count": 26,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44339",
+    "created_at": "2026-02-27T18:08:53Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44311/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44311",
+    "files_url": "https://github.com/huggingface/transformers/pull/44339/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44339",
     "labels": [],
     "merged": false,
-    "number": 44311,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add species bias documentation to model cards and docs",
-    "updated_at": "2026-02-27T14:09:20Z"
+    "number": 44339,
+    "review_comments_count": 231,
+    "state": "open",
+    "title": "model: Add DEIMv2 to Transformers",
+    "updated_at": "2026-04-16T06:56:04Z"
   },
   {
-    "additions": 63,
-    "author": "onel",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Adds documentation for Pruna AI integration to the Transformers ecosystem, following the existing pattern used by vLLM and Unsloth integration docs. ## Changes - Created `docs/source/en/community_integrations/pruna.\u2026",
-    "changed_files": 2,
+    "additions": 3641,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR rework a bit how distributed tests are tested. I tried to keep some of the existing tests and added new tests also. For each of these distributed methods (ddp, fsdp, deepspeeed), we have some common tests li\u2026",
+    "changed_files": 38,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44310",
-    "created_at": "2026-02-27T00:00:31Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44338",
+    "created_at": "2026-02-27T17:50:16Z",
+    "deletions": 3762,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44310/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44310",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44310,
+    "files_url": "https://github.com/huggingface/transformers/pull/44338/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44338",
+    "labels": [],
+    "merged": true,
+    "number": 44338,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "docs: Add Pruna AI integration documentation",
-    "updated_at": "2026-02-27T14:08:21Z"
+    "title": "Update distributed tests",
+    "updated_at": "2026-03-05T23:35:36Z"
   },
   {
-    "additions": 129,
-    "author": "onel",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Adds SkyPilot deployment documentation to the DeepSpeed guide. The new section includes: - Introduction to SkyPilot as a unified framework for running AI workloads across clouds and Kubernetes - Complete example YAM\u2026",
-    "changed_files": 1,
+    "additions": 2,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "- moves `kernels-community/flash-attn2:FlashAttention2` to `from_pretrained(attn_implementation...)` - fix error message for registering a kernel",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44309",
-    "created_at": "2026-02-26T22:44:41Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44337",
+    "created_at": "2026-02-27T17:36:54Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44309/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44309",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44309,
+    "files_url": "https://github.com/huggingface/transformers/pull/44337/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44337",
+    "labels": [],
+    "merged": true,
+    "number": 44337,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add SkyPilot deployment documentation to DeepSpeed guide",
-    "updated_at": "2026-02-27T14:09:55Z"
+    "title": "[docs] kernelconfig fix",
+    "updated_at": "2026-02-27T22:46:30Z"
   },
   {
-    "additions": 5854,
-    "author": "NielsRogge",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 15,
+    "additions": 57,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Refactors the RoFormer model to use the `@capture_outputs` and `@can_return_tuple` decorators, following the established pattern (similar to #44047 for Bloom, #44151 for BioGPT, etc.). ### Changes: - **`RoFormerMod\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44308",
-    "created_at": "2026-02-26T21:26:05Z",
-    "deletions": 9,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44335",
+    "created_at": "2026-02-27T17:23:01Z",
+    "deletions": 172,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44308/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44308",
+    "files_url": "https://github.com/huggingface/transformers/pull/44335/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44335",
     "labels": [],
     "merged": false,
-    "number": 44308,
+    "number": 44335,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Codex/add sam3 litetext model to transformers fuvllg",
-    "updated_at": "2026-02-26T21:35:44Z"
+    "title": "Refactor RoFormer output tracing with @capture_outputs and @can_return_tuple",
+    "updated_at": "2026-03-03T00:29:59Z"
   },
   {
-    "additions": 7,
-    "author": "imstevenpmwork",
+    "additions": 1,
+    "author": "NielsRogge",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR fixes a `TypeError: not all arguments converted during string formatting` caused by incorrectly passing `FutureWarning` as a second argument to `logger.warning_once()` in this file, introduced in https://git\u2026",
+    "body_excerpt": "# What does this PR do? I had some issues with running `transformers-cli add-new-model-like`. This PR fixes it. Fixes #44661.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44307",
-    "created_at": "2026-02-26T20:09:20Z",
-    "deletions": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44334",
+    "created_at": "2026-02-27T17:13:44Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44307/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44307",
+    "files_url": "https://github.com/huggingface/transformers/pull/44334/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44334",
     "labels": [],
     "merged": true,
-    "number": 44307,
-    "review_comments_count": 2,
+    "number": 44334,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(modeling_attn_mask_utils): remove FutureWarning from logger.warning_once()",
-    "updated_at": "2026-02-26T21:29:01Z"
+    "title": "Fix CookieCutter",
+    "updated_at": "2026-03-13T17:13:28Z"
   },
   {
-    "additions": 10,
-    "author": "jashshah999",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes a crash in `PretrainedConfig.update_from_string()` when the input string contains entries without `=` or with multiple `=` signs. **`configuration_utils.py`** - The existing code `dict(x.split(\"=\") for x in up\u2026",
-    "changed_files": 2,
+    "additions": 13,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Refactors the ALBERT model to use named attribute access instead of index-based access on model outputs, and removes redundant `return_dict=True` arguments from inner model calls (already handled by `@capture_outpu\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44306",
-    "created_at": "2026-02-26T20:02:10Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44333",
+    "created_at": "2026-02-27T17:12:25Z",
+    "deletions": 18,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44306/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44306",
+    "files_url": "https://github.com/huggingface/transformers/pull/44333/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44333",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44306,
+    "number": 44333,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix ValueError crash in PretrainedConfig.update_from_string on malformed input",
-    "updated_at": "2026-02-27T14:25:03Z"
+    "title": "Refactor ALBERT to use named attributes and remove redundant return_dict=True",
+    "updated_at": "2026-03-02T13:05:54Z"
   },
   {
-    "additions": 7,
-    "author": "jashshah999",
+    "additions": 3,
+    "author": "tysoncung",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes a mutable default argument and two resource leaks: 1. **`integrations/tpu.py`** - `patched_optimizer_step` used `optimizer_args={}` as a default parameter. Mutable defaults are shared across calls, so any muta\u2026",
-    "changed_files": 3,
+    "body_excerpt": "Fix minor typos found in comments and docstrings: - `orignal` \u2192 `original` in `src/transformers/integrations/peft.py` (lines 245, 284) - Duplicate word `is is` \u2192 `is` in `src/transformers/models/dia/processing_dia.py` (line 89) Small clean\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44305",
-    "created_at": "2026-02-26T19:22:33Z",
-    "deletions": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44332",
+    "created_at": "2026-02-27T16:11:46Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44305/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44305",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44305,
+    "files_url": "https://github.com/huggingface/transformers/pull/44332/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44332",
+    "labels": [],
+    "merged": true,
+    "number": 44332,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix mutable default in TPU optimizer and unclosed file handles",
-    "updated_at": "2026-02-27T14:21:58Z"
+    "title": "Fix typos in comments and docstrings",
+    "updated_at": "2026-02-27T18:02:59Z"
   },
   {
-    "additions": 151,
-    "author": "adil-a",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Adds NeMo Automodel under the community integrations tab. ## Before submitting - [x] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). - [x] Did you read the [contribut\u2026",
+    "additions": 33,
+    "author": "kashif",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? fixed the bfloat16 dtype mismatch and Loss computation shape mismatch. Also added tests for these. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to ap\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44304",
-    "created_at": "2026-02-26T17:57:16Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44331",
+    "created_at": "2026-02-27T15:46:08Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44304/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44304",
+    "files_url": "https://github.com/huggingface/transformers/pull/44331/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44331",
     "labels": [],
     "merged": true,
-    "number": 44304,
+    "number": 44331,
     "review_comments_count": 9,
     "state": "closed",
-    "title": "docs: Add NeMo Automodel community integration docs",
-    "updated_at": "2026-03-03T16:51:48Z"
+    "title": "[timesfm2_5] fix timesfm2.5 loss",
+    "updated_at": "2026-03-03T17:22:56Z"
   },
   {
-    "additions": 28,
-    "author": "michaelbenayoun",
+    "additions": 289,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Integrate the neuron device to TrainingArguments. It enables using the neuron device with the `Trainer` class.",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? As per the title! Follow-up of https://github.com/huggingface/transformers/pull/44181 with more models!",
+    "changed_files": 136,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44302",
-    "created_at": "2026-02-26T15:11:09Z",
-    "deletions": 0,
+    "comments_count": 12,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44330",
+    "created_at": "2026-02-27T15:33:02Z",
+    "deletions": 1682,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44302/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44302",
+    "files_url": "https://github.com/huggingface/transformers/pull/44330/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44330",
     "labels": [],
     "merged": true,
-    "number": 44302,
-    "review_comments_count": 0,
+    "number": 44330,
+    "review_comments_count": 13,
     "state": "closed",
-    "title": "Integrate the Neuron device to TrainingArguments",
-    "updated_at": "2026-03-05T15:11:00Z"
+    "title": "Remove `cache_position` in more models",
+    "updated_at": "2026-03-11T14:47:50Z"
+  },
+  {
+    "additions": 3,
+    "author": "linfeng-du",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44329",
+    "created_at": "2026-02-27T15:27:39Z",
+    "deletions": 3,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44329/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44329",
+    "labels": [],
+    "merged": true,
+    "number": 44329,
+    "review_comments_count": 4,
+    "state": "closed",
+    "title": "Enable Liger Kernel when doing hyperparameter search.",
+    "updated_at": "2026-03-03T13:44:56Z"
   },
   {
-    "additions": 30,
-    "author": "likejazz",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? When fine-tuning Qwen3 with frameworks like TRL, `<think>` blocks are silently omitted from the token sequence, causing chain-of-thought reasoning data to be completely lost during training with no error or warning\u2026",
+    "additions": 92,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR adds guidelines for agents when it comes to add/run trainer tests. This needs to be updated as we modify, refactor the code !",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44301",
-    "created_at": "2026-02-26T14:30:24Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44328",
+    "created_at": "2026-02-27T15:17:24Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44301/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44301",
+    "files_url": "https://github.com/huggingface/transformers/pull/44328/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44328",
     "labels": [],
-    "merged": false,
-    "number": 44301,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44328,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Fix: Qwen3 `<think>` blocks not written during fine-tuning (TRL)",
-    "updated_at": "2026-03-02T17:18:03Z"
+    "title": "Add testing guide for agents for trainer tests",
+    "updated_at": "2026-02-27T17:32:11Z"
   },
   {
-    "additions": 2539,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? The recursive feature is needed for me in https://github.com/huggingface/transformers/pull/44252 to allow timm backbone define its conversion only once. Also it currently allows to delete \"t5gemma2\" from conversion,\u2026",
-    "changed_files": 18,
+    "additions": 38,
+    "author": "overcastbulb",
+    "author_association": "NONE",
+    "body_excerpt": "Adds missing pipeline tutorial example for zero-shot-classification following the existing format of other task examples in pipeline_tutorial.md. Related: #18926",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 44,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44300",
-    "created_at": "2026-02-26T14:09:59Z",
-    "deletions": 470,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44326",
+    "created_at": "2026-02-27T14:37:19Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44300/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44300",
-    "labels": [],
-    "merged": true,
-    "number": 44300,
-    "review_comments_count": 45,
+    "files_url": "https://github.com/huggingface/transformers/pull/44326/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44326",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44326,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Dynamic weight conversion is recursive",
-    "updated_at": "2026-03-26T11:59:06Z"
+    "title": "docs: Add zero-shot-classification example to pipeline tutorial",
+    "updated_at": "2026-02-27T14:46:24Z"
   },
   {
-    "additions": 520,
-    "author": "vasqu",
+    "additions": 4,
+    "author": "kashif",
     "author_association": "MEMBER",
-    "body_excerpt": "As per title, to allow for proper alignment with vllm/sglang Closes #44258",
-    "changed_files": 21,
+    "body_excerpt": "# What does this PR do? The `CLIPMLP` has the bias set to True but timesFM 2.5 uses `bias=False` in the pretrained model <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appe\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44299",
-    "created_at": "2026-02-26T12:50:03Z",
-    "deletions": 282,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44325",
+    "created_at": "2026-02-27T13:18:40Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44299/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44299",
+    "files_url": "https://github.com/huggingface/transformers/pull/44325/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44325",
     "labels": [],
     "merged": true,
-    "number": 44299,
-    "review_comments_count": 4,
+    "number": 44325,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": ":rotating_light: [`Ernie 4.5 VL Moe`] Fix up namings to vllm/sglang convention",
-    "updated_at": "2026-02-26T16:42:50Z"
+    "title": "[timesfm2_5] fix timesfm mlp bias",
+    "updated_at": "2026-02-27T13:36:13Z"
   },
   {
-    "additions": 1145,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? A few issues we did not catch: - https://github.com/huggingface/transformers/blob/47b0e478f324b54f177ea7998a0791870fdd0324/src/transformers/convert_slow_tokenizer.py#L1314-L1315 missing from `GemmaTokenier` - SPM's\u2026",
-    "changed_files": 7,
+    "additions": 16,
+    "author": "tonglei19961121",
+    "author_association": "NONE",
+    "body_excerpt": "Add tutorial example for DocumentQuestionAnswering pipeline following the existing format of other task examples. This PR addresses the first TODO item in #18926. Changes: - Added document-question-answering task example to pipeline_tutori\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44298",
-    "created_at": "2026-02-26T12:34:38Z",
-    "deletions": 29,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44298/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44298",
-    "labels": [],
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44324",
+    "created_at": "2026-02-27T12:29:06Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44324/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44324",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44298,
-    "review_comments_count": 1,
-    "state": "open",
-    "title": "Auto detect wrong mapping models",
-    "updated_at": "2026-03-02T10:13:28Z"
+    "number": 44324,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "docs: Add document-question-answering example to pipeline tutorial",
+    "updated_at": "2026-02-27T14:35:23Z"
   },
   {
-    "additions": 12302,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Needed for https://github.com/huggingface/transformers/pull/41250 to pass the docstring-checker in CI. Our basic checker doesn't handle well dataclasses so we can use `autodocstring`",
-    "changed_files": 512,
+    "additions": 16,
+    "author": "tonglei19961121",
+    "author_association": "NONE",
+    "body_excerpt": "Add tutorial example for DocumentQuestionAnswering pipeline following the existing format of other task examples. Fixes #18926",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44296",
-    "created_at": "2026-02-26T10:29:21Z",
-    "deletions": 37860,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44323",
+    "created_at": "2026-02-27T12:26:00Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44296/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44296",
-    "labels": [],
-    "merged": true,
-    "number": 44296,
-    "review_comments_count": 11,
+    "files_url": "https://github.com/huggingface/transformers/pull/44323/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44323",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44323,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add auto-docstring on configs",
-    "updated_at": "2026-03-06T11:58:10Z"
+    "title": "docs: Add document-question-answering example to pipeline tutorial",
+    "updated_at": "2026-02-27T14:38:43Z"
   },
   {
-    "additions": 1,
-    "author": "mario-sanz",
+    "additions": 12,
+    "author": "kaixuanliu",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "body_excerpt": "@vasqu This PR skipped 2 invalid test cases: ``` tests/models/voxtral_realtime/test_modeling_voxtral_realtime.py::VoxtralRealtimeForConditionalGenerationModelTest::test_generate_with_quant_cache tests/models/voxtral_realtime/test_modeling_\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44294",
-    "created_at": "2026-02-26T08:30:52Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44321",
+    "created_at": "2026-02-27T09:54:14Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44294/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44294",
+    "files_url": "https://github.com/huggingface/transformers/pull/44321/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44321",
     "labels": [],
     "merged": true,
-    "number": 44294,
+    "number": 44321,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix: use `TokenizersBackend` for Olmo3 to preserve custom `pre_tokenizer`",
-    "updated_at": "2026-02-26T10:35:44Z"
+    "title": "skip 2 invalid test cases for voxtral_realtime model",
+    "updated_at": "2026-04-02T03:22:04Z"
   },
   {
-    "additions": 13,
-    "author": "IlyasMoutawwakil",
+    "additions": 5231,
+    "author": "NielsRogge",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 15,
+    "body_excerpt": "# What does this PR do? This PR adds [SAM3-LiteText: An Anatomical Study of the SAM3 Text Encoder for Efficient Vision-Language Segmentation](https://huggingface.co/papers/2602.12173). Fixes #44205",
+    "changed_files": 22,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44293",
-    "created_at": "2026-02-26T08:25:23Z",
-    "deletions": 384,
+    "comments_count": 13,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44320",
+    "created_at": "2026-02-27T08:29:00Z",
+    "deletions": 19,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44293/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44293",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44320/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44320",
+    "labels": [
+      "New model"
+    ],
     "merged": true,
-    "number": 44293,
-    "review_comments_count": 7,
+    "number": 44320,
+    "review_comments_count": 63,
     "state": "closed",
-    "title": "Strict export cleanup",
-    "updated_at": "2026-03-02T09:36:19Z"
+    "title": "Add SAM3-LiteText",
+    "updated_at": "2026-04-13T18:37:07Z"
   },
   {
-    "additions": 548,
-    "author": "stevhliu",
+    "additions": 74,
+    "author": "IlyasMoutawwakil",
     "author_association": "MEMBER",
-    "body_excerpt": "- adds `trainer_recipes.md` to show how to use other practical `Trainer` features outside of the basic training loop - updates hyperparam search docs - updates `optimizers.md` with how to customize it (prebuilt instances, passing a class +\u2026",
-    "changed_files": 6,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44290",
-    "created_at": "2026-02-26T01:02:15Z",
-    "deletions": 210,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44319",
+    "created_at": "2026-02-27T08:20:45Z",
+    "deletions": 56,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44290/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44290",
+    "files_url": "https://github.com/huggingface/transformers/pull/44319/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44319",
     "labels": [],
     "merged": true,
-    "number": 44290,
-    "review_comments_count": 17,
+    "number": 44319,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "[docs] optimizers, hyperparam search, training features",
-    "updated_at": "2026-04-09T20:00:53Z"
+    "title": "Support non-gated experts",
+    "updated_at": "2026-03-02T19:26:38Z"
   },
   {
-    "additions": 8,
-    "author": "haosenwang1018",
+    "additions": 10,
+    "author": "yoginlangalia",
     "author_association": "NONE",
-    "body_excerpt": "Replace bare except clauses with except Exception.",
-    "changed_files": 4,
+    "body_excerpt": "## What does this PR do? Adds input validation for bounding box shape in `LayoutLMv3Tokenizer`. When users pass boxes with fewer (or more) than 4 values per box, the tokenizer now raises a clear `ValueError` instead of a confusing generic\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44289",
-    "created_at": "2026-02-26T00:58:35Z",
-    "deletions": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44318",
+    "created_at": "2026-02-27T06:40:02Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44289/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44289",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44318/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44318",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44289,
+    "number": 44318,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: replace 8 bare except clauses with except Exception",
-    "updated_at": "2026-02-26T12:52:48Z"
+    "title": "Validate bounding box shape in LayoutLMv3Tokenizer",
+    "updated_at": "2026-02-27T14:43:08Z"
   },
   {
-    "additions": 1,
-    "author": "somAzzz",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "# What does this PR do? When loading the new Qwen 3.5 models (e.g., 'Qwen/Qwen3.5-35B-A3B') using the 'transformers' (5.3.0.dev0), the initialization crashes with a 'TypeError' . **Error Traceback Context:** (APIServer pid=98544) File \"...\u2026",
-    "changed_files": 1,
+    "additions": 4,
+    "author": "sxu75374",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Passes `config.rms_norm_eps` explicitly to `q_a_layernorm` and `kv_a_layernorm` in both DeepSeek V2 and V3 MLA attention. Currently these two norms are constructed without `eps`, falling back to the `RMSNorm` class\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44288",
-    "created_at": "2026-02-25T22:43:41Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44317",
+    "created_at": "2026-02-27T04:48:08Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44288/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44288",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44317/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44317",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44288,
-    "review_comments_count": 1,
+    "number": 44317,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix TypeError in modeling_rope_utils.py when ignore_keys_at_rope_vali\u2026",
-    "updated_at": "2026-02-26T20:09:07Z"
+    "title": "fix(deepseek): pass rms_norm_eps to MLA q/kv layernorms",
+    "updated_at": "2026-02-27T14:30:04Z"
   },
   {
-    "additions": 11,
+    "additions": 2,
     "author": "jashshah999",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Fixes mutable default arguments and unclosed file handles across several files. **Mutable defaults** (can cause shared state across calls): - `debug_utils.py`: `DebugUnderflowOverflow.__init__` `trace_batch_nums=[]`\u2026",
-    "changed_files": 4,
+    "body_excerpt": "Fixes #44303 The weight loading progress bar called `pbar.refresh()` on every single parameter, bypassing tqdm's built-in rate-limiting. When output is redirected to a log file (e.g. in CI), this produced one line per parameter -- hundreds\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44287",
-    "created_at": "2026-02-25T22:23:20Z",
-    "deletions": 7,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44316",
+    "created_at": "2026-02-27T03:08:28Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44287/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44287",
+    "files_url": "https://github.com/huggingface/transformers/pull/44316/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44316",
     "labels": [],
-    "merged": true,
-    "number": 44287,
+    "merged": false,
+    "number": 44316,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix mutable default arguments and resource leaks",
-    "updated_at": "2026-03-02T15:17:25Z"
+    "title": "Reduce tqdm verbosity during weight loading",
+    "updated_at": "2026-03-03T17:02:34Z"
   },
   {
-    "additions": 31,
-    "author": "kathrynle20",
+    "additions": 3484,
+    "author": "jp1924",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds on to the Fouroversix integration by adding support for quantized models such as the gpt-oss model by adding weight conversions and an additional config argument. Reference: https://github.com/huggingfa\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Hello, Transformers team! I submitted a PR to add naver-hyperclovax/HyperCLOVAX-SEED-Think-32B (hereafter HCX), developed by the Korean IT company Naver while executing the government's national AI model project. Th\u2026",
+    "changed_files": 24,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44286",
-    "created_at": "2026-02-25T22:15:15Z",
-    "deletions": 14,
+    "comments_count": 22,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44314",
+    "created_at": "2026-02-27T02:01:28Z",
+    "deletions": 18,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44286/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44286",
+    "files_url": "https://github.com/huggingface/transformers/pull/44314/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44314",
     "labels": [],
-    "merged": true,
-    "number": 44286,
-    "review_comments_count": 12,
-    "state": "closed",
-    "title": "Add future model support for Fouroversix",
-    "updated_at": "2026-03-04T16:28:13Z"
-  },
-  {
-    "additions": 3484,
-    "author": "NielsRogge",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds the VidEoMT model, as described in [VidEoMT: Your ViT is Secretly Also a Video Segmentation Model](https://huggingface.co/papers/2602.17807). Gradio demo (running on ZeroGPU): https://huggingface.co/spa\u2026",
-    "changed_files": 17,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 23,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44285",
-    "created_at": "2026-02-25T19:24:39Z",
-    "deletions": 2,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44285/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44285",
-    "labels": [
-      "New model"
-    ],
-    "merged": true,
-    "number": 44285,
-    "review_comments_count": 57,
-    "state": "closed",
-    "title": "Add VidEoMT",
-    "updated_at": "2026-03-25T17:05:47Z"
+    "merged": false,
+    "number": 44314,
+    "review_comments_count": 77,
+    "state": "open",
+    "title": "add HyperClovaX Vision",
+    "updated_at": "2026-04-13T02:23:53Z"
   },
   {
-    "additions": 388,
-    "author": "paipeline",
-    "author_association": "NONE",
-    "body_excerpt": "## Description Fixes #44242 where Mixtral models do not compute auxiliary load balancing loss when `output_router_logits=False`, even when `router_aux_loss_coef > 0`. ## Problem According to the [Mixtral documentation](https://huggingface.\u2026",
-    "changed_files": 4,
+    "additions": 4,
+    "author": "jashshah999",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fixes the same `TypeError: AddedToken() got multiple values for keyword argument 'special'` that #44281 addressed, but for the `extra_special_tokens` code path which was missed. #44281 (commit 8e663c7) correctly added `value.pop(\"special\",\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44284",
-    "created_at": "2026-02-25T18:38:15Z",
-    "deletions": 3,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44313",
+    "created_at": "2026-02-27T01:37:45Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44284/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44284",
+    "files_url": "https://github.com/huggingface/transformers/pull/44313/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44313",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44284,
+    "number": 44313,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix Mixtral auxiliary loss computation when output_router_logits=False",
-    "updated_at": "2026-02-26T12:41:46Z"
+    "title": "Fix AddedToken duplicate 'special' kwarg for extra_special_tokens",
+    "updated_at": "2026-02-27T14:26:28Z"
   },
   {
-    "additions": 1,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "Accidentally caused by #43325, wrong naming --> modular doesn't properly convert some files (e.g. kyutai) Also fixes red CI on main",
-    "changed_files": 1,
+    "additions": 8,
+    "author": "haosenwang1018",
+    "author_association": "NONE",
+    "body_excerpt": "Replace bare `except:` clauses with `except Exception:` for PEP 8 compliance.",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44283",
-    "created_at": "2026-02-25T18:33:17Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44312",
+    "created_at": "2026-02-27T01:00:33Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44283/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44283",
+    "files_url": "https://github.com/huggingface/transformers/pull/44312/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44312",
     "labels": [],
-    "merged": true,
-    "number": 44283,
+    "merged": false,
+    "number": 44312,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[`Modular`] Fix file type regression",
-    "updated_at": "2026-02-25T20:04:41Z"
+    "title": "fix: replace 8 bare except clauses with except Exception",
+    "updated_at": "2026-02-27T03:27:27Z"
   },
   {
-    "additions": 5,
-    "author": "Rocketknight1",
-    "author_association": "MEMBER",
-    "body_excerpt": "Response schema save-loading was broken in #40936, this PR restores it! I did most of this in #42300 but missed an issue with loading/saving.",
-    "changed_files": 1,
+    "additions": 38,
+    "author": "onel",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Adds species bias documentation across the transformers repository to help model authors and users recognize and address potential biases in language models. The updates include guidance on documenting bias categori\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44282",
-    "created_at": "2026-02-25T17:57:54Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44311",
+    "created_at": "2026-02-27T00:02:49Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44282/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44282",
+    "files_url": "https://github.com/huggingface/transformers/pull/44311/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44311",
     "labels": [],
-    "merged": true,
-    "number": 44282,
+    "merged": false,
+    "number": 44311,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Restore response_schema saving-loading",
-    "updated_at": "2026-02-25T18:27:22Z"
+    "title": "Add species bias documentation to model cards and docs",
+    "updated_at": "2026-02-27T14:09:20Z"
   },
   {
-    "additions": 1,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Its a very small fix for #44062",
-    "changed_files": 1,
+    "additions": 63,
+    "author": "onel",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Adds documentation for Pruna AI integration to the Transformers ecosystem, following the existing pattern used by vLLM and Unsloth integration docs. ## Changes - Created `docs/source/en/community_integrations/pruna.\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44281",
-    "created_at": "2026-02-25T16:28:37Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44310",
+    "created_at": "2026-02-27T00:00:31Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44281/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44281",
-    "labels": [],
-    "merged": true,
-    "number": 44281,
+    "files_url": "https://github.com/huggingface/transformers/pull/44310/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44310",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44310,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix special token maps BC",
-    "updated_at": "2026-02-26T10:34:17Z"
+    "title": "docs: Add Pruna AI integration documentation",
+    "updated_at": "2026-02-27T14:08:21Z"
   },
   {
-    "additions": 614,
-    "author": "RishabhMehra",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "# What does this PR do? - Adds an opt-in use_fast_grouping flag to TokenClassificationPipeline to enable a NumPy-vectorised BIO grouping path (~5\u00d7 faster on long sequences) while keeping the legacy path as default. - Improves correctness:\u2026",
-    "changed_files": 3,
+    "additions": 129,
+    "author": "onel",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Adds SkyPilot deployment documentation to the DeepSpeed guide. The new section includes: - Introduction to SkyPilot as a unified framework for running AI workloads across clouds and Kubernetes - Complete example YAM\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44278",
-    "created_at": "2026-02-25T12:49:56Z",
-    "deletions": 63,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44309",
+    "created_at": "2026-02-26T22:44:41Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44278/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44278",
+    "files_url": "https://github.com/huggingface/transformers/pull/44309/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44309",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44278,
+    "number": 44309,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[FEAT] Pipelines - Faster group_entities",
-    "updated_at": "2026-02-25T13:54:58Z"
+    "title": "Add SkyPilot deployment documentation to DeepSpeed guide",
+    "updated_at": "2026-02-27T14:09:55Z"
   },
   {
-    "additions": 105,
-    "author": "tarekziade",
+    "additions": 5854,
+    "author": "NielsRogge",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This patch makes the GLM-ASR doc example runnable by using `runnables` - see https://github.com/huggingface/doc-builder/blob/main/docs/runnable-code-blocks.md",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 36,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44277",
-    "created_at": "2026-02-25T08:49:20Z",
-    "deletions": 19,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44308",
+    "created_at": "2026-02-26T21:26:05Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44277/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44277",
+    "files_url": "https://github.com/huggingface/transformers/pull/44308/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44308",
     "labels": [],
-    "merged": true,
-    "number": 44277,
-    "review_comments_count": 6,
+    "merged": false,
+    "number": 44308,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Use doc-builder runnable example for GLM-ASR",
-    "updated_at": "2026-04-02T16:16:55Z"
+    "title": "Codex/add sam3 litetext model to transformers fuvllg",
+    "updated_at": "2026-02-26T21:35:44Z"
   },
   {
-    "additions": 0,
-    "author": "vishalpatil-45",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? This PR addresses the performance regression where `import transformers` takes ~3.5s. The issue was caused by eager imports of heavy backend libraries (like torch/numpy) during the initial module load. By moving the\u2026",
-    "changed_files": 0,
+    "additions": 7,
+    "author": "imstevenpmwork",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR fixes a `TypeError: not all arguments converted during string formatting` caused by incorrectly passing `FutureWarning` as a second argument to `logger.warning_once()` in this file, introduced in https://git\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44275",
-    "created_at": "2026-02-25T08:27:32Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44307",
+    "created_at": "2026-02-26T20:09:20Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44275/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44275",
+    "files_url": "https://github.com/huggingface/transformers/pull/44307/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44307",
+    "labels": [],
+    "merged": true,
+    "number": 44307,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "fix(modeling_attn_mask_utils): remove FutureWarning from logger.warning_once()",
+    "updated_at": "2026-02-26T21:29:01Z"
+  },
+  {
+    "additions": 10,
+    "author": "jashshah999",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes a crash in `PretrainedConfig.update_from_string()` when the input string contains entries without `=` or with multiple `=` signs. **`configuration_utils.py`** - The existing code `dict(x.split(\"=\") for x in up\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44306",
+    "created_at": "2026-02-26T20:02:10Z",
+    "deletions": 2,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44306/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44306",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44275,
+    "number": 44306,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[Fix] Restore lazy loading to improve import performance (#44273)",
-    "updated_at": "2026-02-25T20:37:18Z"
+    "title": "Fix ValueError crash in PretrainedConfig.update_from_string on malformed input",
+    "updated_at": "2026-02-27T14:25:03Z"
   },
   {
-    "additions": 559,
-    "author": "paipeline",
-    "author_association": "NONE",
-    "body_excerpt": "## Description Fixes #44242 This PR resolves an issue where the auxiliary load balancing loss was not computed when `output_router_logits=False`, even when `router_aux_loss_coef != 0`. ## Problem The auxiliary loss computation was incorrec\u2026",
-    "changed_files": 6,
+    "additions": 7,
+    "author": "jashshah999",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes a mutable default argument and two resource leaks: 1. **`integrations/tpu.py`** - `patched_optimizer_step` used `optimizer_args={}` as a default parameter. Mutable defaults are shared across calls, so any muta\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44274",
-    "created_at": "2026-02-25T06:38:02Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44305",
+    "created_at": "2026-02-26T19:22:33Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44274/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44274",
+    "files_url": "https://github.com/huggingface/transformers/pull/44305/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44305",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44274,
+    "number": 44305,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix auxiliary load balancing loss computation when output_router_logits=False",
-    "updated_at": "2026-02-25T13:36:03Z"
+    "title": "Fix mutable default in TPU optimizer and unclosed file handles",
+    "updated_at": "2026-02-27T14:21:58Z"
   },
   {
-    "additions": 1,
-    "author": "hangjun-ezra",
+    "additions": 151,
+    "author": "adil-a",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes a `TypeError: unsupported operand type(s) for |: 'list' and 'set'` in `RotaryEmbeddingConfigMixin.convert_rope_params_to_dict` when `ignore_keys_at_rope_validation` is a `list` instead of a `set`. ### Root ca\u2026",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Adds NeMo Automodel under the community integrations tab. ## Before submitting - [x] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). - [x] Did you read the [contribut\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44272",
-    "created_at": "2026-02-25T03:52:04Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44304",
+    "created_at": "2026-02-26T17:57:16Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44272/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44272",
+    "files_url": "https://github.com/huggingface/transformers/pull/44304/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44304",
     "labels": [],
     "merged": true,
-    "number": 44272,
-    "review_comments_count": 0,
+    "number": 44304,
+    "review_comments_count": 9,
     "state": "closed",
-    "title": "Fix TypeError in convert_rope_params_to_dict when ignore_keys is a list",
-    "updated_at": "2026-02-25T14:38:36Z"
+    "title": "docs: Add NeMo Automodel community integration docs",
+    "updated_at": "2026-03-03T16:51:48Z"
   },
   {
-    "additions": 1272,
-    "author": "balak4",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary - Add GreedyLR, a metric-based adaptive learning rate scheduler that adjusts the learning rate during training based on the current loss - Based on [\"Dynamic Learning Rate Scheduling based on Loss Changes Leads to Faster Converg\u2026",
-    "changed_files": 10,
+    "additions": 28,
+    "author": "michaelbenayoun",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Integrate the neuron device to TrainingArguments. It enables using the neuron device with the `Trainer` class.",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44271",
-    "created_at": "2026-02-25T01:40:57Z",
-    "deletions": 7,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44302",
+    "created_at": "2026-02-26T15:11:09Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44271/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44271",
+    "files_url": "https://github.com/huggingface/transformers/pull/44302/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44302",
     "labels": [],
     "merged": true,
-    "number": 44271,
-    "review_comments_count": 3,
+    "number": 44302,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add GreedyLR adaptive learning rate scheduler",
-    "updated_at": "2026-03-18T18:45:46Z"
+    "title": "Integrate the Neuron device to TrainingArguments",
+    "updated_at": "2026-03-05T15:11:00Z"
   },
   {
-    "additions": 88,
-    "author": "yonigozlan",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? A lot of ProcessorsKwargs have incorrect/unspecified type hints in their ProcessorsKwargs TypedDict for their images_kwargs attribute. Functionnaly, this did not cause issues as \"_merge_kwargs\" automatically picks u\u2026",
-    "changed_files": 44,
+    "additions": 30,
+    "author": "likejazz",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? When fine-tuning Qwen3 with frameworks like TRL, `<think>` blocks are silently omitted from the token sequence, causing chain-of-thought reasoning data to be completely lost during training with no error or warning\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44270",
-    "created_at": "2026-02-25T00:11:31Z",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44301",
+    "created_at": "2026-02-26T14:30:24Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44270/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44270",
+    "files_url": "https://github.com/huggingface/transformers/pull/44301/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44301",
     "labels": [],
     "merged": false,
-    "number": 44270,
+    "number": 44301,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Add correct typing to custom images_kwargs in ProcessorsKwargs",
-    "updated_at": "2026-02-25T01:12:06Z"
+    "state": "closed",
+    "title": "Fix: Qwen3 `<think>` blocks not written during fine-tuning (TRL)",
+    "updated_at": "2026-03-02T17:18:03Z"
   },
   {
-    "additions": 30,
-    "author": "yonigozlan",
+    "additions": 2539,
+    "author": "zucchini-nlp",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This is a follow-up to https://github.com/huggingface/transformers/pull/43748, and will allow to have clickable links to the full modality kwargs when present in the docstring of a processor or image processor Cc @s\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? The recursive feature is needed for me in https://github.com/huggingface/transformers/pull/44252 to allow timm backbone define its conversion only once. Also it currently allows to delete \"t5gemma2\" from conversion,\u2026",
+    "changed_files": 18,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44269",
-    "created_at": "2026-02-25T00:05:47Z",
-    "deletions": 2,
+    "comments_count": 44,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44300",
+    "created_at": "2026-02-26T14:09:59Z",
+    "deletions": 470,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44269/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44269",
+    "files_url": "https://github.com/huggingface/transformers/pull/44300/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44300",
+    "labels": [],
+    "merged": true,
+    "number": 44300,
+    "review_comments_count": 45,
+    "state": "closed",
+    "title": "Dynamic weight conversion is recursive",
+    "updated_at": "2026-03-26T11:59:06Z"
+  },
+  {
+    "additions": 520,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "As per title, to allow for proper alignment with vllm/sglang Closes #44258",
+    "changed_files": 21,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44299",
+    "created_at": "2026-02-26T12:50:03Z",
+    "deletions": 282,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44299/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44299",
     "labels": [],
     "merged": true,
-    "number": 44269,
-    "review_comments_count": 0,
+    "number": 44299,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Add `ProcessingKwargs` `ImagesKwargs` etc. to docs",
-    "updated_at": "2026-02-27T19:03:15Z"
+    "title": ":rotating_light: [`Ernie 4.5 VL Moe`] Fix up namings to vllm/sglang convention",
+    "updated_at": "2026-02-26T16:42:50Z"
   },
   {
-    "additions": 5,
-    "author": "ethanknights",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Some improvements to the `trainer.py` docs. ## Before submitting - [x] This PR fixes a typo or improves the docs. ## Who can review? Documentation: @stevhliu",
-    "changed_files": 1,
+    "additions": 1145,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? A few issues we did not catch: - https://github.com/huggingface/transformers/blob/47b0e478f324b54f177ea7998a0791870fdd0324/src/transformers/convert_slow_tokenizer.py#L1314-L1315 missing from `GemmaTokenier` - SPM's\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44268",
-    "created_at": "2026-02-24T23:20:16Z",
-    "deletions": 4,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44268/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44268",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44298",
+    "created_at": "2026-02-26T12:34:38Z",
+    "deletions": 29,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44298/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44298",
     "labels": [],
-    "merged": true,
-    "number": 44268,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "chore: fixes in `Trainer` class docs (`compute_loss` & `hyperparameter_search`)",
-    "updated_at": "2026-02-26T00:50:23Z"
+    "merged": false,
+    "number": 44298,
+    "review_comments_count": 1,
+    "state": "open",
+    "title": "Auto detect wrong mapping models",
+    "updated_at": "2026-03-02T10:13:28Z"
   },
   {
-    "additions": 4,
-    "author": "manavshrivastavagit",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary - Update the `DocumentQuestionAnsweringPipeline` docstring to explicitly mention the task summary in the Transformers documentation. - Remove the stale TODO comment now that document question answering is covered in the task sum\u2026",
-    "changed_files": 1,
+    "additions": 12302,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Needed for https://github.com/huggingface/transformers/pull/41250 to pass the docstring-checker in CI. Our basic checker doesn't handle well dataclasses so we can use `autodocstring`",
+    "changed_files": 512,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44267",
-    "created_at": "2026-02-24T20:35:18Z",
-    "deletions": 4,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44296",
+    "created_at": "2026-02-26T10:29:21Z",
+    "deletions": 37860,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44267/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44267",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44267,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44296/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44296",
+    "labels": [],
+    "merged": true,
+    "number": 44296,
+    "review_comments_count": 11,
     "state": "closed",
-    "title": "Docs: point DocumentQuestionAnswering pipeline to task summary",
-    "updated_at": "2026-02-25T13:34:48Z"
+    "title": "Add auto-docstring on configs",
+    "updated_at": "2026-03-06T11:58:10Z"
   },
   {
-    "additions": 27,
-    "author": "harshaljanjani",
+    "additions": 1,
+    "author": "mario-sanz",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following issue was identified and fixed in this PR: \u2192 **Reasoning:** The impact of this fix goes beyond `Mask2Former` and `DeformableDetr` and should fix any model that uses `torch_compilable_check`. Most use\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44266",
-    "created_at": "2026-02-24T20:02:06Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44294",
+    "created_at": "2026-02-26T08:30:52Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44266/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44266",
+    "files_url": "https://github.com/huggingface/transformers/pull/44294/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44294",
     "labels": [],
     "merged": true,
-    "number": 44266,
+    "number": 44294,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(utils): Make torch_compilable_check compatible with torch.export strict mode",
-    "updated_at": "2026-02-26T09:42:47Z"
+    "title": "Fix: use `TokenizersBackend` for Olmo3 to preserve custom `pre_tokenizer`",
+    "updated_at": "2026-02-26T10:35:44Z"
   },
   {
-    "additions": 90,
-    "author": "vasqu",
+    "additions": 13,
+    "author": "IlyasMoutawwakil",
     "author_association": "MEMBER",
-    "body_excerpt": "As per title, WIP --> needs a test",
-    "changed_files": 36,
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44264",
-    "created_at": "2026-02-24T18:06:58Z",
-    "deletions": 210,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44264/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44264",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44293",
+    "created_at": "2026-02-26T08:25:23Z",
+    "deletions": 384,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44293/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44293",
     "labels": [],
-    "merged": false,
-    "number": 44264,
-    "review_comments_count": 3,
-    "state": "open",
-    "title": "[`Moe`] Enable aux loss automatically when in training + coef is not 0",
-    "updated_at": "2026-02-25T18:53:20Z"
+    "merged": true,
+    "number": 44293,
+    "review_comments_count": 7,
+    "state": "closed",
+    "title": "Strict export cleanup",
+    "updated_at": "2026-03-02T09:36:19Z"
   },
   {
-    "additions": 5882,
-    "author": "SunMarc",
+    "additions": 548,
+    "author": "stevhliu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR refactor the common tests that we have in Trainer. I've mainly did the following: - Split the tests that we have in `test_trainer.py` into multiple files. - Fix common tests that were failing in the CI",
-    "changed_files": 18,
+    "body_excerpt": "- adds `trainer_recipes.md` to show how to use other practical `Trainer` features outside of the basic training loop - updates hyperparam search docs - updates `optimizers.md` with how to customize it (prebuilt instances, passing a class +\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44260",
-    "created_at": "2026-02-24T15:51:11Z",
-    "deletions": 6147,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44290",
+    "created_at": "2026-02-26T01:02:15Z",
+    "deletions": 210,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44260/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44260",
+    "files_url": "https://github.com/huggingface/transformers/pull/44290/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44290",
     "labels": [],
     "merged": true,
-    "number": 44260,
-    "review_comments_count": 3,
+    "number": 44290,
+    "review_comments_count": 17,
     "state": "closed",
-    "title": "Update common tests Trainer",
-    "updated_at": "2026-02-27T17:31:59Z"
+    "title": "[docs] optimizers, hyperparam search, training features",
+    "updated_at": "2026-04-09T20:00:53Z"
   },
   {
-    "additions": 1830,
-    "author": "winglian",
-    "author_association": "COLLABORATOR",
-    "body_excerpt": "# What does this PR do? This PR supersedes #43985 to replace the dataset/sampler/dataloader with a data producer that should allow us to more easily get to the next step of async training for RL. <!-- Congratulations! You've made it this f\u2026",
-    "changed_files": 6,
+    "additions": 8,
+    "author": "haosenwang1018",
+    "author_association": "NONE",
+    "body_excerpt": "Replace bare except clauses with except Exception.",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44259",
-    "created_at": "2026-02-24T15:01:56Z",
-    "deletions": 59,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44259/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44259",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44289",
+    "created_at": "2026-02-26T00:58:35Z",
+    "deletions": 8,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44289/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44289",
     "labels": [],
     "merged": false,
-    "number": 44259,
+    "number": 44289,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Async data producer",
-    "updated_at": "2026-02-26T19:57:43Z"
+    "state": "closed",
+    "title": "fix: replace 8 bare except clauses with except Exception",
+    "updated_at": "2026-02-26T12:52:48Z"
   },
   {
-    "additions": 8,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "The old one has a merge conflict and it was easier to just mirror into a new branch / PR Note that this only affects the local big tests that I used to run on the A100s locally; not to be run with the CI (too big)",
-    "changed_files": 2,
+    "additions": 1,
+    "author": "somAzzz",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "# What does this PR do? When loading the new Qwen 3.5 models (e.g., 'Qwen/Qwen3.5-35B-A3B') using the 'transformers' (5.3.0.dev0), the initialization crashes with a 'TypeError' . **Error Traceback Context:** (APIServer pid=98544) File \"...\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44258",
-    "created_at": "2026-02-24T15:00:29Z",
-    "deletions": 8,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44288",
+    "created_at": "2026-02-25T22:43:41Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44258/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44258",
+    "files_url": "https://github.com/huggingface/transformers/pull/44288/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44288",
     "labels": [],
     "merged": false,
-    "number": 44258,
-    "review_comments_count": 0,
+    "number": 44288,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "[`Ernie 4.5 VL Moe`] Change revision",
-    "updated_at": "2026-03-14T19:59:05Z"
+    "title": "Fix TypeError in modeling_rope_utils.py when ignore_keys_at_rope_vali\u2026",
+    "updated_at": "2026-02-26T20:09:07Z"
   },
   {
-    "additions": 3,
-    "author": "winglian",
-    "author_association": "COLLABORATOR",
-    "body_excerpt": "# What does this PR do? When post training using context parallelism, some processes may have their chunk of the sample input masked out leading to a NaN loss for that process. Using `nanmean` allows us to keep the real loss that isn't `Na\u2026",
-    "changed_files": 1,
+    "additions": 11,
+    "author": "jashshah999",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Fixes mutable default arguments and unclosed file handles across several files. **Mutable defaults** (can cause shared state across calls): - `debug_utils.py`: `DebugUnderflowOverflow.__init__` `trace_batch_nums=[]`\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44257",
-    "created_at": "2026-02-24T14:56:42Z",
-    "deletions": 3,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44287",
+    "created_at": "2026-02-25T22:23:20Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44257/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44257",
+    "files_url": "https://github.com/huggingface/transformers/pull/44287/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44287",
     "labels": [],
-    "merged": false,
-    "number": 44257,
-    "review_comments_count": 5,
-    "state": "open",
-    "title": "use nanmean for aggregating loss",
-    "updated_at": "2026-02-25T17:01:08Z"
+    "merged": true,
+    "number": 44287,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix mutable default arguments and resource leaks",
+    "updated_at": "2026-03-02T15:17:25Z"
   },
   {
-    "additions": 10,
-    "author": "albertvillanova",
-    "author_association": "MEMBER",
-    "body_excerpt": "Fix CLI NameError: name 'TypeAdapter' is not defined: - Do not evaluate type annotations in CLI serve ### Problem Calling the CLI raises NameError: > NameError: name 'TypeAdapter' is not defined ```bash transformers --help ``` ```python Tr\u2026",
-    "changed_files": 1,
+    "additions": 31,
+    "author": "kathrynle20",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds on to the Fouroversix integration by adding support for quantized models such as the gpt-oss model by adding weight conversions and an additional config argument. Reference: https://github.com/huggingfa\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44256",
-    "created_at": "2026-02-24T14:54:49Z",
-    "deletions": 9,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44286",
+    "created_at": "2026-02-25T22:15:15Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44256/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44256",
+    "files_url": "https://github.com/huggingface/transformers/pull/44286/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44286",
     "labels": [],
     "merged": true,
-    "number": 44256,
-    "review_comments_count": 0,
+    "number": 44286,
+    "review_comments_count": 12,
     "state": "closed",
-    "title": "Fix CLI NameError: name 'TypeAdapter' is not defined",
-    "updated_at": "2026-03-12T07:57:14Z"
+    "title": "Add future model support for Fouroversix",
+    "updated_at": "2026-03-04T16:28:13Z"
   },
   {
-    "additions": 404,
-    "author": "itazap",
+    "additions": 3484,
+    "author": "NielsRogge",
     "author_association": "MEMBER",
-    "body_excerpt": "## What this PR does Given he different issues that were noticed by @hmellor on vLLM, we wanted to make sure we did not end up with crazy breaks. We ran a full test suite (code can be found in #44298) and the results showed 22 model conver\u2026",
-    "changed_files": 24,
+    "body_excerpt": "# What does this PR do? This PR adds the VidEoMT model, as described in [VidEoMT: Your ViT is Secretly Also a Video Segmentation Model](https://huggingface.co/papers/2602.17807). Gradio demo (running on ZeroGPU): https://huggingface.co/spa\u2026",
+    "changed_files": 17,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44255",
-    "created_at": "2026-02-24T14:17:00Z",
-    "deletions": 205,
+    "comments_count": 23,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44285",
+    "created_at": "2026-02-25T19:24:39Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44255/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44255",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44285/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44285",
+    "labels": [
+      "New model"
+    ],
     "merged": true,
-    "number": 44255,
-    "review_comments_count": 25,
+    "number": 44285,
+    "review_comments_count": 57,
     "state": "closed",
-    "title": "[vllm + v5 fix] handle TokenizersBackend fallback properly for v5",
-    "updated_at": "2026-03-23T11:07:37Z"
+    "title": "Add VidEoMT",
+    "updated_at": "2026-03-25T17:05:47Z"
   },
   {
-    "additions": 16,
-    "author": "mario-sanz",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully refle\u2026",
-    "changed_files": 2,
+    "additions": 388,
+    "author": "paipeline",
+    "author_association": "NONE",
+    "body_excerpt": "## Description Fixes #44242 where Mixtral models do not compute auxiliary load balancing loss when `output_router_logits=False`, even when `router_aux_loss_coef > 0`. ## Problem According to the [Mixtral documentation](https://huggingface.\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44254",
-    "created_at": "2026-02-24T13:54:30Z",
-    "deletions": 13,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44284",
+    "created_at": "2026-02-25T18:38:15Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44254/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44254",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44284/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44284",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44254,
-    "review_comments_count": 5,
+    "number": 44284,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix fast tokenizers overwriting custom `pre_tokenizer` from `tokenizer.json`",
-    "updated_at": "2026-02-26T08:45:56Z"
+    "title": "Fix Mixtral auxiliary loss computation when output_router_logits=False",
+    "updated_at": "2026-02-26T12:41:46Z"
   },
   {
-    "additions": 9,
-    "author": "Rocketknight1",
+    "additions": 1,
+    "author": "vasqu",
     "author_association": "MEMBER",
-    "body_excerpt": "`create_import_structure_from_path` does some redundant `os` calls, so I'm experimenting with changes to see if we can speed up loading a lot. Related to #44246",
+    "body_excerpt": "Accidentally caused by #43325, wrong naming --> modular doesn't properly convert some files (e.g. kyutai) Also fixes red CI on main",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44253",
-    "created_at": "2026-02-24T13:03:40Z",
-    "deletions": 15,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44283",
+    "created_at": "2026-02-25T18:33:17Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44253/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44253",
+    "files_url": "https://github.com/huggingface/transformers/pull/44283/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44283",
     "labels": [],
     "merged": true,
-    "number": 44253,
+    "number": 44283,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Speed create_import_structure up with os.scandir()",
-    "updated_at": "2026-03-10T12:49:42Z"
+    "title": "[`Modular`] Fix file type regression",
+    "updated_at": "2026-02-25T20:04:41Z"
   },
   {
-    "additions": 718,
-    "author": "zucchini-nlp",
+    "additions": 5,
+    "author": "Rocketknight1",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Deprecate timm backbone in favor of keeping all models within one `timm` folder, similar to other vision models. A backbone is just a variation of `PreTrainedModel`",
-    "changed_files": 61,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44252",
-    "created_at": "2026-02-24T13:00:59Z",
-    "deletions": 772,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44252/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44252",
-    "labels": [],
-    "merged": false,
-    "number": 44252,
-    "review_comments_count": 9,
-    "state": "open",
-    "title": "Timm unification continued",
-    "updated_at": "2026-02-26T13:35:44Z"
-  },
-  {
-    "additions": 1951,
-    "author": "Sai-Suraj-27",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Model Page: https://huggingface.co/jinaai/jina-embeddings-v3 Model Paper: https://huggingface.co/papers/2409.10173 Downloads last month > **5.3M** Completes Part of https://github.com/huggingface/transformers/issues\u2026",
-    "changed_files": 13,
+    "body_excerpt": "Response schema save-loading was broken in #40936, this PR restores it! I did most of this in #42300 but missed an issue with loading/saving.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 29,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44251",
-    "created_at": "2026-02-24T12:56:24Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44282",
+    "created_at": "2026-02-25T17:57:54Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44251/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44251",
-    "labels": [
-      "New model"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44282/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44282",
+    "labels": [],
     "merged": true,
-    "number": 44251,
-    "review_comments_count": 74,
+    "number": 44282,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add `Jina-Embeddings-V3` Model",
-    "updated_at": "2026-03-19T10:07:57Z"
+    "title": "Restore response_schema saving-loading",
+    "updated_at": "2026-02-25T18:27:22Z"
   },
   {
-    "additions": 5,
-    "author": "SunMarc",
+    "additions": 1,
+    "author": "ArthurZucker",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR fixes https://github.com/huggingface/transformers/pull/43806#discussion_r2834269455. We removed `self.report_to == \"all\"` functionality by mistake. Adding it back !",
+    "body_excerpt": "# What does this PR do? Its a very small fix for #44062",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44250",
-    "created_at": "2026-02-24T12:38:21Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44281",
+    "created_at": "2026-02-25T16:28:37Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44250/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44250",
+    "files_url": "https://github.com/huggingface/transformers/pull/44281/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44281",
     "labels": [],
     "merged": true,
-    "number": 44250,
+    "number": 44281,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix regression report_to \"all\"",
-    "updated_at": "2026-02-24T12:55:06Z"
+    "title": "Fix special token maps BC",
+    "updated_at": "2026-02-26T10:34:17Z"
   },
   {
-    "additions": 9,
-    "author": "Ryan-J-MAX",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR fix? This PR adds backward compatibility for the deprecated `grouped_entities` parameter in the `TokenClassificationPipeline`. ## Problem The `grouped_entities` parameter was deprecated in favor of `aggregation_strateg\u2026",
-    "changed_files": 1,
+    "additions": 614,
+    "author": "RishabhMehra",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "# What does this PR do? - Adds an opt-in use_fast_grouping flag to TokenClassificationPipeline to enable a NumPy-vectorised BIO grouping path (~5\u00d7 faster on long sequences) while keeping the legacy path as default. - Improves correctness:\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44249",
-    "created_at": "2026-02-24T10:48:54Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44278",
+    "created_at": "2026-02-25T12:49:56Z",
+    "deletions": 63,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44249/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44249",
+    "files_url": "https://github.com/huggingface/transformers/pull/44278/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44278",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44249,
+    "number": 44278,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: add backward compatibility for grouped_entities parameter",
-    "updated_at": "2026-02-24T12:31:26Z"
+    "title": "[FEAT] Pipelines - Faster group_entities",
+    "updated_at": "2026-02-25T13:54:58Z"
   },
   {
-    "additions": 12,
-    "author": "yonigozlan",
+    "additions": 105,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fix backward compatibility with remote code for old processors not defining valid_kwargs (e.g. phi4) Cc @zucchini-nlp Fix `test_processor_override` for phi3v and phi4 in vllm @hmellor",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? This patch makes the GLM-ASR doc example runnable by using `runnables` - see https://github.com/huggingface/doc-builder/blob/main/docs/runnable-code-blocks.md",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44245",
-    "created_at": "2026-02-23T21:47:19Z",
-    "deletions": 4,
+    "comments_count": 36,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44277",
+    "created_at": "2026-02-25T08:49:20Z",
+    "deletions": 19,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44245/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44245",
+    "files_url": "https://github.com/huggingface/transformers/pull/44277/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44277",
     "labels": [],
     "merged": true,
-    "number": 44245,
-    "review_comments_count": 3,
+    "number": 44277,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "Fix image processors `from_dict` backward compatibility with old remote code",
-    "updated_at": "2026-02-24T15:17:37Z"
+    "title": "Use doc-builder runnable example for GLM-ASR",
+    "updated_at": "2026-04-02T16:16:55Z"
   },
   {
-    "additions": 63,
-    "author": "thakoreh",
+    "additions": 0,
+    "author": "vishalpatil-45",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44242 Load balancing loss was not being added when `output_router_logits=False` in Mixtral models. ## Changes - Fixed loss calculation to include load balancing even when router logits are not output - Added test case ##\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? This PR addresses the performance regression where `import transformers` takes ~3.5s. The issue was caused by eager imports of heavy backend libraries (like torch/numpy) during the initial module load. By moving the\u2026",
+    "changed_files": 0,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44243",
-    "created_at": "2026-02-23T21:27:09Z",
-    "deletions": 5,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44275",
+    "created_at": "2026-02-25T08:27:32Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44243/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44243",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44275/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44275",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44243,
+    "number": 44275,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: add load balancing loss when output_router_logits=False",
-    "updated_at": "2026-02-23T21:54:11Z"
+    "title": "[Fix] Restore lazy loading to improve import performance (#44273)",
+    "updated_at": "2026-02-25T20:37:18Z"
   },
   {
-    "additions": 9,
-    "author": "yushiran",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Adds missing `-> bool`, `-> int`, and `-> str | None` return type annotations to public utility functions in `utils/generic.py`, making them consistent with the newer `is_timm_config_dict` and `is_timm_local_checkpoint` function\u2026",
-    "changed_files": 1,
+    "additions": 559,
+    "author": "paipeline",
+    "author_association": "NONE",
+    "body_excerpt": "## Description Fixes #44242 This PR resolves an issue where the auxiliary load balancing loss was not computed when `output_router_logits=False`, even when `router_aux_loss_coef != 0`. ## Problem The auxiliary loss computation was incorrec\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44241",
-    "created_at": "2026-02-23T19:50:05Z",
-    "deletions": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44274",
+    "created_at": "2026-02-25T06:38:02Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44241/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44241",
-    "labels": [],
-    "merged": true,
-    "number": 44241,
+    "files_url": "https://github.com/huggingface/transformers/pull/44274/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44274",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44274,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: add missing return type annotations to type-checking utilities in generic.py",
-    "updated_at": "2026-02-24T13:27:11Z"
+    "title": "Fix auxiliary load balancing loss computation when output_router_logits=False",
+    "updated_at": "2026-02-25T13:36:03Z"
   },
   {
-    "additions": 2,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Makes sure `find_bad_commit` always return the result `dict`",
+    "additions": 1,
+    "author": "hangjun-ezra",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes a `TypeError: unsupported operand type(s) for |: 'list' and 'set'` in `RotaryEmbeddingConfigMixin.convert_rope_params_to_dict` when `ignore_keys_at_rope_validation` is a `list` instead of a `set`. ### Root ca\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44240",
-    "created_at": "2026-02-23T19:12:49Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44272",
+    "created_at": "2026-02-25T03:52:04Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44240/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44240",
+    "files_url": "https://github.com/huggingface/transformers/pull/44272/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44272",
     "labels": [],
     "merged": true,
-    "number": 44240,
+    "number": 44272,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix return value - fixes #44238",
-    "updated_at": "2026-02-24T13:02:59Z"
+    "title": "Fix TypeError in convert_rope_params_to_dict when ignore_keys is a list",
+    "updated_at": "2026-02-25T14:38:36Z"
   },
   {
-    "additions": 253,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "part 2 of refactoring the training docs adds new dedicated guide to callbacks and data collators todo: - [x] backlink to `## Next steps` in `trainer.md` once https://github.com/huggingface/transformers/pull/44185 is merged",
-    "changed_files": 7,
+    "additions": 1272,
+    "author": "balak4",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary - Add GreedyLR, a metric-based adaptive learning rate scheduler that adjusts the learning rate during training based on the current loss - Based on [\"Dynamic Learning Rate Scheduling based on Loss Changes Leads to Faster Converg\u2026",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44239",
-    "created_at": "2026-02-23T18:54:55Z",
-    "deletions": 47,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44271",
+    "created_at": "2026-02-25T01:40:57Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44239/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44239",
+    "files_url": "https://github.com/huggingface/transformers/pull/44271/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44271",
     "labels": [],
     "merged": true,
-    "number": 44239,
-    "review_comments_count": 10,
+    "number": 44271,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "[docs] callbacks and collators",
-    "updated_at": "2026-02-24T22:12:46Z"
+    "title": "Add GreedyLR adaptive learning rate scheduler",
+    "updated_at": "2026-03-18T18:45:46Z"
   },
   {
-    "additions": 1,
-    "author": "eustlb",
+    "additions": 88,
+    "author": "yonigozlan",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? small nit but will be misleading if not fixed",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? A lot of ProcessorsKwargs have incorrect/unspecified type hints in their ProcessorsKwargs TypedDict for their images_kwargs attribute. Functionnaly, this did not cause issues as \"_merge_kwargs\" automatically picks u\u2026",
+    "changed_files": 44,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44237",
-    "created_at": "2026-02-23T17:52:17Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44270",
+    "created_at": "2026-02-25T00:11:31Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44237/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44237",
+    "files_url": "https://github.com/huggingface/transformers/pull/44270/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44270",
     "labels": [],
-    "merged": true,
-    "number": 44237,
+    "merged": false,
+    "number": 44270,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[mimi] nit",
-    "updated_at": "2026-02-24T15:43:55Z"
+    "state": "open",
+    "title": "Add correct typing to custom images_kwargs in ProcessorsKwargs",
+    "updated_at": "2026-02-25T01:12:06Z"
   },
   {
-    "additions": 109,
-    "author": "SunMarc",
+    "additions": 30,
+    "author": "yonigozlan",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Supersedes https://github.com/huggingface/transformers/pull/43847 When using zero3 + from_config, the model was incorrectly initialized as we were not gathering the params. Added a test also. cc @tohtana",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? This is a follow-up to https://github.com/huggingface/transformers/pull/43748, and will allow to have clickable links to the full modality kwargs when present in the docstring of a processor or image processor Cc @s\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44236",
-    "created_at": "2026-02-23T17:20:01Z",
-    "deletions": 3,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44269",
+    "created_at": "2026-02-25T00:05:47Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44236/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44236",
+    "files_url": "https://github.com/huggingface/transformers/pull/44269/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44269",
     "labels": [],
     "merged": true,
-    "number": 44236,
+    "number": 44269,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix zero3 init config",
-    "updated_at": "2026-02-27T11:36:19Z"
+    "title": "Add `ProcessingKwargs` `ImagesKwargs` etc. to docs",
+    "updated_at": "2026-02-27T19:03:15Z"
   },
   {
-    "additions": 1,
-    "author": "itazap",
-    "author_association": "MEMBER",
-    "body_excerpt": "UPDATE TO: https://github.com/huggingface/transformers/pull/44179/changes Models with incorrect tokenizer_class in tokenization_config.json that should use TokenziersBackend",
+    "additions": 5,
+    "author": "ethanknights",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Some improvements to the `trainer.py` docs. ## Before submitting - [x] This PR fixes a typo or improves the docs. ## Who can review? Documentation: @stevhliu",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44235",
-    "created_at": "2026-02-23T17:06:54Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44268",
+    "created_at": "2026-02-24T23:20:16Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44235/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44235",
+    "files_url": "https://github.com/huggingface/transformers/pull/44268/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44268",
     "labels": [],
     "merged": true,
-    "number": 44235,
+    "number": 44268,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "update fuyu tokenizer class",
-    "updated_at": "2026-02-23T17:36:22Z"
+    "title": "chore: fixes in `Trainer` class docs (`compute_loss` & `hyperparameter_search`)",
+    "updated_at": "2026-02-26T00:50:23Z"
   },
   {
-    "additions": 249,
-    "author": "yonigozlan",
-    "author_association": "MEMBER",
-    "body_excerpt": "Cc @zucchini-nlp",
-    "changed_files": 3,
+    "additions": 4,
+    "author": "manavshrivastavagit",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Update the `DocumentQuestionAnsweringPipeline` docstring to explicitly mention the task summary in the Transformers documentation. - Remove the stale TODO comment now that document question answering is covered in the task sum\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44234",
-    "created_at": "2026-02-23T17:03:05Z",
-    "deletions": 55,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44267",
+    "created_at": "2026-02-24T20:35:18Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44234/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44234",
-    "labels": [],
-    "merged": true,
-    "number": 44234,
-    "review_comments_count": 2,
+    "files_url": "https://github.com/huggingface/transformers/pull/44267/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44267",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44267,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add processing tests for phi4 multimodal",
-    "updated_at": "2026-02-23T22:08:11Z"
+    "title": "Docs: point DocumentQuestionAnswering pipeline to task summary",
+    "updated_at": "2026-02-25T13:34:48Z"
   },
   {
-    "additions": 219,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "Extends `ty` coverage to `src/transformers/generation` - Added a dedicated type-check wrapper script: `utils/check_types.py`. - Updated `Makefile` to run `ty` checks through the wrapper in both `style` and `check-repo`. - merged all typing\u2026",
-    "changed_files": 15,
+    "additions": 27,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? The following issue was identified and fixed in this PR: \u2192 **Reasoning:** The impact of this fix goes beyond `Mask2Former` and `DeformableDetr` and should fix any model that uses `torch_compilable_check`. Most use\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 12,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44233",
-    "created_at": "2026-02-23T16:23:24Z",
-    "deletions": 101,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44266",
+    "created_at": "2026-02-24T20:02:06Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44233/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44233",
+    "files_url": "https://github.com/huggingface/transformers/pull/44266/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44266",
     "labels": [],
     "merged": true,
-    "number": 44233,
-    "review_comments_count": 33,
+    "number": 44266,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "chore(typing): Add type checking to `src/transformers/generation`",
-    "updated_at": "2026-03-04T17:24:37Z"
+    "title": "fix(utils): Make torch_compilable_check compatible with torch.export strict mode",
+    "updated_at": "2026-02-26T09:42:47Z"
   },
   {
-    "additions": 11,
-    "author": "tarekziade",
+    "additions": 90,
+    "author": "vasqu",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? per https://code.claude.com/docs/en/claude-code-on-the-web#best-practices `CLAUDE.md` can alias directly into `AGENTS.md`",
-    "changed_files": 2,
+    "body_excerpt": "As per title, WIP --> needs a test",
+    "changed_files": 36,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44232",
-    "created_at": "2026-02-23T16:10:15Z",
-    "deletions": 109,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44232/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44232",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44264",
+    "created_at": "2026-02-24T18:06:58Z",
+    "deletions": 210,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44264/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44264",
     "labels": [],
-    "merged": true,
-    "number": 44232,
-    "review_comments_count": 2,
-    "state": "closed",
-    "title": "chore: added CLAUDE.md alias",
-    "updated_at": "2026-02-24T14:48:36Z"
+    "merged": false,
+    "number": 44264,
+    "review_comments_count": 3,
+    "state": "open",
+    "title": "[`Moe`] Enable aux loss automatically when in training + coef is not 0",
+    "updated_at": "2026-02-25T18:53:20Z"
   },
   {
-    "additions": 413,
-    "author": "IlyasMoutawwakil",
+    "additions": 5882,
+    "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 5,
+    "body_excerpt": "# What does this PR do? This PR refactor the common tests that we have in Trainer. I've mainly did the following: - Split the tests that we have in `test_trainer.py` into multiple files. - Fix common tests that were failing in the CI",
+    "changed_files": 18,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44231",
-    "created_at": "2026-02-23T15:45:47Z",
-    "deletions": 578,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44260",
+    "created_at": "2026-02-24T15:51:11Z",
+    "deletions": 6147,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44231/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44231",
+    "files_url": "https://github.com/huggingface/transformers/pull/44260/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44260",
     "labels": [],
     "merged": true,
-    "number": 44231,
-    "review_comments_count": 18,
+    "number": 44260,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "[Performance] FP8 Grouped and Batched Matmuls",
-    "updated_at": "2026-03-11T08:51:02Z"
+    "title": "Update common tests Trainer",
+    "updated_at": "2026-02-27T17:31:59Z"
   },
   {
-    "additions": 4,
-    "author": "alvarobartt",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR adds the missing backtick (`) on the `AnyToAnyPipeline.__call__` docstrings, as those were showing as in the screenshot below instead. <img width=\"1023\" height=\"400\" alt=\"image\" src=\"https://github.com/user-\u2026",
-    "changed_files": 1,
+    "additions": 1830,
+    "author": "winglian",
+    "author_association": "COLLABORATOR",
+    "body_excerpt": "# What does this PR do? This PR supersedes #43985 to replace the dataset/sampler/dataloader with a data producer that should allow us to more easily get to the next step of async training for RL. <!-- Congratulations! You've made it this f\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44229",
-    "created_at": "2026-02-23T15:25:47Z",
-    "deletions": 4,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44229/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44229",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44259",
+    "created_at": "2026-02-24T15:01:56Z",
+    "deletions": 59,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44259/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44259",
     "labels": [],
-    "merged": true,
-    "number": 44229,
+    "merged": false,
+    "number": 44259,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Add missing backtick in `AnyToAnyPipeline.__call__` docstring",
-    "updated_at": "2026-02-23T19:21:08Z"
+    "state": "open",
+    "title": "Async data producer",
+    "updated_at": "2026-02-26T19:57:43Z"
   },
   {
-    "additions": 35,
-    "author": "JonoLF",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "additions": 8,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "The old one has a merge conflict and it was easier to just mirror into a new branch / PR Note that this only affects the local big tests that I used to run on the A100s locally; not to be run with the CI (too big)",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44228",
-    "created_at": "2026-02-23T15:09:05Z",
-    "deletions": 7,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44258",
+    "created_at": "2026-02-24T15:00:29Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44228/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44228",
+    "files_url": "https://github.com/huggingface/transformers/pull/44258/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44258",
     "labels": [],
     "merged": false,
-    "number": 44228,
+    "number": 44258,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[Quantisation] account for nested tensors from quantisers",
-    "updated_at": "2026-03-17T11:57:53Z"
+    "state": "closed",
+    "title": "[`Ernie 4.5 VL Moe`] Change revision",
+    "updated_at": "2026-03-14T19:59:05Z"
   },
   {
-    "additions": 21,
-    "author": "remi-or",
-    "author_association": "MEMBER",
-    "body_excerpt": "This PR adds a logging message when infering the behavior of use async and fixes an error when evicting a graph from the graph buffer.",
-    "changed_files": 4,
+    "additions": 3,
+    "author": "winglian",
+    "author_association": "COLLABORATOR",
+    "body_excerpt": "# What does this PR do? When post training using context parallelism, some processes may have their chunk of the sample input masked out leading to a NaN loss for that process. Using `nanmean` allows us to keep the real loss that isn't `Na\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44227",
-    "created_at": "2026-02-23T14:53:53Z",
-    "deletions": 13,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44257",
+    "created_at": "2026-02-24T14:56:42Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44227/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44227",
+    "files_url": "https://github.com/huggingface/transformers/pull/44257/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44257",
     "labels": [],
-    "merged": true,
-    "number": 44227,
-    "review_comments_count": 9,
-    "state": "closed",
-    "title": "[CB] Small fixes",
-    "updated_at": "2026-03-03T13:40:10Z"
+    "merged": false,
+    "number": 44257,
+    "review_comments_count": 5,
+    "state": "open",
+    "title": "use nanmean for aggregating loss",
+    "updated_at": "2026-02-25T17:01:08Z"
   },
   {
-    "additions": 86,
-    "author": "Cyrilvallez",
+    "additions": 10,
+    "author": "albertvillanova",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. It looks like some models (xlnet and kosmos2_5) and most audio models sometimes rely on the full previous input_ids to prepare inputs. Note that this cannot be compatible with restarting generation\u2026",
-    "changed_files": 12,
+    "body_excerpt": "Fix CLI NameError: name 'TypeAdapter' is not defined: - Do not evaluate type annotations in CLI serve ### Problem Calling the CLI raises NameError: > NameError: name 'TypeAdapter' is not defined ```bash transformers --help ``` ```python Tr\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44226",
-    "created_at": "2026-02-23T13:27:23Z",
-    "deletions": 66,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44256",
+    "created_at": "2026-02-24T14:54:49Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44226/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44226",
+    "files_url": "https://github.com/huggingface/transformers/pull/44256/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44256",
     "labels": [],
     "merged": true,
-    "number": 44226,
-    "review_comments_count": 17,
+    "number": 44256,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[generate] Always pass full input_ids in `prepare_inputs_for_generation`",
-    "updated_at": "2026-02-24T10:45:49Z"
+    "title": "Fix CLI NameError: name 'TypeAdapter' is not defined",
+    "updated_at": "2026-03-12T07:57:14Z"
   },
   {
-    "additions": 169,
-    "author": "zucchini-nlp",
+    "additions": 404,
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per title, we weren't running these test for some time because they were being filtered into `non-model` tests. But `non-model` doesn't run tests that are marked as generation, so I moved it back to `generation`\u2026",
-    "changed_files": 5,
+    "body_excerpt": "## What this PR does Given he different issues that were noticed by @hmellor on vLLM, we wanted to make sure we did not end up with crazy breaks. We ran a full test suite (code can be found in #44298) and the results showed 22 model conver\u2026",
+    "changed_files": 24,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44225",
-    "created_at": "2026-02-23T12:09:40Z",
-    "deletions": 270,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44255",
+    "created_at": "2026-02-24T14:17:00Z",
+    "deletions": 205,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44225/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44225",
+    "files_url": "https://github.com/huggingface/transformers/pull/44255/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44255",
     "labels": [],
     "merged": true,
-    "number": 44225,
-    "review_comments_count": 18,
+    "number": 44255,
+    "review_comments_count": 25,
     "state": "closed",
-    "title": "Fix generation integration tests",
-    "updated_at": "2026-02-25T09:19:39Z"
+    "title": "[vllm + v5 fix] handle TokenizersBackend fallback properly for v5",
+    "updated_at": "2026-03-23T11:07:37Z"
   },
   {
-    "additions": 5,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Small fixes after https://github.com/huggingface/transformers/pull/44130. See https://huggingface.co/datasets/hf-internal-testing/transformers_daily_ci/raw/8785954cca2fdca181de0b9567059471bcadb959/2026-02-21/ci_resu\u2026",
-    "changed_files": 3,
+    "additions": 16,
+    "author": "mario-sanz",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully refle\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44224",
-    "created_at": "2026-02-23T10:48:19Z",
-    "deletions": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44254",
+    "created_at": "2026-02-24T13:54:30Z",
+    "deletions": 13,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44224/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44224",
+    "files_url": "https://github.com/huggingface/transformers/pull/44254/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44254",
     "labels": [],
     "merged": false,
-    "number": 44224,
-    "review_comments_count": 0,
+    "number": 44254,
+    "review_comments_count": 5,
     "state": "closed",
-    "title": "Small fixes",
-    "updated_at": "2026-02-24T10:06:14Z"
+    "title": "Fix fast tokenizers overwriting custom `pre_tokenizer` from `tokenizer.json`",
+    "updated_at": "2026-02-26T08:45:56Z"
   },
   {
-    "additions": 1,
-    "author": "albertvillanova",
+    "additions": 9,
+    "author": "Rocketknight1",
     "author_association": "MEMBER",
-    "body_excerpt": "Fix type of `TrainingArguments.logging_steps`. This PR makes a minor update to the `TrainingArguments` class, so `logging_steps` parameter accepts both integers and floats, rather than only floats. Note these are the expected types in the\u2026",
+    "body_excerpt": "`create_import_structure_from_path` does some redundant `os` calls, so I'm experimenting with changes to see if we can speed up loading a lot. Related to #44246",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44223",
-    "created_at": "2026-02-23T08:50:04Z",
-    "deletions": 1,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44253",
+    "created_at": "2026-02-24T13:03:40Z",
+    "deletions": 15,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44223/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44223",
+    "files_url": "https://github.com/huggingface/transformers/pull/44253/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44253",
     "labels": [],
-    "merged": false,
-    "number": 44223,
+    "merged": true,
+    "number": 44253,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix type of TrainingArguments.logging_steps",
-    "updated_at": "2026-02-23T09:08:18Z"
+    "title": "Speed create_import_structure up with os.scandir()",
+    "updated_at": "2026-03-10T12:49:42Z"
   },
   {
-    "additions": 1,
-    "author": "matisgagneux21",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary - fix a typo in the Italian add-new-model guide: `docstirng` -> `docstring`. ## Why Small docs quality fix that avoids confusion for readers following the contribution guide.",
-    "changed_files": 1,
+    "additions": 718,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Deprecate timm backbone in favor of keeping all models within one `timm` folder, similar to other vision models. A backbone is just a variation of `PreTrainedModel`",
+    "changed_files": 61,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44219",
-    "created_at": "2026-02-23T00:43:59Z",
-    "deletions": 1,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44252",
+    "created_at": "2026-02-24T13:00:59Z",
+    "deletions": 772,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44219/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44219",
+    "files_url": "https://github.com/huggingface/transformers/pull/44252/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44252",
     "labels": [],
-    "merged": true,
-    "number": 44219,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Docs(it): fix typo in docstring wording",
-    "updated_at": "2026-02-23T15:04:51Z"
+    "merged": false,
+    "number": 44252,
+    "review_comments_count": 9,
+    "state": "open",
+    "title": "Timm unification continued",
+    "updated_at": "2026-02-26T13:35:44Z"
   },
   {
-    "additions": 1,
-    "author": "matisgagneux21",
+    "additions": 1951,
+    "author": "Sai-Suraj-27",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary - fix a typo in the Italian migration guide install command: - `stentencepiece` -> `sentencepiece` ## Why The current command fails if copied as-is. This makes the installation snippet runnable for users reading the Italian docs.",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Model Page: https://huggingface.co/jinaai/jina-embeddings-v3 Model Paper: https://huggingface.co/papers/2409.10173 Downloads last month > **5.3M** Completes Part of https://github.com/huggingface/transformers/issues\u2026",
+    "changed_files": 13,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44218",
-    "created_at": "2026-02-23T00:32:49Z",
-    "deletions": 1,
+    "comments_count": 29,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44251",
+    "created_at": "2026-02-24T12:56:24Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44218/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44218",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44251/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44251",
+    "labels": [
+      "New model"
+    ],
     "merged": true,
-    "number": 44218,
-    "review_comments_count": 0,
+    "number": 44251,
+    "review_comments_count": 74,
     "state": "closed",
-    "title": "Docs(it): fix typo in sentencepiece install command",
-    "updated_at": "2026-02-23T15:05:17Z"
+    "title": "Add `Jina-Embeddings-V3` Model",
+    "updated_at": "2026-03-19T10:07:57Z"
   },
   {
-    "additions": 1,
-    "author": "matisgagneux21",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary - Update the Italian migration guide to avoid pointing users to the deprecated `grouped_entities` flag. - Clarify that `aggregation_strategy` is the current option (with a note that it was previously `grouped_entities`). ## Why\u2026",
+    "additions": 5,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR fixes https://github.com/huggingface/transformers/pull/43806#discussion_r2834269455. We removed `self.report_to == \"all\"` functionality by mistake. Adding it back !",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44217",
-    "created_at": "2026-02-23T00:10:48Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44250",
+    "created_at": "2026-02-24T12:38:21Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44217/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44217",
+    "files_url": "https://github.com/huggingface/transformers/pull/44250/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44250",
     "labels": [],
-    "merged": false,
-    "number": 44217,
+    "merged": true,
+    "number": 44250,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Docs(it): update deprecated grouped_entities reference",
-    "updated_at": "2026-02-23T14:56:19Z"
+    "title": "fix regression report_to \"all\"",
+    "updated_at": "2026-02-24T12:55:06Z"
   },
   {
-    "additions": 13,
-    "author": "nikste",
+    "additions": 9,
+    "author": "Ryan-J-MAX",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Automated fix for #44016: Syntax error in Transformer section 3 (Transformers, what can they do?) notebook Fixes #44016 ## What does this PR do? This PR addresses issue #44016 by implementing the fix described in the issue. ---\u2026",
+    "body_excerpt": "## What does this PR fix? This PR adds backward compatibility for the deprecated `grouped_entities` parameter in the `TokenClassificationPipeline`. ## Problem The `grouped_entities` parameter was deprecated in favor of `aggregation_strateg\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44216",
-    "created_at": "2026-02-22T23:40:56Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44249",
+    "created_at": "2026-02-24T10:48:54Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44216/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44216",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44249/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44249",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44216,
+    "number": 44249,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix #44016: Syntax error in Transformer section 3 (Transformers, what can they do?) notebook",
-    "updated_at": "2026-02-22T23:50:57Z"
+    "title": "fix: add backward compatibility for grouped_entities parameter",
+    "updated_at": "2026-02-24T12:31:26Z"
   },
   {
-    "additions": 187,
-    "author": "jmriosal",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "# What does this PR do? Add sequence classification capabilities to the family of Granite models (Granite, GraniteMoe, GraniteMoeHybrid, and GraniteMoeShared). Fixes #44214, #35720 ## Why The Granite models currently only have the base mod\u2026",
-    "changed_files": 17,
+    "additions": 12,
+    "author": "yonigozlan",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fix backward compatibility with remote code for old processors not defining valid_kwargs (e.g. phi4) Cc @zucchini-nlp Fix `test_processor_override` for phi3v and phi4 in vllm @hmellor",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44215",
-    "created_at": "2026-02-22T23:24:43Z",
-    "deletions": 13,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44245",
+    "created_at": "2026-02-23T21:47:19Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44215/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44215",
+    "files_url": "https://github.com/huggingface/transformers/pull/44245/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44245",
     "labels": [],
-    "merged": false,
-    "number": 44215,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Add sequence classification capability to Granite models",
-    "updated_at": "2026-02-24T20:39:37Z"
+    "merged": true,
+    "number": 44245,
+    "review_comments_count": 3,
+    "state": "closed",
+    "title": "Fix image processors `from_dict` backward compatibility with old remote code",
+    "updated_at": "2026-02-24T15:17:37Z"
   },
   {
-    "additions": 70,
-    "author": "parthchopra07",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "# What does this PR do? This PR refreshes the BEiT model documentation to align it with the current Transformers vision docs style and features. It updates the usage examples, clarifies configuration details, and improves the resources sec\u2026",
-    "changed_files": 1,
+    "additions": 63,
+    "author": "thakoreh",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #44242 Load balancing loss was not being added when `output_router_logits=False` in Mixtral models. ## Changes - Fixed loss calculation to include load balancing even when router logits are not output - Added test case ##\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44213",
-    "created_at": "2026-02-22T18:32:16Z",
-    "deletions": 29,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44243",
+    "created_at": "2026-02-23T21:27:09Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44213/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44213",
+    "files_url": "https://github.com/huggingface/transformers/pull/44243/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44243",
     "labels": [],
     "merged": false,
-    "number": 44213,
+    "number": 44243,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Update BEiT model card",
-    "updated_at": "2026-02-28T14:33:57Z"
+    "title": "fix: add load balancing loss when output_router_logits=False",
+    "updated_at": "2026-02-23T21:54:11Z"
   },
   {
-    "additions": 1,
-    "author": "alexandercarruthers",
+    "additions": 9,
+    "author": "yushiran",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Updates a broken link in the summarization guide. https://huggingface.co/docs/transformers/tasks/summarization https://huggingface.co/billsum/datasets results in a 404. New URL is https://huggingface.co/datasets/Fis\u2026",
+    "body_excerpt": "## Summary Adds missing `-> bool`, `-> int`, and `-> str | None` return type annotations to public utility functions in `utils/generic.py`, making them consistent with the newer `is_timm_config_dict` and `is_timm_local_checkpoint` function\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44212",
-    "created_at": "2026-02-22T18:02:43Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44241",
+    "created_at": "2026-02-23T19:50:05Z",
+    "deletions": 9,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44212/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44212",
+    "files_url": "https://github.com/huggingface/transformers/pull/44241/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44241",
     "labels": [],
     "merged": true,
-    "number": 44212,
+    "number": 44241,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Update 404ing BillSum dataset URL on Summarization Task guide",
-    "updated_at": "2026-02-23T14:46:11Z"
+    "title": "fix: add missing return type annotations to type-checking utilities in generic.py",
+    "updated_at": "2026-02-24T13:27:11Z"
   },
   {
-    "additions": 10,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## Fix for #44016 The `grouped_entities` parameter in `TokenClassificationPipeline._sanitize_parameters` was removed without a deprecation period, causing a `TypeError` when users pass `grouped_entities=True` to the `pipeline()` call (as s\u2026",
+    "additions": 2,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Makes sure `find_bad_commit` always return the result `dict`",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44211",
-    "created_at": "2026-02-22T17:04:50Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44240",
+    "created_at": "2026-02-23T19:12:49Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44211/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44211",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44211,
+    "files_url": "https://github.com/huggingface/transformers/pull/44240/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44240",
+    "labels": [],
+    "merged": true,
+    "number": 44240,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add backward compatibility for deprecated grouped_entities parameter",
-    "updated_at": "2026-02-23T16:26:02Z"
+    "title": "Fix return value - fixes #44238",
+    "updated_at": "2026-02-24T13:02:59Z"
   },
   {
-    "additions": 1,
-    "author": "nightcityblade",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes #44206 ## Problem PR #43769 (\"Add Voxtral Realtime\") added a `center` parameter to `LasrFeatureExtractor.__call__()` and passed it to `_torch_extract_fbank_features()`, but that method does not accept it. This causes a `TypeError` on\u2026",
-    "changed_files": 1,
+    "additions": 253,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "part 2 of refactoring the training docs adds new dedicated guide to callbacks and data collators todo: - [x] backlink to `## Next steps` in `trainer.md` once https://github.com/huggingface/transformers/pull/44185 is merged",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44210",
-    "created_at": "2026-02-22T16:06:16Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44239",
+    "created_at": "2026-02-23T18:54:55Z",
+    "deletions": 47,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44210/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44210",
+    "files_url": "https://github.com/huggingface/transformers/pull/44239/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44239",
     "labels": [],
-    "merged": false,
-    "number": 44210,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44239,
+    "review_comments_count": 10,
     "state": "closed",
-    "title": "fix(lasr): remove spurious center arg from _torch_extract_fbank_features call",
-    "updated_at": "2026-02-23T14:32:41Z"
+    "title": "[docs] callbacks and collators",
+    "updated_at": "2026-02-24T22:12:46Z"
   },
   {
-    "additions": 197,
-    "author": "paipeline",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? This PR fixes a critical bug in LayoutLMv2Tokenizer where passing `word_labels` for NER token classification tasks would crash with `AttributeError`. The issue was that `word_ids` and `sequence_ids` were being acce\u2026",
-    "changed_files": 3,
+    "additions": 1,
+    "author": "eustlb",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? small nit but will be misleading if not fixed",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44209",
-    "created_at": "2026-02-22T14:37:25Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44237",
+    "created_at": "2026-02-23T17:52:17Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44209/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44209",
+    "files_url": "https://github.com/huggingface/transformers/pull/44237/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44237",
     "labels": [],
-    "merged": false,
-    "number": 44209,
+    "merged": true,
+    "number": 44237,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix LayoutLMv2Tokenizer NER crashes with word_labels",
-    "updated_at": "2026-02-23T10:30:26Z"
+    "title": "[mimi] nit",
+    "updated_at": "2026-02-24T15:43:55Z"
   },
   {
-    "additions": 1,
-    "author": "ainergiz",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR fixes a LASR regression introduced in `#43769` (released in `v5.2.0`). `LasrFeatureExtractor.__call__` passes `center` into `_torch_extract_fbank_features(...)`, but `_torch_extract_fbank_features` did not a\u2026",
-    "changed_files": 3,
+    "additions": 109,
+    "author": "SunMarc",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Supersedes https://github.com/huggingface/transformers/pull/43847 When using zero3 + from_config, the model was incorrectly initialized as we were not gathering the params. Added a test also. cc @tohtana",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44207",
-    "created_at": "2026-02-21T20:56:49Z",
-    "deletions": 70,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44236",
+    "created_at": "2026-02-23T17:20:01Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44207/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44207",
+    "files_url": "https://github.com/huggingface/transformers/pull/44236/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44236",
     "labels": [],
     "merged": true,
-    "number": 44207,
+    "number": 44236,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix LASR feature extractor regression from invalid center argument",
-    "updated_at": "2026-02-23T10:01:35Z"
+    "title": "fix zero3 init config",
+    "updated_at": "2026-02-27T11:36:19Z"
   },
   {
     "additions": 1,
-    "author": "nightcityblade",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes #44186 `LayoutLMv2Tokenizer.__init__` passes `only_label_first_subword` to `super().__init__()` but never stores it as `self.only_label_first_subword`. This causes an `AttributeError` when `word_labels` is pa\u2026",
+    "author": "itazap",
+    "author_association": "MEMBER",
+    "body_excerpt": "UPDATE TO: https://github.com/huggingface/transformers/pull/44179/changes Models with incorrect tokenizer_class in tokenization_config.json that should use TokenziersBackend",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44204",
-    "created_at": "2026-02-21T16:06:46Z",
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44235",
+    "created_at": "2026-02-23T17:06:54Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44204/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44204",
+    "files_url": "https://github.com/huggingface/transformers/pull/44235/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44235",
     "labels": [],
-    "merged": false,
-    "number": 44204,
+    "merged": true,
+    "number": 44235,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(layoutlmv2): store only_label_first_subword attribute in tokenizer",
-    "updated_at": "2026-02-23T10:30:19Z"
+    "title": "update fuyu tokenizer class",
+    "updated_at": "2026-02-23T17:36:22Z"
   },
   {
-    "additions": 22,
-    "author": "nightcityblade",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #44075 `_get_sgd()`, `_get_adagrad()`, and `_get_rmsprop()` in `trainer_optimizer.py` only returned `ctx.optimizer_kwargs` (which contains just `lr`), completely ignoring `ctx.optim_args`. This meant that parameters specif\u2026",
-    "changed_files": 1,
+    "additions": 249,
+    "author": "yonigozlan",
+    "author_association": "MEMBER",
+    "body_excerpt": "Cc @zucchini-nlp",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44203",
-    "created_at": "2026-02-21T15:12:17Z",
-    "deletions": 3,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44234",
+    "created_at": "2026-02-23T17:03:05Z",
+    "deletions": 55,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44203/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44203",
+    "files_url": "https://github.com/huggingface/transformers/pull/44234/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44234",
     "labels": [],
     "merged": true,
-    "number": 44203,
-    "review_comments_count": 0,
+    "number": 44234,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "fix(trainer): pass optim_args to SGD, Adagrad, and RMSprop optimizers",
-    "updated_at": "2026-02-25T16:04:20Z"
+    "title": "Add processing tests for phi4 multimodal",
+    "updated_at": "2026-02-23T22:08:11Z"
   },
   {
-    "additions": 63,
-    "author": "GS-GOAT",
-    "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? When `torch.compile` is used, [_ignore_bidirectional_mask_sdpa](cci:1://file:///c:/Users/BIT/Desktop/proj/gitrepo_clones/transformers/src/transformers/masking_utils.py:303:0-338:16) behaves differently than in eager\u2026",
-    "changed_files": 2,
+    "additions": 219,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "Extends `ty` coverage to `src/transformers/generation` - Added a dedicated type-check wrapper script: `utils/check_types.py`. - Updated `Makefile` to run `ty` checks through the wrapper in both `style` and `check-repo`. - merged all typing\u2026",
+    "changed_files": 15,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44202",
-    "created_at": "2026-02-21T13:45:15Z",
-    "deletions": 1,
+    "comments_count": 12,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44233",
+    "created_at": "2026-02-23T16:23:24Z",
+    "deletions": 101,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44202/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44202",
+    "files_url": "https://github.com/huggingface/transformers/pull/44233/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44233",
     "labels": [],
-    "merged": false,
-    "number": 44202,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44233,
+    "review_comments_count": 33,
     "state": "closed",
-    "title": "Fix: bidirectional mask skip when attention dropout is active (#44188)",
-    "updated_at": "2026-03-09T10:31:41Z"
+    "title": "chore(typing): Add type checking to `src/transformers/generation`",
+    "updated_at": "2026-03-04T17:24:37Z"
   },
   {
-    "additions": 18,
+    "additions": 11,
     "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? main is currently failing with ``` FAILED tests/models/higgs_audio_v2/test_modeling_higgs_audio_v2.py::HiggsAudioV2ModelTest::test_generate_compilation_all_outputs - AssertionError: Lists differ: [torch.Size([2, 15,\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? per https://code.claude.com/docs/en/claude-code-on-the-web#best-practices `CLAUDE.md` can alias directly into `AGENTS.md`",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44201",
-    "created_at": "2026-02-21T10:03:41Z",
-    "deletions": 13,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44232",
+    "created_at": "2026-02-23T16:10:15Z",
+    "deletions": 109,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44201/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44201",
+    "files_url": "https://github.com/huggingface/transformers/pull/44232/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44232",
     "labels": [],
     "merged": true,
-    "number": 44201,
-    "review_comments_count": 3,
+    "number": 44232,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "fix: HiggsAudioV2 cached decode inputs in compiled generation",
-    "updated_at": "2026-02-23T12:39:19Z"
+    "title": "chore: added CLAUDE.md alias",
+    "updated_at": "2026-02-24T14:48:36Z"
   },
   {
-    "additions": 3,
-    "author": "pragnyanramtha",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Fixes #43782 The `weights_only` parameter passed to `from_pretrained()` was not being forwarded to `load_state_dict()` when loading `.bin` checkpoint files in the non-DeepSpeed code path. This caused `weights_only` to always default to `Tr\u2026",
-    "changed_files": 1,
+    "additions": 413,
+    "author": "IlyasMoutawwakil",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44200",
-    "created_at": "2026-02-21T06:24:17Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44231",
+    "created_at": "2026-02-23T15:45:47Z",
+    "deletions": 578,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44200/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44200",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44200,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44231/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44231",
+    "labels": [],
+    "merged": true,
+    "number": 44231,
+    "review_comments_count": 18,
     "state": "closed",
-    "title": "fix: propagate `weights_only` param to `load_state_dict` in .bin loading path (#43782)",
-    "updated_at": "2026-02-23T14:20:12Z"
+    "title": "[Performance] FP8 Grouped and Batched Matmuls",
+    "updated_at": "2026-03-11T08:51:02Z"
   },
   {
-    "additions": 3,
-    "author": "gowthamr-tech",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? This PR fixes an issue in `run_image_classification_no_trainer.py` where the script always loaded `dataset_name` (e.g., CIFAR10) even when `--train_dir` or `--validation_dir` was provided. Now, when local dataset d\u2026",
+    "additions": 4,
+    "author": "alvarobartt",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? This PR adds the missing backtick (`) on the `AnyToAnyPipeline.__call__` docstrings, as those were showing as in the screenshot below instead. <img width=\"1023\" height=\"400\" alt=\"image\" src=\"https://github.com/user-\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44199",
-    "created_at": "2026-02-21T06:03:29Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44229",
+    "created_at": "2026-02-23T15:25:47Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44199/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44199",
+    "files_url": "https://github.com/huggingface/transformers/pull/44229/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44229",
     "labels": [],
     "merged": true,
-    "number": 44199,
+    "number": 44229,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix local dataset loading priority in run_image_classification_no_tra\u2026",
-    "updated_at": "2026-02-24T15:10:17Z"
+    "title": "Add missing backtick in `AnyToAnyPipeline.__call__` docstring",
+    "updated_at": "2026-02-23T19:21:08Z"
   },
   {
-    "additions": 71,
-    "author": "danielalanbates",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #43975 ## Summary This PR fixes: `deepseek-ai/deepseek-coder-6.7b-instruct` incorrectly detokenizes in v5 ## Changes ``` src/transformers/tokenization_utils_tokenizers.py | 12 ++++- tests/models/llama/test_tokenization_llama.py | 60\u2026",
+    "additions": 35,
+    "author": "JonoLF",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44198",
-    "created_at": "2026-02-21T04:54:47Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44228",
+    "created_at": "2026-02-23T15:09:05Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44198/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44198",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44228/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44228",
+    "labels": [],
     "merged": false,
-    "number": 44198,
+    "number": 44228,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Fix #43975: `deepseek-ai/deepseek-coder-6.7b-instruct` incorrectly detok",
-    "updated_at": "2026-02-23T14:10:47Z"
+    "state": "open",
+    "title": "[Quantisation] account for nested tensors from quantisers",
+    "updated_at": "2026-03-17T11:57:53Z"
   },
   {
-    "additions": 37,
-    "author": "danielalanbates",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #43937 ## Summary This PR fixes: [GLM-5] ValueError: GenerationConfig is invalid ## Changes ``` src/transformers/generation/configuration_utils.py | 13 +++++++++++- src/transformers/modeling_utils.py | 2 +- tests/generation/test_conf\u2026",
-    "changed_files": 3,
+    "additions": 21,
+    "author": "remi-or",
+    "author_association": "MEMBER",
+    "body_excerpt": "This PR adds a logging message when infering the behavior of use async and fixes an error when evicting a graph from the graph buffer.",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44197",
-    "created_at": "2026-02-21T04:47:32Z",
-    "deletions": 2,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44227",
+    "created_at": "2026-02-23T14:53:53Z",
+    "deletions": 13,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44197/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44197",
+    "files_url": "https://github.com/huggingface/transformers/pull/44227/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44227",
     "labels": [],
-    "merged": false,
-    "number": 44197,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44227,
+    "review_comments_count": 9,
     "state": "closed",
-    "title": "Fix #43937: [GLM-5] ValueError: GenerationConfig is invalid",
-    "updated_at": "2026-02-23T09:42:54Z"
+    "title": "[CB] Small fixes",
+    "updated_at": "2026-03-03T13:40:10Z"
   },
   {
-    "additions": 12,
-    "author": "danielalanbates",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #43881 ## Summary This PR fixes: glm-4v-9b loading failed ## Changes ``` src/transformers/configuration_utils.py | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) ``` ## Testing Please review the changes carefully. T\u2026",
-    "changed_files": 1,
+    "additions": 86,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title. It looks like some models (xlnet and kosmos2_5) and most audio models sometimes rely on the full previous input_ids to prepare inputs. Note that this cannot be compatible with restarting generation\u2026",
+    "changed_files": 12,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44196",
-    "created_at": "2026-02-21T04:41:02Z",
-    "deletions": 1,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44226",
+    "created_at": "2026-02-23T13:27:23Z",
+    "deletions": 66,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44196/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44196",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44196,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44226/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44226",
+    "labels": [],
+    "merged": true,
+    "number": 44226,
+    "review_comments_count": 17,
     "state": "closed",
-    "title": "Fix #43881: glm-4v-9b loading failed",
-    "updated_at": "2026-02-23T09:45:03Z"
+    "title": "[generate] Always pass full input_ids in `prepare_inputs_for_generation`",
+    "updated_at": "2026-02-24T10:45:49Z"
   },
   {
-    "additions": 2,
-    "author": "danielalanbates",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #44062 ## Summary This PR fixes: TypeError: tokenizers.AddedToken() got multiple values for keyword argument 'special' ## Changes ``` src/transformers/tokenization_utils_base.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-\u2026",
-    "changed_files": 1,
+    "additions": 169,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per title, we weren't running these test for some time because they were being filtered into `non-model` tests. But `non-model` doesn't run tests that are marked as generation, so I moved it back to `generation`\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44195",
-    "created_at": "2026-02-21T04:38:14Z",
-    "deletions": 2,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44225",
+    "created_at": "2026-02-23T12:09:40Z",
+    "deletions": 270,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44195/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44195",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44195,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44225/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44225",
+    "labels": [],
+    "merged": true,
+    "number": 44225,
+    "review_comments_count": 18,
     "state": "closed",
-    "title": "Fix #44062: TypeError: tokenizers.AddedToken() got multiple values for k",
-    "updated_at": "2026-02-23T14:10:30Z"
+    "title": "Fix generation integration tests",
+    "updated_at": "2026-02-25T09:19:39Z"
   },
   {
-    "additions": 16,
-    "author": "danielalanbates",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #44075 ## Summary This PR fixes: Optimizer SGD args are not used ## Changes ``` src/transformers/trainer_optimizer.py | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) ``` ## Testing Please review the changes carefully. The fix\u2026",
-    "changed_files": 1,
+    "additions": 5,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Small fixes after https://github.com/huggingface/transformers/pull/44130. See https://huggingface.co/datasets/hf-internal-testing/transformers_daily_ci/raw/8785954cca2fdca181de0b9567059471bcadb959/2026-02-21/ci_resu\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44194",
-    "created_at": "2026-02-21T04:35:53Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44224",
+    "created_at": "2026-02-23T10:48:19Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44194/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44194",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44224/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44224",
+    "labels": [],
     "merged": false,
-    "number": 44194,
+    "number": 44224,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix #44075: Optimizer SGD args are not used",
-    "updated_at": "2026-02-23T14:10:20Z"
+    "title": "Small fixes",
+    "updated_at": "2026-02-24T10:06:14Z"
   },
   {
-    "additions": 2,
-    "author": "danielalanbates",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #43986 ## Summary This PR fixes: Confusing crash when loading a video model through AutoProcessor without torchvision installed ## Changes ``` src/transformers/models/auto/video_processing_auto.py | 2 ++ 1 file changed, 2 insertions(\u2026",
+    "additions": 1,
+    "author": "albertvillanova",
+    "author_association": "MEMBER",
+    "body_excerpt": "Fix type of `TrainingArguments.logging_steps`. This PR makes a minor update to the `TrainingArguments` class, so `logging_steps` parameter accepts both integers and floats, rather than only floats. Note these are the expected types in the\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44193",
-    "created_at": "2026-02-21T04:34:37Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44223",
+    "created_at": "2026-02-23T08:50:04Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44193/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44193",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44223/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44223",
+    "labels": [],
     "merged": false,
-    "number": 44193,
+    "number": 44223,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix #43986: Confusing crash when loading a video model through AutoProce",
-    "updated_at": "2026-02-23T09:46:15Z"
+    "title": "Fix type of TrainingArguments.logging_steps",
+    "updated_at": "2026-02-23T09:08:18Z"
   },
   {
-    "additions": 3,
-    "author": "danielalanbates",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #44079 ## Summary This PR fixes: `ModelOutput` keys aren't correctly assigned if key was previously None ## Changes ``` src/transformers/utils/generic.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) ``` ## Testing Please r\u2026",
+    "additions": 1,
+    "author": "matisgagneux21",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary - fix a typo in the Italian add-new-model guide: `docstirng` -> `docstring`. ## Why Small docs quality fix that avoids confusion for readers following the contribution guide.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44192",
-    "created_at": "2026-02-21T04:33:52Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44219",
+    "created_at": "2026-02-23T00:43:59Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44192/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44192",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44192,
+    "files_url": "https://github.com/huggingface/transformers/pull/44219/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44219",
+    "labels": [],
+    "merged": true,
+    "number": 44219,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix #44079: `ModelOutput` keys aren't correctly assigned if key was prev",
-    "updated_at": "2026-02-23T14:10:14Z"
+    "title": "Docs(it): fix typo in docstring wording",
+    "updated_at": "2026-02-23T15:04:51Z"
   },
   {
-    "additions": 95,
-    "author": "danielalanbates",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #44155 ## Summary This PR fixes: [AudioFlamingo3] Batched inference produces incorrect results due to embedding/token leak between tracks ## Changes ``` .../audioflamingo3/modeling_audioflamingo3.py | 51 +++++++++++++++++++--- .../au\u2026",
-    "changed_files": 3,
+    "additions": 1,
+    "author": "matisgagneux21",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary - fix a typo in the Italian migration guide install command: - `stentencepiece` -> `sentencepiece` ## Why The current command fails if copied as-is. This makes the installation snippet runnable for users reading the Italian docs.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44191",
-    "created_at": "2026-02-21T04:32:30Z",
-    "deletions": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44218",
+    "created_at": "2026-02-23T00:32:49Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44191/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44191",
-    "labels": [
-      "Audio"
-    ],
-    "merged": false,
-    "number": 44191,
+    "files_url": "https://github.com/huggingface/transformers/pull/44218/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44218",
+    "labels": [],
+    "merged": true,
+    "number": 44218,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix #44155: [AudioFlamingo3] Batched inference produces incorrect result",
-    "updated_at": "2026-03-19T16:16:17Z"
+    "title": "Docs(it): fix typo in sentencepiece install command",
+    "updated_at": "2026-02-23T15:05:17Z"
   },
   {
-    "additions": 3,
-    "author": "excepshenal",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Under fp16_full_eval or bf16_full_eval, still don't move model to device if using another dist train backend. This is causing bugs with FSDP2 + bf16_full_eval. The dist train backend would still be in charge of movi\u2026",
+    "additions": 1,
+    "author": "matisgagneux21",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary - Update the Italian migration guide to avoid pointing users to the deprecated `grouped_entities` flag. - Clarify that `aggregation_strategy` is the current option (with a note that it was previously `grouped_entities`). ## Why\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44189",
-    "created_at": "2026-02-21T00:06:16Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44217",
+    "created_at": "2026-02-23T00:10:48Z",
     "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44189/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44189",
+    "files_url": "https://github.com/huggingface/transformers/pull/44217/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44217",
     "labels": [],
     "merged": false,
-    "number": 44189,
+    "number": 44217,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "fix: don't move model to device under other dist train backends",
-    "updated_at": "2026-02-21T00:06:16Z"
+    "state": "closed",
+    "title": "Docs(it): update deprecated grouped_entities reference",
+    "updated_at": "2026-02-23T14:56:19Z"
   },
   {
-    "additions": 3,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following issues were identified and fixed in this PR: \u2192 The NER/token classification issue and the downstream bug uncovered in the batched preprocessing use case with `LayoutLMv2Tokenizer`. \u2192 **Reasoning:** T\u2026",
+    "additions": 13,
+    "author": "nikste",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Automated fix for #44016: Syntax error in Transformer section 3 (Transformers, what can they do?) notebook Fixes #44016 ## What does this PR do? This PR addresses issue #44016 by implementing the fix described in the issue. ---\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44187",
-    "created_at": "2026-02-20T20:02:04Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44216",
+    "created_at": "2026-02-22T23:40:56Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44187/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44187",
+    "files_url": "https://github.com/huggingface/transformers/pull/44216/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44216",
     "labels": [],
-    "merged": true,
-    "number": 44187,
+    "merged": false,
+    "number": 44216,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(models): Fix LayoutLMv2 NER crash and broken batched truncation/padding",
-    "updated_at": "2026-02-23T10:30:51Z"
+    "title": "Fix #44016: Syntax error in Transformer section 3 (Transformers, what can they do?) notebook",
+    "updated_at": "2026-02-22T23:50:57Z"
   },
   {
-    "additions": 361,
-    "author": "stevhliu",
-    "author_association": "MEMBER",
-    "body_excerpt": "part 1 of refactoring the `Trainer` docs - restructure the `toctree` a bit to accommodate new sections and docs - slim down `trainer.md` to be a clearer entry point (will expand the `## Next steps` section as we continue for better navigat\u2026",
-    "changed_files": 7,
+    "additions": 187,
+    "author": "jmriosal",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "# What does this PR do? Add sequence classification capabilities to the family of Granite models (Granite, GraniteMoe, GraniteMoeHybrid, and GraniteMoeShared). Fixes #44214, #35720 ## Why The Granite models currently only have the base mod\u2026",
+    "changed_files": 17,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44185",
-    "created_at": "2026-02-20T19:25:07Z",
-    "deletions": 578,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44215",
+    "created_at": "2026-02-22T23:24:43Z",
+    "deletions": 13,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44185/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44185",
+    "files_url": "https://github.com/huggingface/transformers/pull/44215/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44215",
     "labels": [],
-    "merged": true,
-    "number": 44185,
-    "review_comments_count": 19,
-    "state": "closed",
-    "title": "[docs] trainer part 1",
-    "updated_at": "2026-02-24T21:18:42Z"
+    "merged": false,
+    "number": 44215,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Add sequence classification capability to Granite models",
+    "updated_at": "2026-02-24T20:39:37Z"
   },
   {
-    "additions": 191,
-    "author": "mariam851",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR implements the initial architecture for CircuitGPT (based on OpenAI's research), as discussed in #44121. Key implementations: SparseLinear: Custom layer with Top-K weight sparsity logic. CircuitGpt Components: Attention, MLP, and C\u2026",
-    "changed_files": 3,
+    "additions": 70,
+    "author": "parthchopra07",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "# What does this PR do? This PR refreshes the BEiT model documentation to align it with the current Transformers vision docs style and features. It updates the usage examples, clarifies configuration details, and improves the resources sec\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44184",
-    "created_at": "2026-02-20T16:58:27Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44213",
+    "created_at": "2026-02-22T18:32:16Z",
+    "deletions": 29,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44184/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44184",
+    "files_url": "https://github.com/huggingface/transformers/pull/44213/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44213",
     "labels": [],
     "merged": false,
-    "number": 44184,
+    "number": 44213,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "feat: add OpenAI CircuitGPT core architecture and sparse linear layers",
-    "updated_at": "2026-02-20T17:18:44Z"
+    "state": "closed",
+    "title": "Update BEiT model card",
+    "updated_at": "2026-02-28T14:33:57Z"
   },
   {
     "additions": 1,
-    "author": "Rocketknight1",
-    "author_association": "MEMBER",
-    "body_excerpt": "Our code has some references to the `grouped_entities` arg to the token classification pipeline, but this is no longer usable. This PR cleans them up entirely! Fixes #44016",
-    "changed_files": 2,
+    "author": "alexandercarruthers",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Updates a broken link in the summarization guide. https://huggingface.co/docs/transformers/tasks/summarization https://huggingface.co/billsum/datasets results in a 404. New URL is https://huggingface.co/datasets/Fis\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44182",
-    "created_at": "2026-02-20T15:28:26Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44212",
+    "created_at": "2026-02-22T18:02:43Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44182/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44182",
+    "files_url": "https://github.com/huggingface/transformers/pull/44212/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44212",
     "labels": [],
     "merged": true,
-    "number": 44182,
+    "number": 44212,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Remove refs to grouped_entities",
-    "updated_at": "2026-02-24T16:07:24Z"
+    "title": "Update 404ing BillSum dataset URL on Summarization Task guide",
+    "updated_at": "2026-02-23T14:46:11Z"
   },
   {
-    "additions": 898,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title! Follow up of https://github.com/huggingface/transformers/pull/44130 and https://github.com/huggingface/transformers/pull/44226. Finally remove the `cache_position` everywhere (not ALL models, but a\u2026",
-    "changed_files": 169,
+    "additions": 10,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## Fix for #44016 The `grouped_entities` parameter in `TokenClassificationPipeline._sanitize_parameters` was removed without a deprecation period, causing a `TypeError` when users pass `grouped_entities=True` to the `pipeline()` call (as s\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 11,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44181",
-    "created_at": "2026-02-20T15:24:39Z",
-    "deletions": 2698,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44211",
+    "created_at": "2026-02-22T17:04:50Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44181/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44181",
-    "labels": [],
-    "merged": true,
-    "number": 44181,
-    "review_comments_count": 32,
+    "files_url": "https://github.com/huggingface/transformers/pull/44211/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44211",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44211,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[core] \ud83d\udea8 Completely remove cache positions",
-    "updated_at": "2026-03-04T18:08:42Z"
+    "title": "Add backward compatibility for deprecated grouped_entities parameter",
+    "updated_at": "2026-02-23T16:26:02Z"
   },
   {
-    "additions": 28,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes a flaky test in IdeficsForVisionText2TextTest::test_generate_continue_from_inputs_embeds. The flakiness can be reproduced with: ``` pytest -q -p no:rerunfailures --flake-finder --flake-runs=20 \\ tests/models/i\u2026",
+    "additions": 1,
+    "author": "nightcityblade",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fixes #44206 ## Problem PR #43769 (\"Add Voxtral Realtime\") added a `center` parameter to `LasrFeatureExtractor.__call__()` and passed it to `_torch_extract_fbank_features()`, but that method does not accept it. This causes a `TypeError` on\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 10,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44180",
-    "created_at": "2026-02-20T14:30:46Z",
-    "deletions": 2,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44210",
+    "created_at": "2026-02-22T16:06:16Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44180/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44180",
+    "files_url": "https://github.com/huggingface/transformers/pull/44210/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44210",
     "labels": [],
-    "merged": true,
-    "number": 44180,
+    "merged": false,
+    "number": 44210,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(flaky): idefics generate cache flake",
-    "updated_at": "2026-02-26T16:18:18Z"
+    "title": "fix(lasr): remove spurious center arg from _torch_extract_fbank_features call",
+    "updated_at": "2026-02-23T14:32:41Z"
   },
   {
-    "additions": 27,
-    "author": "itazap",
-    "author_association": "MEMBER",
-    "body_excerpt": "Models with incorrect tokenizer_class in tokenization_config.json that should use TokenziersBackend",
-    "changed_files": 1,
+    "additions": 197,
+    "author": "paipeline",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? This PR fixes a critical bug in LayoutLMv2Tokenizer where passing `word_labels` for NER token classification tasks would crash with `AttributeError`. The issue was that `word_ids` and `sequence_ids` were being acce\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44179",
-    "created_at": "2026-02-20T13:51:44Z",
-    "deletions": 0,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44209",
+    "created_at": "2026-02-22T14:37:25Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44179/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44179",
+    "files_url": "https://github.com/huggingface/transformers/pull/44209/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44209",
     "labels": [],
-    "merged": true,
-    "number": 44179,
-    "review_comments_count": 1,
+    "merged": false,
+    "number": 44209,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Models with incorrect tokenizer_class in tokenization_config.json tha\u2026",
-    "updated_at": "2026-02-23T08:33:13Z"
+    "title": "Fix LayoutLMv2Tokenizer NER crashes with word_labels",
+    "updated_at": "2026-02-23T10:30:26Z"
   },
   {
-    "additions": 2993,
-    "author": "ebezzam",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Re-opening https://github.com/huggingface/transformers/pull/37868 TODO - [x] recompute expected outputs - [x] passthrough code given new conventions - [x] check for unused code paths / configuration parameters Origi\u2026",
-    "changed_files": 27,
+    "additions": 1,
+    "author": "ainergiz",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR fixes a LASR regression introduced in `#43769` (released in `v5.2.0`). `LasrFeatureExtractor.__call__` passes `center` into `_torch_extract_fbank_features(...)`, but `_torch_extract_fbank_features` did not a\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44178",
-    "created_at": "2026-02-20T12:36:21Z",
-    "deletions": 50,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44178/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44178",
-    "labels": [],
-    "merged": false,
-    "number": 44178,
-    "review_comments_count": 9,
-    "state": "open",
-    "title": "Add xcodec2 model",
-    "updated_at": "2026-04-13T14:26:21Z"
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44207",
+    "created_at": "2026-02-21T20:56:49Z",
+    "deletions": 70,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44207/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44207",
+    "labels": [],
+    "merged": true,
+    "number": 44207,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix LASR feature extractor regression from invalid center argument",
+    "updated_at": "2026-02-23T10:01:35Z"
   },
   {
-    "additions": 41,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "As per title, spiritual successor to #44081 Why? Because as is - Only defaults for fa2/fa3, not on other requested kernels - Limits implementations to one kernel/implementation while I suspect that there will be multiple viable versions (i\u2026",
-    "changed_files": 7,
+    "additions": 1,
+    "author": "nightcityblade",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes #44186 `LayoutLMv2Tokenizer.__init__` passes `only_label_first_subword` to `super().__init__()` but never stores it as `self.only_label_first_subword`. This causes an `AttributeError` when `word_labels` is pa\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44177",
-    "created_at": "2026-02-20T12:13:30Z",
-    "deletions": 71,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44204",
+    "created_at": "2026-02-21T16:06:46Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44177/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44177",
+    "files_url": "https://github.com/huggingface/transformers/pull/44204/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44204",
     "labels": [],
-    "merged": true,
-    "number": 44177,
-    "review_comments_count": 2,
+    "merged": false,
+    "number": 44204,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[`Flash Attn`] Enable compatible implementations",
-    "updated_at": "2026-02-20T12:43:35Z"
+    "title": "fix(layoutlmv2): store only_label_first_subword attribute in tokenizer",
+    "updated_at": "2026-02-23T10:30:19Z"
   },
   {
-    "additions": 271,
-    "author": "vasqu",
-    "author_association": "MEMBER",
-    "body_excerpt": "Our kernel loading is incompatible with the original packages as they do not expose the same import structure: - Kernels seem to expose things in the init (and not in the original path) - Original packages seem to expose only within their\u2026",
-    "changed_files": 14,
+    "additions": 22,
+    "author": "nightcityblade",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Fixes #44075 `_get_sgd()`, `_get_adagrad()`, and `_get_rmsprop()` in `trainer_optimizer.py` only returned `ctx.optimizer_kwargs` (which contains just `lr`), completely ignoring `ctx.optim_args`. This meant that parameters specif\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44176",
-    "created_at": "2026-02-20T11:36:01Z",
-    "deletions": 124,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44203",
+    "created_at": "2026-02-21T15:12:17Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44176/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44176",
+    "files_url": "https://github.com/huggingface/transformers/pull/44203/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44203",
     "labels": [],
     "merged": true,
-    "number": 44176,
-    "review_comments_count": 2,
+    "number": 44203,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[`Mamba`] Fix kernel loading",
-    "updated_at": "2026-02-20T16:19:06Z"
+    "title": "fix(trainer): pass optim_args to SGD, Adagrad, and RMSprop optimizers",
+    "updated_at": "2026-02-25T16:04:20Z"
   },
   {
-    "additions": 1,
-    "author": "itazap",
-    "author_association": "MEMBER",
-    "body_excerpt": "add jamba tokenizer mapping to PreTrainedTokenizerFast (v4/v5 BC) for vllm: https://buildkite.com/vllm/ci/builds/52260/steps/canvas?sid=019c76ad-c8f2-4e59-a2f4-5f3b5bbc204c&tab=output",
-    "changed_files": 1,
+    "additions": 63,
+    "author": "GS-GOAT",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? When `torch.compile` is used, [_ignore_bidirectional_mask_sdpa](cci:1://file:///c:/Users/BIT/Desktop/proj/gitrepo_clones/transformers/src/transformers/masking_utils.py:303:0-338:16) behaves differently than in eager\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44175",
-    "created_at": "2026-02-20T11:00:18Z",
-    "deletions": 0,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44202",
+    "created_at": "2026-02-21T13:45:15Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44175/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44175",
+    "files_url": "https://github.com/huggingface/transformers/pull/44202/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44202",
     "labels": [],
     "merged": false,
-    "number": 44175,
+    "number": 44202,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "add jamba tokenizer mapping to PreTrainedTokenizerFast (v4/v5 BC)",
-    "updated_at": "2026-02-20T16:19:31Z"
+    "title": "Fix: bidirectional mask skip when attention dropout is active (#44188)",
+    "updated_at": "2026-03-09T10:31:41Z"
   },
   {
-    "additions": 1367,
+    "additions": 18,
     "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "This draft expands `utils/check_modeling_structure.py` into a rule-driven linter for model code, with new checks and tests, while keeping runtime very low. Key features: - The checker is intentionally AST-only (no heavy imports/execution),\u2026",
-    "changed_files": 4,
+    "body_excerpt": "# What does this PR do? main is currently failing with ``` FAILED tests/models/higgs_audio_v2/test_modeling_higgs_audio_v2.py::HiggsAudioV2ModelTest::test_generate_compilation_all_outputs - AssertionError: Lists differ: [torch.Size([2, 15,\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44174",
-    "created_at": "2026-02-20T10:38:11Z",
-    "deletions": 24,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44201",
+    "created_at": "2026-02-21T10:03:41Z",
+    "deletions": 13,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44174/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44174",
+    "files_url": "https://github.com/huggingface/transformers/pull/44201/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44201",
     "labels": [],
     "merged": true,
-    "number": 44174,
-    "review_comments_count": 38,
+    "number": 44201,
+    "review_comments_count": 3,
     "state": "closed",
-    "title": "Expand model-structure lint rules with a fast AST-based, ruff-like framework",
-    "updated_at": "2026-03-12T06:42:21Z"
+    "title": "fix: HiggsAudioV2 cached decode inputs in compiled generation",
+    "updated_at": "2026-02-23T12:39:19Z"
   },
   {
-    "additions": 20,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "Fixes flaky GLM OCR generation behavior when 2D `position_ids` are passed explicitly. Reproducible locally with: ``` pytest tests/models/glm_ocr/test_modeling_glm_ocr.py::GlmOcrModelTest::test_generate_with_and_without_position_ids --flake\u2026",
+    "additions": 3,
+    "author": "pragnyanramtha",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Fixes #43782 The `weights_only` parameter passed to `from_pretrained()` was not being forwarded to `load_state_dict()` when loading `.bin` checkpoint files in the non-DeepSpeed code path. This caused `weights_only` to always default to `Tr\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 14,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44173",
-    "created_at": "2026-02-20T09:28:48Z",
-    "deletions": 0,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44200",
+    "created_at": "2026-02-21T06:24:17Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44173/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44173",
-    "labels": [],
-    "merged": true,
-    "number": 44173,
-    "review_comments_count": 10,
+    "files_url": "https://github.com/huggingface/transformers/pull/44200/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44200",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44200,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(flaky): `test_generate_with_and_without_position_ids` in GLM ORC",
-    "updated_at": "2026-02-20T19:06:19Z"
+    "title": "fix: propagate `weights_only` param to `load_state_dict` in .bin loading path (#43782)",
+    "updated_at": "2026-02-23T14:20:12Z"
   },
   {
-    "additions": 2,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "## Summary I've noticed `test_synthidtext_watermark_processor_distributional_convergence_*` was our slowest tests in CircleCI This PR speeds up the slowest SynthID distributional convergence tests by optimizing a hot path in SynthIDTextWat\u2026",
+    "additions": 3,
+    "author": "gowthamr-tech",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? This PR fixes an issue in `run_image_classification_no_trainer.py` where the script always loaded `dataset_name` (e.g., CIFAR10) even when `--train_dir` or `--validation_dir` was provided. Now, when local dataset d\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44172",
-    "created_at": "2026-02-20T08:59:30Z",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44199",
+    "created_at": "2026-02-21T06:03:29Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44172/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44172",
+    "files_url": "https://github.com/huggingface/transformers/pull/44199/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44199",
     "labels": [],
     "merged": true,
-    "number": 44172,
+    "number": 44199,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "perf: Optimize SynthID logits processor batch index construction",
-    "updated_at": "2026-02-27T09:32:43Z"
+    "title": "Fix local dataset loading priority in run_image_classification_no_tra\u2026",
+    "updated_at": "2026-02-24T15:10:17Z"
   },
   {
-    "additions": 2067,
-    "author": "lmaksym",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds TDT decoder support for Parakeet ASR models, extending the existing CTC-only implementation. It incorporates the initial TDT integration work from [#41545](https://github.com/huggingface/transformers/pu\u2026",
-    "changed_files": 32,
+    "additions": 71,
+    "author": "danielalanbates",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #43975 ## Summary This PR fixes: `deepseek-ai/deepseek-coder-6.7b-instruct` incorrectly detokenizes in v5 ## Changes ``` src/transformers/tokenization_utils_tokenizers.py | 12 ++++- tests/models/llama/test_tokenization_llama.py | 60\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 16,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44171",
-    "created_at": "2026-02-20T08:44:46Z",
-    "deletions": 283,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44198",
+    "created_at": "2026-02-21T04:54:47Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44171/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44171",
+    "files_url": "https://github.com/huggingface/transformers/pull/44198/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44198",
     "labels": [
-      "New model",
-      "Audio"
+      "Code agent slop"
     ],
     "merged": false,
-    "number": 44171,
-    "review_comments_count": 73,
-    "state": "open",
-    "title": "Parakeet tdt",
-    "updated_at": "2026-04-15T15:47:45Z"
+    "number": 44198,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix #43975: `deepseek-ai/deepseek-coder-6.7b-instruct` incorrectly detok",
+    "updated_at": "2026-02-23T14:10:47Z"
   },
   {
-    "additions": 74,
-    "author": "veeceey",
+    "additions": 37,
+    "author": "danielalanbates",
     "author_association": "NONE",
-    "body_excerpt": "## Summary - Adds `GitForCausalLM` to `MODEL_FOR_VISUAL_QUESTION_ANSWERING_MAPPING_NAMES` so GIT can be used with the `visual-question-answering` pipeline - Filters tokenizer outputs in VQA pipeline `preprocess` to only pass keys accepted\u2026",
-    "changed_files": 4,
+    "body_excerpt": "Fixes #43937 ## Summary This PR fixes: [GLM-5] ValueError: GenerationConfig is invalid ## Changes ``` src/transformers/generation/configuration_utils.py | 13 +++++++++++- src/transformers/modeling_utils.py | 2 +- tests/generation/test_conf\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44170",
-    "created_at": "2026-02-20T08:28:05Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44197",
+    "created_at": "2026-02-21T04:47:32Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44170/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44170",
+    "files_url": "https://github.com/huggingface/transformers/pull/44197/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44197",
     "labels": [],
     "merged": false,
-    "number": 44170,
+    "number": 44197,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add GIT model support in VQA pipeline",
-    "updated_at": "2026-02-20T09:34:31Z"
+    "title": "Fix #43937: [GLM-5] ValueError: GenerationConfig is invalid",
+    "updated_at": "2026-02-23T09:42:54Z"
   },
   {
-    "additions": 415,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "Initial ty integration. To avoid a gigantic, risky patch, let's start with a baby step where we add the tooling to make repo-check and activate it on a subset of the repo. That gives us a human-readable patch, and allows us to get conforta\u2026",
-    "changed_files": 25,
+    "additions": 12,
+    "author": "danielalanbates",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #43881 ## Summary This PR fixes: glm-4v-9b loading failed ## Changes ``` src/transformers/configuration_utils.py | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) ``` ## Testing Please review the changes carefully. T\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 12,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44167",
-    "created_at": "2026-02-20T07:39:44Z",
-    "deletions": 210,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44196",
+    "created_at": "2026-02-21T04:41:02Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44167/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44167",
-    "labels": [],
-    "merged": true,
-    "number": 44167,
-    "review_comments_count": 2,
+    "files_url": "https://github.com/huggingface/transformers/pull/44196/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44196",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44196,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "chore(typing): initial ty integration",
-    "updated_at": "2026-02-20T19:08:09Z"
+    "title": "Fix #43881: glm-4v-9b loading failed",
+    "updated_at": "2026-02-23T09:45:03Z"
   },
   {
-    "additions": 73,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "The CI does not output useful info on this flaky test - `tests.models.olmo.test_modeling_olmo.OlmoModelTest testMethod=test_generate_with_static_cache` and makes it harder to determine the root problem when not reproducible locally. This p\u2026",
-    "changed_files": 8,
+    "additions": 2,
+    "author": "danielalanbates",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #44062 ## Summary This PR fixes: TypeError: tokenizers.AddedToken() got multiple values for keyword argument 'special' ## Changes ``` src/transformers/tokenization_utils_base.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44166",
-    "created_at": "2026-02-20T07:20:15Z",
-    "deletions": 61,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44195",
+    "created_at": "2026-02-21T04:38:14Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44166/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44166",
-    "labels": [],
-    "merged": true,
-    "number": 44166,
-    "review_comments_count": 4,
+    "files_url": "https://github.com/huggingface/transformers/pull/44195/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44195",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44195,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Improve `has_similar_generate_outputs` assertions",
-    "updated_at": "2026-02-27T08:26:13Z"
+    "title": "Fix #44062: TypeError: tokenizers.AddedToken() got multiple values for k",
+    "updated_at": "2026-02-23T14:10:30Z"
   },
   {
-    "additions": 29,
-    "author": "alexmalyshev",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "With Python 3.14 and PEP 649, you can no longer expect `cls.__dict__[\"__annotations__\"]` to contain annotations for the exact class, it will be loaded lazily and can make it seem like the class doesn't have any annotations. The recommended\u2026",
+    "additions": 16,
+    "author": "danielalanbates",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #44075 ## Summary This PR fixes: Optimizer SGD args are not used ## Changes ``` src/transformers/trainer_optimizer.py | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) ``` ## Testing Please review the changes carefully. The fix\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44165",
-    "created_at": "2026-02-20T04:47:07Z",
-    "deletions": 14,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44194",
+    "created_at": "2026-02-21T04:35:53Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44165/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44165",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44194/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44194",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44165,
+    "number": 44194,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix how PreTrainedModel checks annotations on Python 3.14+",
-    "updated_at": "2026-03-20T17:31:11Z"
+    "title": "Fix #44075: Optimizer SGD args are not used",
+    "updated_at": "2026-02-23T14:10:20Z"
   },
   {
-    "additions": 7,
-    "author": "lhallee",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes: https://github.com/huggingface/transformers/issues/44162 @ArthurZucker @Cyrilvallez",
-    "changed_files": 2,
+    "additions": 2,
+    "author": "danielalanbates",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #43986 ## Summary This PR fixes: Confusing crash when loading a video model through AutoProcessor without torchvision installed ## Changes ``` src/transformers/models/auto/video_processing_auto.py | 2 ++ 1 file changed, 2 insertions(\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44163",
-    "created_at": "2026-02-19T21:44:25Z",
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44193",
+    "created_at": "2026-02-21T04:34:37Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44163/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44163",
-    "labels": [],
-    "merged": true,
-    "number": 44163,
+    "files_url": "https://github.com/huggingface/transformers/pull/44193/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44193",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44193,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "ESM2 attention_mask and token_dropout fix",
-    "updated_at": "2026-02-20T15:17:31Z"
+    "title": "Fix #43986: Confusing crash when loading a video model through AutoProce",
+    "updated_at": "2026-02-23T09:46:15Z"
   },
   {
-    "additions": 379,
-    "author": "cogniera",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "What does this PR do? This PR refactors the LongT5 model to use the @capture_outputs and @can_return_tuple decorators for standardized output handling across the model stack. The refactor removes manual handling of: output_attentions outpu\u2026",
+    "additions": 3,
+    "author": "danielalanbates",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #44079 ## Summary This PR fixes: `ModelOutput` keys aren't correctly assigned if key was previously None ## Changes ``` src/transformers/utils/generic.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) ``` ## Testing Please r\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44161",
-    "created_at": "2026-02-19T20:46:49Z",
-    "deletions": 170,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44192",
+    "created_at": "2026-02-21T04:33:52Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44161/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44161",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44192/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44192",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44161,
+    "number": 44192,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Refactor LongT5 to use @capture_outputs and @can_return_tuple decorators for unified output handling (Fixes #43979)",
-    "updated_at": "2026-02-20T17:28:12Z"
+    "state": "closed",
+    "title": "Fix #44079: `ModelOutput` keys aren't correctly assigned if key was prev",
+    "updated_at": "2026-02-23T14:10:14Z"
   },
   {
-    "additions": 2104,
-    "author": "molbap",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Hey yall, I started porting the pi0 model so Transformers can be a backend for VLAs directly. I checked it against LeRobot on fix/lerobot_openpi: outputs seem to match and for sure lerobot/pi0_base loads cleanly (no\u2026",
-    "changed_files": 22,
+    "additions": 95,
+    "author": "danielalanbates",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #44155 ## Summary This PR fixes: [AudioFlamingo3] Batched inference produces incorrect results due to embedding/token leak between tracks ## Changes ``` .../audioflamingo3/modeling_audioflamingo3.py | 51 +++++++++++++++++++--- .../au\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 13,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44160",
-    "created_at": "2026-02-19T17:16:29Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44191",
+    "created_at": "2026-02-21T04:32:30Z",
+    "deletions": 11,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44160/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44160",
+    "files_url": "https://github.com/huggingface/transformers/pull/44191/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44191",
     "labels": [
-      "New model"
+      "Audio"
     ],
-    "merged": true,
-    "number": 44160,
-    "review_comments_count": 32,
+    "merged": false,
+    "number": 44191,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add model lerobot PI0 to transformers",
-    "updated_at": "2026-03-16T10:23:14Z"
+    "title": "Fix #44155: [AudioFlamingo3] Batched inference produces incorrect result",
+    "updated_at": "2026-03-19T16:16:17Z"
   },
   {
-    "additions": 67,
-    "author": "samuelleecong",
+    "additions": 3,
+    "author": "excepshenal",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary Closes #28103 - Refactor `OwlViTAttention` to use `ALL_ATTENTION_FUNCTIONS` for dynamic attention backend dispatch (same pattern as CLIP) - Add `eager_attention_forward` standalone function with the standardized interface - Resh\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Under fp16_full_eval or bf16_full_eval, still don't move model to device if using another dist train backend. This is causing bugs with FSDP2 + bf16_full_eval. The dist train backend would still be in charge of movi\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44159",
-    "created_at": "2026-02-19T16:31:44Z",
-    "deletions": 61,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44189",
+    "created_at": "2026-02-21T00:06:16Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44159/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44159",
+    "files_url": "https://github.com/huggingface/transformers/pull/44189/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44189",
     "labels": [],
     "merged": false,
-    "number": 44159,
+    "number": 44189,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add SDPA and Flash Attention support for OWL-ViT",
-    "updated_at": "2026-02-24T12:53:10Z"
+    "title": "fix: don't move model to device under other dist train backends",
+    "updated_at": "2026-02-21T00:06:16Z"
   },
   {
-    "additions": 141,
-    "author": "leopold-tzafon",
+    "additions": 3,
+    "author": "harshaljanjani",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# Fix issue where `use_cache=False`, corrupts model Qwen3vl output. Tested with: ``` import torch from transformers import Qwen3VLForConditionalGeneration, AutoProcessor MODEL_NAME = \"Qwen/Qwen3-VL-4B-Instruct\" DEVICE = \"cuda\" DTYPE = torc\u2026",
-    "changed_files": 9,
+    "body_excerpt": "### What does this PR do? The following issues were identified and fixed in this PR: \u2192 The NER/token classification issue and the downstream bug uncovered in the batched preprocessing use case with `LayoutLMv2Tokenizer`. \u2192 **Reasoning:** T\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 19,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44158",
-    "created_at": "2026-02-19T15:45:13Z",
-    "deletions": 36,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44187",
+    "created_at": "2026-02-20T20:02:04Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44158/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44158",
+    "files_url": "https://github.com/huggingface/transformers/pull/44187/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44187",
     "labels": [],
     "merged": true,
-    "number": 44158,
+    "number": 44187,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix bug with position_ids on qwen3-vl models, such that position_ids include text position",
-    "updated_at": "2026-02-23T14:53:33Z"
+    "title": "fix(models): Fix LayoutLMv2 NER crash and broken batched truncation/padding",
+    "updated_at": "2026-02-23T10:30:51Z"
+  },
+  {
+    "additions": 361,
+    "author": "stevhliu",
+    "author_association": "MEMBER",
+    "body_excerpt": "part 1 of refactoring the `Trainer` docs - restructure the `toctree` a bit to accommodate new sections and docs - slim down `trainer.md` to be a clearer entry point (will expand the `## Next steps` section as we continue for better navigat\u2026",
+    "changed_files": 7,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44185",
+    "created_at": "2026-02-20T19:25:07Z",
+    "deletions": 578,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44185/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44185",
+    "labels": [],
+    "merged": true,
+    "number": 44185,
+    "review_comments_count": 19,
+    "state": "closed",
+    "title": "[docs] trainer part 1",
+    "updated_at": "2026-02-24T21:18:42Z"
   },
   {
-    "additions": 689,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per title, gets rid of `if/else` per attn implementation",
-    "changed_files": 24,
+    "additions": 191,
+    "author": "mariam851",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "This PR implements the initial architecture for CircuitGPT (based on OpenAI's research), as discussed in #44121. Key implementations: SparseLinear: Custom layer with Top-K weight sparsity logic. CircuitGpt Components: Attention, MLP, and C\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44157",
-    "created_at": "2026-02-19T14:49:49Z",
-    "deletions": 834,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44184",
+    "created_at": "2026-02-20T16:58:27Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44157/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44157",
+    "files_url": "https://github.com/huggingface/transformers/pull/44184/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44184",
     "labels": [],
     "merged": false,
-    "number": 44157,
-    "review_comments_count": 10,
+    "number": 44184,
+    "review_comments_count": 0,
     "state": "open",
-    "title": "Use correct mask for packed inputs in Qwen-VL ",
-    "updated_at": "2026-02-24T13:13:43Z"
+    "title": "feat: add OpenAI CircuitGPT core architecture and sparse linear layers",
+    "updated_at": "2026-02-20T17:18:44Z"
   },
   {
-    "additions": 2,
-    "author": "SunMarc",
+    "additions": 1,
+    "author": "Rocketknight1",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? This PR fixes a minor error when using aqml quantization. We specified the wrong argument.",
+    "body_excerpt": "Our code has some references to the `grouped_entities` arg to the token classification pipeline, but this is no longer usable. This PR cleans them up entirely! Fixes #44016",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44156",
-    "created_at": "2026-02-19T14:35:38Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44182",
+    "created_at": "2026-02-20T15:28:26Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44156/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44156",
+    "files_url": "https://github.com/huggingface/transformers/pull/44182/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44182",
     "labels": [],
-    "merged": false,
-    "number": 44156,
+    "merged": true,
+    "number": 44182,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix aqml `modules_to_not_convert`",
-    "updated_at": "2026-03-27T16:50:02Z"
+    "title": "Remove refs to grouped_entities",
+    "updated_at": "2026-02-24T16:07:24Z"
   },
   {
-    "additions": 44,
-    "author": "Aatman09",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
-    "cluster_id": "cluster-43979-11",
-    "cluster_ids": [
-      "cluster-43979-11"
-    ],
-    "cluster_role": "member",
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44154",
-    "created_at": "2026-02-19T12:17:56Z",
-    "deletions": 52,
+    "additions": 898,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title! Follow up of https://github.com/huggingface/transformers/pull/44130 and https://github.com/huggingface/transformers/pull/44226. Finally remove the `cache_position` everywhere (not ALL models, but a\u2026",
+    "changed_files": 169,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 11,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44181",
+    "created_at": "2026-02-20T15:24:39Z",
+    "deletions": 2698,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44154/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44154",
+    "files_url": "https://github.com/huggingface/transformers/pull/44181/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44181",
     "labels": [],
-    "merged": false,
-    "number": 44154,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Refactored vits to match standardized output collection interface",
-    "updated_at": "2026-02-19T12:18:56Z"
+    "merged": true,
+    "number": 44181,
+    "review_comments_count": 32,
+    "state": "closed",
+    "title": "[core] \ud83d\udea8 Completely remove cache positions",
+    "updated_at": "2026-03-04T18:08:42Z"
   },
   {
-    "additions": 79,
-    "author": "IlyasMoutawwakil",
+    "additions": 28,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do? Fixes a flaky test in IdeficsForVisionText2TextTest::test_generate_continue_from_inputs_embeds. The flakiness can be reproduced with: ``` pytest -q -p no:rerunfailures --flake-finder --flake-runs=20 \\ tests/models/i\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44152",
-    "created_at": "2026-02-19T09:37:51Z",
-    "deletions": 45,
+    "comments_count": 10,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44180",
+    "created_at": "2026-02-20T14:30:46Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44152/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44152",
+    "files_url": "https://github.com/huggingface/transformers/pull/44180/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44180",
     "labels": [],
     "merged": true,
-    "number": 44152,
-    "review_comments_count": 2,
+    "number": 44180,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "AutoGrad support for grouped_mm fallback",
-    "updated_at": "2026-02-20T11:15:23Z"
+    "title": "fix(flaky): idefics generate cache flake",
+    "updated_at": "2026-02-26T16:18:18Z"
   },
   {
-    "additions": 58,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "Part of the ongoing effort to migrate models to standardized output collection (ref #43979). Replaces manual output_hidden_states/output_attentions/return_dict handling with @capture_outputs and @can_return_tuple decorators in the BioGPT m\u2026",
-    "changed_files": 2,
+    "additions": 27,
+    "author": "itazap",
+    "author_association": "MEMBER",
+    "body_excerpt": "Models with incorrect tokenizer_class in tokenization_config.json that should use TokenziersBackend",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44151",
-    "created_at": "2026-02-19T06:55:43Z",
-    "deletions": 134,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44179",
+    "created_at": "2026-02-20T13:51:44Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44151/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44151",
+    "files_url": "https://github.com/huggingface/transformers/pull/44179/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44179",
     "labels": [],
-    "merged": false,
-    "number": 44151,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44179,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Refactor BioGPT output tracing with @capture_outputs and @can_return_tuple",
-    "updated_at": "2026-03-03T00:30:01Z"
+    "title": "Models with incorrect tokenizer_class in tokenization_config.json tha\u2026",
+    "updated_at": "2026-02-23T08:33:13Z"
   },
   {
-    "additions": 22,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "Part of the ongoing effort to migrate models to standardized output collection (ref #43979). Replaces manual output_hidden_states/output_attentions/return_dict handling with @capture_outputs and @can_return_tuple decorators in the MPT mode\u2026",
-    "changed_files": 1,
+    "additions": 2993,
+    "author": "ebezzam",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Re-opening https://github.com/huggingface/transformers/pull/37868 TODO - [x] recompute expected outputs - [x] passthrough code given new conventions - [x] check for unused code paths / configuration parameters Origi\u2026",
+    "changed_files": 27,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44150",
-    "created_at": "2026-02-19T06:54:09Z",
-    "deletions": 73,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44178",
+    "created_at": "2026-02-20T12:36:21Z",
+    "deletions": 50,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44150/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44150",
+    "files_url": "https://github.com/huggingface/transformers/pull/44178/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44178",
     "labels": [],
     "merged": false,
-    "number": 44150,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Refactor MPT output tracing with @capture_outputs and @can_return_tuple",
-    "updated_at": "2026-03-03T00:30:02Z"
+    "number": 44178,
+    "review_comments_count": 9,
+    "state": "open",
+    "title": "Add xcodec2 model",
+    "updated_at": "2026-04-13T14:26:21Z"
   },
   {
-    "additions": 85,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "Part of the ongoing effort to migrate models to standardized output collection (ref #43979). Replaces manual output_hidden_states/output_attentions/return_dict handling with @capture_outputs and @can_return_tuple decorators in the CPMAnt m\u2026",
-    "changed_files": 4,
+    "additions": 41,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "As per title, spiritual successor to #44081 Why? Because as is - Only defaults for fa2/fa3, not on other requested kernels - Limits implementations to one kernel/implementation while I suspect that there will be multiple viable versions (i\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44149",
-    "created_at": "2026-02-19T06:51:06Z",
-    "deletions": 201,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44177",
+    "created_at": "2026-02-20T12:13:30Z",
+    "deletions": 71,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44149/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44149",
+    "files_url": "https://github.com/huggingface/transformers/pull/44177/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44177",
     "labels": [],
-    "merged": false,
-    "number": 44149,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44177,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Refactor CPMAnt output tracing with @capture_outputs and @can_return_tuple",
-    "updated_at": "2026-03-03T00:30:03Z"
+    "title": "[`Flash Attn`] Enable compatible implementations",
+    "updated_at": "2026-02-20T12:43:35Z"
   },
   {
-    "additions": 33,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "Part of the ongoing effort to migrate models to standardized output collection (ref #43979). Replaces manual output_hidden_states/output_attentions/return_dict handling with @capture_outputs and @can_return_tuple decorators in the Bros mod\u2026",
-    "changed_files": 4,
+    "additions": 271,
+    "author": "vasqu",
+    "author_association": "MEMBER",
+    "body_excerpt": "Our kernel loading is incompatible with the original packages as they do not expose the same import structure: - Kernels seem to expose things in the init (and not in the original path) - Original packages seem to expose only within their\u2026",
+    "changed_files": 14,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44148",
-    "created_at": "2026-02-19T06:46:24Z",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44176",
+    "created_at": "2026-02-20T11:36:01Z",
     "deletions": 124,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44148/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44148",
+    "files_url": "https://github.com/huggingface/transformers/pull/44176/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44176",
     "labels": [],
-    "merged": false,
-    "number": 44148,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Refactor Bros output tracing with @capture_outputs and @can_return_tuple",
-    "updated_at": "2026-03-03T00:30:05Z"
+    "merged": true,
+    "number": 44176,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "[`Mamba`] Fix kernel loading",
+    "updated_at": "2026-02-20T16:19:06Z"
   },
   {
-    "additions": 11,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "Part of the ongoing effort to migrate models to standardized output collection (ref #43979). Replaces manual output_hidden_states/output_attentions/return_dict handling with @capture_outputs and @can_return_tuple decorators in the CTRL mod\u2026",
+    "additions": 1,
+    "author": "itazap",
+    "author_association": "MEMBER",
+    "body_excerpt": "add jamba tokenizer mapping to PreTrainedTokenizerFast (v4/v5 BC) for vllm: https://buildkite.com/vllm/ci/builds/52260/steps/canvas?sid=019c76ad-c8f2-4e59-a2f4-5f3b5bbc204c&tab=output",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44147",
-    "created_at": "2026-02-19T06:45:32Z",
-    "deletions": 47,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44175",
+    "created_at": "2026-02-20T11:00:18Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44147/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44147",
+    "files_url": "https://github.com/huggingface/transformers/pull/44175/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44175",
     "labels": [],
     "merged": false,
-    "number": 44147,
+    "number": 44175,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor CTRL output tracing with @capture_outputs and @can_return_tuple",
-    "updated_at": "2026-03-03T00:30:06Z"
+    "title": "add jamba tokenizer mapping to PreTrainedTokenizerFast (v4/v5 BC)",
+    "updated_at": "2026-02-20T16:19:31Z"
   },
   {
-    "additions": 38,
-    "author": "khushali9",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? When using a step-based evaluation strategy (IntervalStrategy.STEPS), the trainer may skip evaluation at the final step if the last step does not align with eval_steps. This avoids missing the final evaluation while\u2026",
+    "additions": 1367,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "This draft expands `utils/check_modeling_structure.py` into a rule-driven linter for model code, with new checks and tests, while keeping runtime very low. Key features: - The checker is intentionally AST-only (no heavy imports/execution),\u2026",
     "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 18,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44146",
-    "created_at": "2026-02-19T05:29:21Z",
-    "deletions": 11,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44146/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44146",
-    "labels": [],
-    "merged": true,
-    "number": 44146,
-    "review_comments_count": 14,
-    "state": "closed",
-    "title": "Ensure final evaluation runs with step-based evaluation strategy",
-    "updated_at": "2026-03-26T16:30:40Z"
-  },
-  {
-    "additions": 400,
-    "author": "balvisio",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds support for sequence packing in the ESM2 model. Currently, the RotaryEmbedding class of the ESM2 model supports BSHD format. This PR makes the RotayEmbedding class aware of the`position_ids` and builds\u2026",
-    "changed_files": 8,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 27,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44145",
-    "created_at": "2026-02-19T02:58:50Z",
-    "deletions": 216,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44174",
+    "created_at": "2026-02-20T10:38:11Z",
+    "deletions": 24,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44145/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44145",
+    "files_url": "https://github.com/huggingface/transformers/pull/44174/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44174",
     "labels": [],
     "merged": true,
-    "number": 44145,
-    "review_comments_count": 9,
+    "number": 44174,
+    "review_comments_count": 38,
     "state": "closed",
-    "title": "Add THD support in ESM",
-    "updated_at": "2026-04-09T14:40:26Z"
+    "title": "Expand model-structure lint rules with a fast AST-based, ruff-like framework",
+    "updated_at": "2026-03-12T06:42:21Z"
   },
   {
-    "additions": 1481,
-    "author": "TinderZ",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR adds 5 Chinese translations for common NLP task tutorials that were missing from the `docs/source/zh/tasks/` directory. The following files are added: - `tasks/sequence_classification.md` - \u6587\u672c\u5206\u7c7b - `tasks/tok\u2026",
-    "changed_files": 6,
+    "additions": 20,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "Fixes flaky GLM OCR generation behavior when 2D `position_ids` are passed explicitly. Reproducible locally with: ``` pytest tests/models/glm_ocr/test_modeling_glm_ocr.py::GlmOcrModelTest::test_generate_with_and_without_position_ids --flake\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44144",
-    "created_at": "2026-02-19T02:35:08Z",
+    "comments_count": 14,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44173",
+    "created_at": "2026-02-20T09:28:48Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44144/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44144",
+    "files_url": "https://github.com/huggingface/transformers/pull/44173/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44173",
     "labels": [],
     "merged": true,
-    "number": 44144,
-    "review_comments_count": 0,
+    "number": 44173,
+    "review_comments_count": 10,
     "state": "closed",
-    "title": "[docs] Add Chinese translations for common NLP task tutorials",
-    "updated_at": "2026-02-20T16:50:29Z"
+    "title": "fix(flaky): `test_generate_with_and_without_position_ids` in GLM ORC",
+    "updated_at": "2026-02-20T19:06:19Z"
   },
   {
     "additions": 2,
-    "author": "nightcityblade",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Fixes minor typos in the `GenerationConfig` class docstring: - \"overriden\" \u2192 \"overridden\" - \"field that are\" \u2192 \"fields that are\" - \"Arg:\" \u2192 \"Args:\" (consistent with the rest of the docstring) No code changes, docum\u2026",
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "## Summary I've noticed `test_synthidtext_watermark_processor_distributional_convergence_*` was our slowest tests in CircleCI This PR speeds up the slowest SynthID distributional convergence tests by optimizing a hot path in SynthIDTextWat\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44143",
-    "created_at": "2026-02-18T23:07:23Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44172",
+    "created_at": "2026-02-20T08:59:30Z",
     "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44143/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44143",
+    "files_url": "https://github.com/huggingface/transformers/pull/44172/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44172",
     "labels": [],
     "merged": true,
-    "number": 44143,
+    "number": 44172,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[docs] Fix typos in GenerationConfig docstring",
-    "updated_at": "2026-02-19T13:24:09Z"
+    "title": "perf: Optimize SynthID logits processor batch index construction",
+    "updated_at": "2026-02-27T09:32:43Z"
   },
   {
-    "additions": 72,
-    "author": "eustlb",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? so @Deep-unlearning noticed, benchmarking for the Open ASR leaderbaord, that the current implem is particularly slow. That would make sense since we go through every layer of the encoder forward, and that the stream\u2026",
-    "changed_files": 3,
+    "additions": 2418,
+    "author": "lmaksym",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds TDT decoder support for Parakeet ASR models, extending the existing CTC-only implementation. It incorporates the initial TDT integration work from [#41545](https://github.com/huggingface/transformers/pu\u2026",
+    "changed_files": 34,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 6,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44142",
-    "created_at": "2026-02-18T21:44:11Z",
-    "deletions": 12,
+    "comments_count": 16,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44171",
+    "created_at": "2026-02-20T08:44:46Z",
+    "deletions": 293,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44142/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44142",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44171/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44171",
+    "labels": [
+      "New model",
+      "Audio"
+    ],
     "merged": false,
-    "number": 44142,
-    "review_comments_count": 0,
+    "number": 44171,
+    "review_comments_count": 81,
     "state": "open",
-    "title": "[voxtral-realtime] get more perfs!",
-    "updated_at": "2026-02-23T17:25:45Z"
-  },
-  {
-    "additions": 42,
-    "author": "dario-fumarola",
-    "author_association": "NONE",
-    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44102 (original account: @fumadari). ## Summary Refactors the `ibert` model to use the new `@capture_outputs` and `@can_return_tuple` decorators for output tracing, as part of\u2026",
-    "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44141",
-    "created_at": "2026-02-18T21:14:53Z",
-    "deletions": 154,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44141/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44141",
-    "labels": [],
-    "merged": false,
-    "number": 44141,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "Refactor ibert output tracing with capture_outputs",
-    "updated_at": "2026-02-22T02:28:47Z"
+    "title": "Parakeet tdt",
+    "updated_at": "2026-04-16T16:36:59Z"
   },
   {
-    "additions": 66,
-    "author": "dario-fumarola",
+    "additions": 74,
+    "author": "veeceey",
     "author_association": "NONE",
-    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44104 (original account: @fumadari). ## Summary - Replace manual `hidden_states`/`attentions`/`cross_attentions` collection in `MegatronBertEncoder` with the `@capture_outputs`\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## Summary - Adds `GitForCausalLM` to `MODEL_FOR_VISUAL_QUESTION_ANSWERING_MAPPING_NAMES` so GIT can be used with the `visual-question-answering` pipeline - Filters tokenizer outputs in VQA pipeline `preprocess` to only pass keys accepted\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44140",
-    "created_at": "2026-02-18T21:14:50Z",
-    "deletions": 207,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44170",
+    "created_at": "2026-02-20T08:28:05Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44140/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44140",
+    "files_url": "https://github.com/huggingface/transformers/pull/44170/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44170",
     "labels": [],
     "merged": false,
-    "number": 44140,
+    "number": 44170,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor megatron_bert to use automatic output tracing",
-    "updated_at": "2026-02-22T02:28:48Z"
+    "title": "Add GIT model support in VQA pipeline",
+    "updated_at": "2026-02-20T09:34:31Z"
   },
   {
-    "additions": 39,
-    "author": "dario-fumarola",
-    "author_association": "NONE",
-    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44105 (original account: @fumadari). ## Summary - Replace manual `hidden_states`/`attentions` collection in `LiltEncoder` with the `@capture_outputs` decorator and forward hook\u2026",
-    "changed_files": 1,
+    "additions": 415,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "Initial ty integration. To avoid a gigantic, risky patch, let's start with a baby step where we add the tooling to make repo-check and activate it on a subset of the repo. That gives us a human-readable patch, and allows us to get conforta\u2026",
+    "changed_files": 25,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44139",
-    "created_at": "2026-02-18T21:14:46Z",
-    "deletions": 127,
+    "comments_count": 12,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44167",
+    "created_at": "2026-02-20T07:39:44Z",
+    "deletions": 210,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44139/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44139",
+    "files_url": "https://github.com/huggingface/transformers/pull/44167/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44167",
     "labels": [],
-    "merged": false,
-    "number": 44139,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44167,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Refactor lilt to use automatic output tracing",
-    "updated_at": "2026-02-22T02:28:48Z"
+    "title": "chore(typing): initial ty integration",
+    "updated_at": "2026-02-20T19:08:09Z"
   },
   {
-    "additions": 51,
-    "author": "dario-fumarola",
-    "author_association": "NONE",
-    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44106 (original account: @fumadari). ## Summary - Replace manual `hidden_states`/`attentions` collection in `YosoEncoder` with the `@capture_outputs` decorator and forward hook\u2026",
-    "changed_files": 2,
+    "additions": 73,
+    "author": "tarekziade",
+    "author_association": "MEMBER",
+    "body_excerpt": "The CI does not output useful info on this flaky test - `tests.models.olmo.test_modeling_olmo.OlmoModelTest testMethod=test_generate_with_static_cache` and makes it harder to determine the root problem when not reproducible locally. This p\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44138",
-    "created_at": "2026-02-18T21:14:42Z",
-    "deletions": 132,
+    "cluster_role": null,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44166",
+    "created_at": "2026-02-20T07:20:15Z",
+    "deletions": 61,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44138/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44138",
+    "files_url": "https://github.com/huggingface/transformers/pull/44166/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44166",
     "labels": [],
-    "merged": false,
-    "number": 44138,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44166,
+    "review_comments_count": 4,
     "state": "closed",
-    "title": "Refactor yoso to use automatic output tracing",
-    "updated_at": "2026-02-22T02:28:49Z"
+    "title": "Improve `has_similar_generate_outputs` assertions",
+    "updated_at": "2026-02-27T08:26:13Z"
   },
   {
-    "additions": 43,
-    "author": "dario-fumarola",
-    "author_association": "NONE",
-    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44107 (original account: @fumadari). ## Summary - Replaces manual `output_hidden_states`/`return_dict` resolution in `MraModel` with `@merge_with_config_defaults` and `@capture\u2026",
+    "additions": 29,
+    "author": "alexmalyshev",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "With Python 3.14 and PEP 649, you can no longer expect `cls.__dict__[\"__annotations__\"]` to contain annotations for the exact class, it will be loaded lazily and can make it seem like the class doesn't have any annotations. The recommended\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44137",
-    "created_at": "2026-02-18T21:14:39Z",
-    "deletions": 113,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44165",
+    "created_at": "2026-02-20T04:47:07Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44137/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44137",
+    "files_url": "https://github.com/huggingface/transformers/pull/44165/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44165",
     "labels": [],
     "merged": false,
-    "number": 44137,
+    "number": 44165,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "refactor(mra): use output tracing decorators",
-    "updated_at": "2026-02-22T02:28:50Z"
+    "title": "Fix how PreTrainedModel checks annotations on Python 3.14+",
+    "updated_at": "2026-03-20T17:31:11Z"
   },
   {
-    "additions": 37,
-    "author": "dario-fumarola",
-    "author_association": "NONE",
-    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44108 (original account: @fumadari). ## Summary - Adds `@merge_with_config_defaults` and `@capture_outputs` to both `VitDetModel` and `VitDetBackbone`, removing manual `output_\u2026",
+    "additions": 7,
+    "author": "lhallee",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fixes: https://github.com/huggingface/transformers/issues/44162 @ArthurZucker @Cyrilvallez",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44136",
-    "created_at": "2026-02-18T21:14:35Z",
-    "deletions": 86,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44163",
+    "created_at": "2026-02-19T21:44:25Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44136/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44136",
+    "files_url": "https://github.com/huggingface/transformers/pull/44163/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44163",
     "labels": [],
-    "merged": false,
-    "number": 44136,
+    "merged": true,
+    "number": 44163,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "refactor(vitdet): use output tracing decorators",
-    "updated_at": "2026-02-22T02:28:50Z"
+    "title": "ESM2 attention_mask and token_dropout fix",
+    "updated_at": "2026-02-20T15:17:31Z"
   },
   {
-    "additions": 48,
-    "author": "dario-fumarola",
-    "author_association": "NONE",
-    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44109 (original account: @fumadari). ## Summary - Part of #43979 \u2014 refactors `hgnet_v2` to use the `capture_outputs` and `merge_with_config_defaults` decorators - Simplifies `H\u2026",
-    "changed_files": 2,
+    "additions": 379,
+    "author": "cogniera",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "What does this PR do? This PR refactors the LongT5 model to use the @capture_outputs and @can_return_tuple decorators for standardized output handling across the model stack. The refactor removes manual handling of: output_attentions outpu\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44135",
-    "created_at": "2026-02-18T21:14:31Z",
-    "deletions": 87,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44161",
+    "created_at": "2026-02-19T20:46:49Z",
+    "deletions": 170,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44135/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44135",
+    "files_url": "https://github.com/huggingface/transformers/pull/44161/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44161",
     "labels": [],
     "merged": false,
-    "number": 44135,
+    "number": 44161,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "refactor(hgnet_v2): use capture_outputs for output tracing",
-    "updated_at": "2026-02-22T02:28:51Z"
+    "state": "open",
+    "title": "Refactor LongT5 to use @capture_outputs and @can_return_tuple decorators for unified output handling (Fixes #43979)",
+    "updated_at": "2026-02-20T17:28:12Z"
   },
   {
-    "additions": 28,
-    "author": "dario-fumarola",
-    "author_association": "NONE",
-    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44110 (original account: @fumadari). ## Summary - Part of #43979 \u2014 refactors `tvp` to use the `capture_outputs`, `can_return_tuple`, and `merge_with_config_defaults` decorators\u2026",
-    "changed_files": 1,
+    "additions": 2104,
+    "author": "molbap",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Hey yall, I started porting the pi0 model so Transformers can be a backend for VLAs directly. I checked it against LeRobot on fix/lerobot_openpi: outputs seem to match and for sure lerobot/pi0_base loads cleanly (no\u2026",
+    "changed_files": 22,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44134",
-    "created_at": "2026-02-18T21:14:27Z",
-    "deletions": 101,
+    "comments_count": 13,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44160",
+    "created_at": "2026-02-19T17:16:29Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44134/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44134",
-    "labels": [],
-    "merged": false,
-    "number": 44134,
-    "review_comments_count": 0,
+    "files_url": "https://github.com/huggingface/transformers/pull/44160/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44160",
+    "labels": [
+      "New model"
+    ],
+    "merged": true,
+    "number": 44160,
+    "review_comments_count": 32,
     "state": "closed",
-    "title": "refactor(tvp): use capture_outputs for output tracing",
-    "updated_at": "2026-02-22T02:28:51Z"
+    "title": "Add model lerobot PI0 to transformers",
+    "updated_at": "2026-03-16T10:23:14Z"
   },
   {
-    "additions": 30,
-    "author": "dario-fumarola",
-    "author_association": "NONE",
-    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44111 (original account: @fumadari). ## Summary - Part of #43979 \u2014 refactors `poolformer` to use the `capture_outputs`, `can_return_tuple`, and `merge_with_config_defaults` dec\u2026",
-    "changed_files": 1,
+    "additions": 67,
+    "author": "samuelleecong",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary Closes #28103 - Refactor `OwlViTAttention` to use `ALL_ATTENTION_FUNCTIONS` for dynamic attention backend dispatch (same pattern as CLIP) - Add `eager_attention_forward` standalone function with the standardized interface - Resh\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44133",
-    "created_at": "2026-02-18T21:12:22Z",
-    "deletions": 59,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44159",
+    "created_at": "2026-02-19T16:31:44Z",
+    "deletions": 61,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44133/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44133",
+    "files_url": "https://github.com/huggingface/transformers/pull/44159/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44159",
     "labels": [],
     "merged": false,
-    "number": 44133,
+    "number": 44159,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "refactor(poolformer): use capture_outputs for output tracing",
-    "updated_at": "2026-02-22T02:28:52Z"
+    "state": "open",
+    "title": "Add SDPA and Flash Attention support for OWL-ViT",
+    "updated_at": "2026-02-24T12:53:10Z"
   },
   {
-    "additions": 13,
-    "author": "eustlb",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? #43674 broke voxtral processor",
-    "changed_files": 1,
+    "additions": 141,
+    "author": "leopold-tzafon",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# Fix issue where `use_cache=False`, corrupts model Qwen3vl output. Tested with: ``` import torch from transformers import Qwen3VLForConditionalGeneration, AutoProcessor MODEL_NAME = \"Qwen/Qwen3-VL-4B-Instruct\" DEVICE = \"cuda\" DTYPE = torc\u2026",
+    "changed_files": 9,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44132",
-    "created_at": "2026-02-18T20:13:15Z",
-    "deletions": 34,
+    "comments_count": 19,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44158",
+    "created_at": "2026-02-19T15:45:13Z",
+    "deletions": 36,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44132/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44132",
+    "files_url": "https://github.com/huggingface/transformers/pull/44158/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44158",
     "labels": [],
     "merged": true,
-    "number": 44132,
-    "review_comments_count": 6,
+    "number": 44158,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[voxtral] fix voxtral proc",
-    "updated_at": "2026-02-19T16:41:53Z"
+    "title": "fix bug with position_ids on qwen3-vl models, such that position_ids include text position",
+    "updated_at": "2026-02-23T14:53:33Z"
   },
   {
-    "additions": 2,
-    "author": "cluster2600",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## What Two small corrections in `docs/source/en/quantization/overview.md`: 1. **Typo fix**: `AuoQuant Notebook` \u2192 `AutoQuant Notebook` in the *User-Friendly Quantization Tools* section. The letter `t` was missing from the link text. 2. **\u2026",
-    "changed_files": 1,
+    "additions": 689,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per title, gets rid of `if/else` per attn implementation",
+    "changed_files": 24,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44131",
-    "created_at": "2026-02-18T19:25:52Z",
-    "deletions": 2,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44157",
+    "created_at": "2026-02-19T14:49:49Z",
+    "deletions": 834,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44131/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44131",
+    "files_url": "https://github.com/huggingface/transformers/pull/44157/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44157",
     "labels": [],
-    "merged": true,
-    "number": 44131,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "docs: fix typo 'AuoQuant' \u2192 'AutoQuant' and clarify FINEGRAINED_FP8 library column",
-    "updated_at": "2026-02-18T20:49:47Z"
+    "merged": false,
+    "number": 44157,
+    "review_comments_count": 10,
+    "state": "open",
+    "title": "Use correct mask for packed inputs in Qwen-VL ",
+    "updated_at": "2026-02-24T13:13:43Z"
   },
   {
-    "additions": 302,
-    "author": "Cyrilvallez",
+    "additions": 2,
+    "author": "SunMarc",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. This PR is the first big step towards removing the `cache_position` everywhere, as they are not needed in general and everything can be inferred from the cache itself. The major changes are the fol\u2026",
-    "changed_files": 23,
+    "body_excerpt": "# What does this PR do? This PR fixes a minor error when using aqml quantization. We specified the wrong argument.",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 7,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44130",
-    "created_at": "2026-02-18T11:58:54Z",
-    "deletions": 886,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44156",
+    "created_at": "2026-02-19T14:35:38Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44130/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44130",
+    "files_url": "https://github.com/huggingface/transformers/pull/44156/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44156",
     "labels": [],
-    "merged": true,
-    "number": 44130,
-    "review_comments_count": 16,
+    "merged": false,
+    "number": 44156,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[generate] Completely stop relying on `cache_position` to prepare inputs",
-    "updated_at": "2026-02-20T18:46:19Z"
+    "title": "Fix aqml `modules_to_not_convert`",
+    "updated_at": "2026-03-27T16:50:02Z"
   },
   {
-    "additions": 76,
-    "author": "preetam1407",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Refactors SpeechT5 to the standardized output tracing interface. - Adds `@capture_outputs` to base encoder/decoder forwards. - Adds `_can_record_outputs` mappings for hidden states and attentions. - Adds `@can_retur\u2026",
+    "additions": 44,
+    "author": "Aatman09",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 1,
     "cluster_id": "cluster-43979-11",
     "cluster_ids": [
@@ -24761,2832 +24575,2876 @@
     ],
     "cluster_role": "member",
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44129",
-    "created_at": "2026-02-18T11:24:13Z",
-    "deletions": 222,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44154",
+    "created_at": "2026-02-19T12:17:56Z",
+    "deletions": 52,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44129/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44129",
+    "files_url": "https://github.com/huggingface/transformers/pull/44154/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44154",
     "labels": [],
     "merged": false,
-    "number": 44129,
+    "number": 44154,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Refactor SpeechT5 output tracing to standardized output capture",
-    "updated_at": "2026-02-18T11:25:19Z"
-  },
-  {
-    "additions": 59,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Just makes sure we trigger dev version update",
-    "changed_files": 2,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44128",
-    "created_at": "2026-02-18T10:42:21Z",
-    "deletions": 5,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44128/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44128",
-    "labels": [],
-    "merged": false,
-    "number": 44128,
-    "review_comments_count": 4,
-    "state": "closed",
-    "title": "update release workflow",
-    "updated_at": "2026-03-30T13:40:19Z"
+    "title": "Refactored vits to match standardized output collection interface",
+    "updated_at": "2026-02-19T12:18:56Z"
   },
   {
-    "additions": 3,
-    "author": "itazap",
+    "additions": 79,
+    "author": "IlyasMoutawwakil",
     "author_association": "MEMBER",
-    "body_excerpt": "when the model_type isn't in `TOKENIZER_MAPPING_NAMES` (ex. \"llama\"), `TOKENIZER_MAPPING_NAMES.get(\"llama\", \"\")` --> \"\". Then we compare `\"\" != \"LlamaTokenizer\"` (the `tokenizer_class` in `tokenizer_config.json`). Since that's true we earl\u2026",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44127",
-    "created_at": "2026-02-18T10:41:48Z",
-    "deletions": 8,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44152",
+    "created_at": "2026-02-19T09:37:51Z",
+    "deletions": 45,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44127/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44127",
+    "files_url": "https://github.com/huggingface/transformers/pull/44152/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44152",
     "labels": [],
     "merged": true,
-    "number": 44127,
-    "review_comments_count": 0,
+    "number": 44152,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "AutoTokenizer ignores config when model_type is None",
-    "updated_at": "2026-02-18T14:47:52Z"
+    "title": "AutoGrad support for grouped_mm fallback",
+    "updated_at": "2026-02-20T11:15:23Z"
   },
   {
-    "additions": 17,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. Let's simplify after https://github.com/huggingface/transformers/pull/42848",
+    "additions": 58,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "Part of the ongoing effort to migrate models to standardized output collection (ref #43979). Replaces manual output_hidden_states/output_attentions/return_dict handling with @capture_outputs and @can_return_tuple decorators in the BioGPT m\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44126",
-    "created_at": "2026-02-18T09:58:49Z",
-    "deletions": 40,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44151",
+    "created_at": "2026-02-19T06:55:43Z",
+    "deletions": 134,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44126/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44126",
+    "files_url": "https://github.com/huggingface/transformers/pull/44151/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44151",
     "labels": [],
-    "merged": true,
-    "number": 44126,
+    "merged": false,
+    "number": 44151,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Simplify input preparation in generate",
-    "updated_at": "2026-02-18T10:30:48Z"
+    "title": "Refactor BioGPT output tracing with @capture_outputs and @can_return_tuple",
+    "updated_at": "2026-03-03T00:30:01Z"
   },
   {
-    "additions": 8,
-    "author": "zucchini-nlp",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/43986",
+    "additions": 22,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "Part of the ongoing effort to migrate models to standardized output collection (ref #43979). Replaces manual output_hidden_states/output_attentions/return_dict handling with @capture_outputs and @can_return_tuple decorators in the MPT mode\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44125",
-    "created_at": "2026-02-18T09:34:54Z",
-    "deletions": 7,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44150",
+    "created_at": "2026-02-19T06:54:09Z",
+    "deletions": 73,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44125/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44125",
+    "files_url": "https://github.com/huggingface/transformers/pull/44150/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44150",
     "labels": [],
-    "merged": true,
-    "number": 44125,
-    "review_comments_count": 2,
+    "merged": false,
+    "number": 44150,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Raise informative error when loading video processors",
-    "updated_at": "2026-02-20T08:23:35Z"
+    "title": "Refactor MPT output tracing with @capture_outputs and @can_return_tuple",
+    "updated_at": "2026-03-03T00:30:02Z"
   },
   {
-    "additions": 10,
-    "author": "mariam851",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Description: Adds eval_on_end to TrainingArguments to force evaluation at the end of training, even if the last step doesn't align with eval_steps. Changes: training_args.py: Added eval_on_end field. trainer.py: Added logic to call evaluat\u2026",
-    "changed_files": 2,
+    "additions": 85,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "Part of the ongoing effort to migrate models to standardized output collection (ref #43979). Replaces manual output_hidden_states/output_attentions/return_dict handling with @capture_outputs and @can_return_tuple decorators in the CPMAnt m\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44124",
-    "created_at": "2026-02-18T08:52:23Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44149",
+    "created_at": "2026-02-19T06:51:06Z",
+    "deletions": 201,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44124/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44124",
+    "files_url": "https://github.com/huggingface/transformers/pull/44149/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44149",
     "labels": [],
     "merged": false,
-    "number": 44124,
+    "number": 44149,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "feat: add eval_on_end to Trainer for final evaluation",
-    "updated_at": "2026-02-18T14:14:16Z"
+    "title": "Refactor CPMAnt output tracing with @capture_outputs and @can_return_tuple",
+    "updated_at": "2026-03-03T00:30:03Z"
   },
   {
     "additions": 33,
-    "author": "cyyever",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR avoids device sync in training loss accumulation by ```torch.where```. The `is_torch_xla_available` condition is also removed.",
-    "changed_files": 1,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "Part of the ongoing effort to migrate models to standardized output collection (ref #43979). Replaces manual output_hidden_states/output_attentions/return_dict handling with @capture_outputs and @can_return_tuple decorators in the Bros mod\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44123",
-    "created_at": "2026-02-18T08:22:57Z",
-    "deletions": 22,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44148",
+    "created_at": "2026-02-19T06:46:24Z",
+    "deletions": 124,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44123/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44123",
+    "files_url": "https://github.com/huggingface/transformers/pull/44148/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44148",
     "labels": [],
     "merged": false,
-    "number": 44123,
+    "number": 44148,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Avoid device sync in training loss accumulation",
-    "updated_at": "2026-03-30T07:57:16Z"
+    "state": "closed",
+    "title": "Refactor Bros output tracing with @capture_outputs and @can_return_tuple",
+    "updated_at": "2026-03-03T00:30:05Z"
   },
   {
-    "additions": 158,
-    "author": "adityuhkapoor",
+    "additions": 11,
+    "author": "ManasVardhan",
     "author_association": "NONE",
-    "body_excerpt": "# What does this PR do? Adds 4-bit embedding quantization for BitsAndBytes, mirroring TorchAO's existing `include_input_output_embeddings` and `untie_embedding_weights` pattern (PRs #37802, #37905, #37935). Large-vocabulary models (Llama 3\u2026",
-    "changed_files": 4,
+    "body_excerpt": "Part of the ongoing effort to migrate models to standardized output collection (ref #43979). Replaces manual output_hidden_states/output_attentions/return_dict handling with @capture_outputs and @can_return_tuple decorators in the CTRL mod\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44122",
-    "created_at": "2026-02-18T06:35:09Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44147",
+    "created_at": "2026-02-19T06:45:32Z",
+    "deletions": 47,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44122/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44122",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44147/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44147",
+    "labels": [],
     "merged": false,
-    "number": 44122,
+    "number": 44147,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add BnB 4-bit embedding quantization support",
-    "updated_at": "2026-02-18T14:27:25Z"
+    "title": "Refactor CTRL output tracing with @capture_outputs and @can_return_tuple",
+    "updated_at": "2026-03-03T00:30:06Z"
   },
   {
-    "additions": 14,
-    "author": "tirth8205",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #34920 After applying `normalize()`, images can have negative values. Calling `resize()` on such images fails because it internally converts to PIL, which requires values in [0, 1] or [0, 255]. ### Fix When the image has values outsi\u2026",
-    "changed_files": 1,
+    "additions": 38,
+    "author": "khushali9",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? When using a step-based evaluation strategy (IntervalStrategy.STEPS), the trainer may skip evaluation at the final step if the last step does not align with eval_steps. This avoids missing the final evaluation while\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44120",
-    "created_at": "2026-02-17T23:56:48Z",
-    "deletions": 0,
+    "comments_count": 18,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44146",
+    "created_at": "2026-02-19T05:29:21Z",
+    "deletions": 11,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44120/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44120",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44120,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix: allow image_transforms.resize to handle negative values after normalization",
-    "updated_at": "2026-02-18T14:08:54Z"
+    "files_url": "https://github.com/huggingface/transformers/pull/44146/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44146",
+    "labels": [],
+    "merged": true,
+    "number": 44146,
+    "review_comments_count": 14,
+    "state": "closed",
+    "title": "Ensure final evaluation runs with step-based evaluation strategy",
+    "updated_at": "2026-03-26T16:30:40Z"
   },
   {
-    "additions": 1,
-    "author": "tirth8205",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #44117 `TOKENIZER_MAPPING_NAMES.get(config_model_type, \"\")` returns `None` when the key exists with value `None`, causing `AttributeError: 'NoneType' object has no attribute 'replace'` when loading models like `google/siglip2-so400m-\u2026",
-    "changed_files": 1,
+    "additions": 400,
+    "author": "balvisio",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds support for sequence packing in the ESM2 model. Currently, the RotaryEmbedding class of the ESM2 model supports BSHD format. This PR makes the RotayEmbedding class aware of the`position_ids` and builds\u2026",
+    "changed_files": 8,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44119",
-    "created_at": "2026-02-17T23:53:20Z",
-    "deletions": 1,
+    "comments_count": 27,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44145",
+    "created_at": "2026-02-19T02:58:50Z",
+    "deletions": 216,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44119/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44119",
+    "files_url": "https://github.com/huggingface/transformers/pull/44145/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44145",
     "labels": [],
-    "merged": false,
-    "number": 44119,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44145,
+    "review_comments_count": 9,
     "state": "closed",
-    "title": "fix: handle None value from TOKENIZER_MAPPING_NAMES.get() in AutoTokenizer",
-    "updated_at": "2026-02-18T14:04:47Z"
+    "title": "Add THD support in ESM",
+    "updated_at": "2026-04-09T14:40:26Z"
   },
   {
-    "additions": 32,
-    "author": "tirth8205",
-    "author_association": "NONE",
-    "body_excerpt": "## Fix Fixes #44079 When a `ModelOutput` dataclass field is initialized as `None`, it is correctly excluded from the OrderedDict keys. However, **subsequently setting that field to a non-None value** via attribute assignment (e.g. `outputs\u2026",
-    "changed_files": 2,
+    "additions": 1481,
+    "author": "TinderZ",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR adds 5 Chinese translations for common NLP task tutorials that were missing from the `docs/source/zh/tasks/` directory. The following files are added: - `tasks/sequence_classification.md` - \u6587\u672c\u5206\u7c7b - `tasks/tok\u2026",
+    "changed_files": 6,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44118",
-    "created_at": "2026-02-17T23:31:31Z",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44144",
+    "created_at": "2026-02-19T02:35:08Z",
     "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44118/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44118",
-    "labels": [
-      "Code agent slop"
-    ],
-    "merged": false,
-    "number": 44118,
+    "files_url": "https://github.com/huggingface/transformers/pull/44144/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44144",
+    "labels": [],
+    "merged": true,
+    "number": 44144,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "fix: ModelOutput keys not updated when setting previously-None dataclass fields",
-    "updated_at": "2026-02-18T14:18:12Z"
+    "title": "[docs] Add Chinese translations for common NLP task tutorials",
+    "updated_at": "2026-02-20T16:50:29Z"
   },
   {
-    "additions": 27,
-    "author": "dtiourine",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Migrate Flaubert to the @capture_outputs and @can_return_tuple decorator pattern for output handling, as part of #43979. # What does this PR do? - Add `_can_record_outputs = {\"attentions\": MultiHeadAttention}` on `FlaubertPreTrainedModel`\u2026",
+    "additions": 2,
+    "author": "nightcityblade",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Fixes minor typos in the `GenerationConfig` class docstring: - \"overriden\" \u2192 \"overridden\" - \"field that are\" \u2192 \"fields that are\" - \"Arg:\" \u2192 \"Args:\" (consistent with the rest of the docstring) No code changes, docum\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44116",
-    "created_at": "2026-02-17T21:52:13Z",
-    "deletions": 102,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44143",
+    "created_at": "2026-02-18T23:07:23Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44116/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44116",
+    "files_url": "https://github.com/huggingface/transformers/pull/44143/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44143",
     "labels": [],
-    "merged": false,
-    "number": 44116,
+    "merged": true,
+    "number": 44143,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[WIP] [Flaubert] Refactor output tracing to decorator-based interface",
-    "updated_at": "2026-02-17T21:53:23Z"
+    "state": "closed",
+    "title": "[docs] Fix typos in GenerationConfig docstring",
+    "updated_at": "2026-02-19T13:24:09Z"
   },
   {
-    "additions": 2,
-    "author": "Deep-unlearning",
+    "additions": 72,
+    "author": "eustlb",
     "author_association": "MEMBER",
-    "body_excerpt": "## Summary - Fix broken `[chat template](./chat_templating)` links in `docs/source/en/tasks/` - `./chat_templating` resolves within `tasks/` (doesn't exist); corrected to `../chat_templating` - Affected files: `tasks/image_text_to_text.md`\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? so @Deep-unlearning noticed, benchmarking for the Open ASR leaderbaord, that the current implem is particularly slow. That would make sense since we go through every layer of the encoder forward, and that the stream\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44115",
-    "created_at": "2026-02-17T21:32:55Z",
-    "deletions": 2,
+    "comments_count": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44142",
+    "created_at": "2026-02-18T21:44:11Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44115/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44115",
+    "files_url": "https://github.com/huggingface/transformers/pull/44142/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44142",
     "labels": [],
-    "merged": true,
-    "number": 44115,
+    "merged": false,
+    "number": 44142,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[docs] fix broken chat_templating links in tasks docs",
-    "updated_at": "2026-02-23T16:27:57Z"
+    "state": "open",
+    "title": "[voxtral-realtime] get more perfs!",
+    "updated_at": "2026-02-23T17:25:45Z"
   },
   {
-    "additions": 716,
-    "author": "23atharvaS",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary This PR migrates the `wav2vec2` family to the standardized output-capturing interface (`@capture_outputs` + `@can_return_tuple`) and includes follow-up compatibility fixes required to make full CI green. ## What changed ### Core\u2026",
-    "changed_files": 19,
+    "additions": 42,
+    "author": "dario-fumarola",
+    "author_association": "NONE",
+    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44102 (original account: @fumadari). ## Summary Refactors the `ibert` model to use the new `@capture_outputs` and `@can_return_tuple` decorators for output tracing, as part of\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44114",
-    "created_at": "2026-02-17T21:17:35Z",
-    "deletions": 1237,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44141",
+    "created_at": "2026-02-18T21:14:53Z",
+    "deletions": 154,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44114/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44114",
+    "files_url": "https://github.com/huggingface/transformers/pull/44141/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44141",
     "labels": [],
     "merged": false,
-    "number": 44114,
+    "number": 44141,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Migrate wav2vec2, wav2vec2_conformer, and wav2vec2_bert to standardized output collection decorators",
-    "updated_at": "2026-02-18T20:34:53Z"
+    "state": "closed",
+    "title": "Refactor ibert output tracing with capture_outputs",
+    "updated_at": "2026-02-22T02:28:47Z"
   },
   {
-    "additions": 5,
-    "author": "harshaljanjani",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "### What does this PR do? The following issue was identified and fixed in this PR: \u2192 Updates the stale `test_device_override` in `test_processing_granite_speech.py` to verify that the device param controls where speech inputs are placed, r\u2026",
+    "additions": 66,
+    "author": "dario-fumarola",
+    "author_association": "NONE",
+    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44104 (original account: @fumadari). ## Summary - Replace manual `hidden_states`/`attentions`/`cross_attentions` collection in `MegatronBertEncoder` with the `@capture_outputs`\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44113",
-    "created_at": "2026-02-17T20:01:32Z",
-    "deletions": 7,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44140",
+    "created_at": "2026-02-18T21:14:50Z",
+    "deletions": 207,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44113/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44113",
+    "files_url": "https://github.com/huggingface/transformers/pull/44140/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44140",
     "labels": [],
-    "merged": true,
-    "number": 44113,
-    "review_comments_count": 2,
+    "merged": false,
+    "number": 44140,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "fix(testing): Update stale device override test in GraniteSpeech",
-    "updated_at": "2026-02-19T11:24:29Z"
+    "title": "Refactor megatron_bert to use automatic output tracing",
+    "updated_at": "2026-02-22T02:28:48Z"
   },
   {
-    "additions": 30,
-    "author": "fumadari",
+    "additions": 39,
+    "author": "dario-fumarola",
     "author_association": "NONE",
-    "body_excerpt": "## Summary - Part of #43979 \u2014 refactors `poolformer` to use the `capture_outputs`, `can_return_tuple`, and `merge_with_config_defaults` decorators - Simplifies `PoolFormerLayer` to return a single tensor instead of a 1-tuple - Simplifies `\u2026",
+    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44105 (original account: @fumadari). ## Summary - Replace manual `hidden_states`/`attentions` collection in `LiltEncoder` with the `@capture_outputs` decorator and forward hook\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44111",
-    "created_at": "2026-02-17T19:38:02Z",
-    "deletions": 59,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44139",
+    "created_at": "2026-02-18T21:14:46Z",
+    "deletions": 127,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44111/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44111",
+    "files_url": "https://github.com/huggingface/transformers/pull/44139/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44139",
     "labels": [],
     "merged": false,
-    "number": 44111,
+    "number": 44139,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "refactor(poolformer): use capture_outputs for output tracing",
-    "updated_at": "2026-02-18T21:19:22Z"
+    "title": "Refactor lilt to use automatic output tracing",
+    "updated_at": "2026-02-22T02:28:48Z"
   },
   {
-    "additions": 28,
-    "author": "fumadari",
+    "additions": 51,
+    "author": "dario-fumarola",
     "author_association": "NONE",
-    "body_excerpt": "## Summary - Part of #43979 \u2014 refactors `tvp` to use the `capture_outputs`, `can_return_tuple`, and `merge_with_config_defaults` decorators - Simplifies `TvpAttention` to always return `(output, attention_probs)` (hooks decide what to capt\u2026",
-    "changed_files": 1,
+    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44106 (original account: @fumadari). ## Summary - Replace manual `hidden_states`/`attentions` collection in `YosoEncoder` with the `@capture_outputs` decorator and forward hook\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44110",
-    "created_at": "2026-02-17T19:32:55Z",
-    "deletions": 101,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44138",
+    "created_at": "2026-02-18T21:14:42Z",
+    "deletions": 132,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44110/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44110",
+    "files_url": "https://github.com/huggingface/transformers/pull/44138/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44138",
     "labels": [],
     "merged": false,
-    "number": 44110,
+    "number": 44138,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "refactor(tvp): use capture_outputs for output tracing",
-    "updated_at": "2026-02-18T21:19:24Z"
+    "title": "Refactor yoso to use automatic output tracing",
+    "updated_at": "2026-02-22T02:28:49Z"
   },
   {
-    "additions": 48,
-    "author": "fumadari",
+    "additions": 43,
+    "author": "dario-fumarola",
     "author_association": "NONE",
-    "body_excerpt": "## Summary - Part of #43979 \u2014 refactors `hgnet_v2` to use the `capture_outputs` and `merge_with_config_defaults` decorators - Simplifies `HGNetV2Encoder` by removing `return_dict` parameter (always returns `BaseModelOutputWithNoAttention`)\u2026",
-    "changed_files": 2,
+    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44107 (original account: @fumadari). ## Summary - Replaces manual `output_hidden_states`/`return_dict` resolution in `MraModel` with `@merge_with_config_defaults` and `@capture\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44109",
-    "created_at": "2026-02-17T19:23:03Z",
-    "deletions": 87,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44137",
+    "created_at": "2026-02-18T21:14:39Z",
+    "deletions": 113,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44109/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44109",
+    "files_url": "https://github.com/huggingface/transformers/pull/44137/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44137",
     "labels": [],
     "merged": false,
-    "number": 44109,
+    "number": 44137,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "refactor(hgnet_v2): use capture_outputs for output tracing",
-    "updated_at": "2026-02-18T21:19:25Z"
+    "title": "refactor(mra): use output tracing decorators",
+    "updated_at": "2026-02-22T02:28:50Z"
   },
   {
-    "additions": 33,
-    "author": "fumadari",
+    "additions": 37,
+    "author": "dario-fumarola",
     "author_association": "NONE",
-    "body_excerpt": "## Summary - Adds `@merge_with_config_defaults` and `@capture_outputs` to both `VitDetModel` and `VitDetBackbone`, removing manual `output_attentions`/`return_dict` resolution - Adds `_can_record_outputs = {\"attentions\": VitDetAttention}`\u2026",
-    "changed_files": 1,
+    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44108 (original account: @fumadari). ## Summary - Adds `@merge_with_config_defaults` and `@capture_outputs` to both `VitDetModel` and `VitDetBackbone`, removing manual `output_\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44108",
-    "created_at": "2026-02-17T19:15:00Z",
-    "deletions": 82,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44136",
+    "created_at": "2026-02-18T21:14:35Z",
+    "deletions": 86,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44108/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44108",
+    "files_url": "https://github.com/huggingface/transformers/pull/44136/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44136",
     "labels": [],
     "merged": false,
-    "number": 44108,
+    "number": 44136,
     "review_comments_count": 0,
     "state": "closed",
     "title": "refactor(vitdet): use output tracing decorators",
-    "updated_at": "2026-02-18T21:19:27Z"
+    "updated_at": "2026-02-22T02:28:50Z"
   },
   {
-    "additions": 40,
-    "author": "fumadari",
+    "additions": 48,
+    "author": "dario-fumarola",
     "author_association": "NONE",
-    "body_excerpt": "## Summary - Replaces manual `output_hidden_states`/`return_dict` resolution in `MraModel` with `@merge_with_config_defaults` and `@capture_outputs` decorators - Simplifies `MraEncoder` to a plain loop returning a single tensor, removing `\u2026",
-    "changed_files": 1,
+    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44109 (original account: @fumadari). ## Summary - Part of #43979 \u2014 refactors `hgnet_v2` to use the `capture_outputs` and `merge_with_config_defaults` decorators - Simplifies `H\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44107",
-    "created_at": "2026-02-17T19:04:42Z",
-    "deletions": 112,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44135",
+    "created_at": "2026-02-18T21:14:31Z",
+    "deletions": 87,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44107/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44107",
+    "files_url": "https://github.com/huggingface/transformers/pull/44135/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44135",
     "labels": [],
     "merged": false,
-    "number": 44107,
+    "number": 44135,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "refactor(mra): use output tracing decorators",
-    "updated_at": "2026-02-18T21:19:29Z"
+    "title": "refactor(hgnet_v2): use capture_outputs for output tracing",
+    "updated_at": "2026-02-22T02:28:51Z"
   },
   {
-    "additions": 47,
-    "author": "fumadari",
+    "additions": 28,
+    "author": "dario-fumarola",
     "author_association": "NONE",
-    "body_excerpt": "## Summary - Replace manual `hidden_states`/`attentions` collection in `YosoEncoder` with the `@capture_outputs` decorator and forward hooks - Add `@can_return_tuple` to all 5 wrapper model classes, eliminating manual `return_dict` handlin\u2026",
+    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44110 (original account: @fumadari). ## Summary - Part of #43979 \u2014 refactors `tvp` to use the `capture_outputs`, `can_return_tuple`, and `merge_with_config_defaults` decorators\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44106",
-    "created_at": "2026-02-17T18:59:25Z",
-    "deletions": 132,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44134",
+    "created_at": "2026-02-18T21:14:27Z",
+    "deletions": 101,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44106/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44106",
+    "files_url": "https://github.com/huggingface/transformers/pull/44134/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44134",
     "labels": [],
     "merged": false,
-    "number": 44106,
+    "number": 44134,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor yoso to use automatic output tracing",
-    "updated_at": "2026-02-18T21:19:30Z"
+    "title": "refactor(tvp): use capture_outputs for output tracing",
+    "updated_at": "2026-02-22T02:28:51Z"
   },
   {
-    "additions": 39,
-    "author": "fumadari",
+    "additions": 30,
+    "author": "dario-fumarola",
     "author_association": "NONE",
-    "body_excerpt": "## Summary - Replace manual `hidden_states`/`attentions` collection in `LiltEncoder` with the `@capture_outputs` decorator and forward hooks - Add `@can_return_tuple` to all 3 wrapper model classes, eliminating manual `return_dict` handlin\u2026",
+    "body_excerpt": "Migrated from https://github.com/huggingface/transformers/pull/44111 (original account: @fumadari). ## Summary - Part of #43979 \u2014 refactors `poolformer` to use the `capture_outputs`, `can_return_tuple`, and `merge_with_config_defaults` dec\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44105",
-    "created_at": "2026-02-17T18:54:40Z",
-    "deletions": 127,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44133",
+    "created_at": "2026-02-18T21:12:22Z",
+    "deletions": 59,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44105/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44105",
+    "files_url": "https://github.com/huggingface/transformers/pull/44133/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44133",
     "labels": [],
     "merged": false,
-    "number": 44105,
+    "number": 44133,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor lilt to use automatic output tracing",
-    "updated_at": "2026-02-18T21:19:32Z"
+    "title": "refactor(poolformer): use capture_outputs for output tracing",
+    "updated_at": "2026-02-22T02:28:52Z"
   },
   {
-    "additions": 66,
-    "author": "fumadari",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary - Replace manual `hidden_states`/`attentions`/`cross_attentions` collection in `MegatronBertEncoder` with the `@capture_outputs` decorator and forward hooks - Add `@can_return_tuple` to all 8 wrapper model classes, eliminating m\u2026",
+    "additions": 13,
+    "author": "eustlb",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? #43674 broke voxtral processor",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44104",
-    "created_at": "2026-02-17T18:43:44Z",
-    "deletions": 207,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44132",
+    "created_at": "2026-02-18T20:13:15Z",
+    "deletions": 34,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44104/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44104",
+    "files_url": "https://github.com/huggingface/transformers/pull/44132/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44132",
     "labels": [],
-    "merged": false,
-    "number": 44104,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44132,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "Refactor megatron_bert to use automatic output tracing",
-    "updated_at": "2026-02-18T21:19:34Z"
+    "title": "[voxtral] fix voxtral proc",
+    "updated_at": "2026-02-19T16:41:53Z"
   },
   {
-    "additions": 53,
-    "author": "engmohamedsalah",
-    "author_association": "NONE",
-    "body_excerpt": "Fixes #44052 Now and then, the indexer ran into trouble switching between masks and cache. Most of the test failures came from these hiccups: - Indexer cache: the old if seq_len > 1: reset cache heuristic broke assisted decoding (multi-tok\u2026",
-    "changed_files": 3,
+    "additions": 2,
+    "author": "cluster2600",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## What Two small corrections in `docs/source/en/quantization/overview.md`: 1. **Typo fix**: `AuoQuant Notebook` \u2192 `AutoQuant Notebook` in the *User-Friendly Quantization Tools* section. The letter `t` was missing from the link text. 2. **\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44103",
-    "created_at": "2026-02-17T18:04:48Z",
-    "deletions": 76,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44131",
+    "created_at": "2026-02-18T19:25:52Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44103/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44103",
+    "files_url": "https://github.com/huggingface/transformers/pull/44131/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44131",
     "labels": [],
-    "merged": false,
-    "number": 44103,
+    "merged": true,
+    "number": 44131,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix glm_moe_dsa",
-    "updated_at": "2026-02-18T19:38:11Z"
+    "title": "docs: fix typo 'AuoQuant' \u2192 'AutoQuant' and clarify FINEGRAINED_FP8 library column",
+    "updated_at": "2026-02-18T20:49:47Z"
   },
   {
-    "additions": 42,
-    "author": "fumadari",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Refactors the `ibert` model to use the new `@capture_outputs` and `@can_return_tuple` decorators for output tracing, as part of the meta-issue #43979. **Key changes:** - Added `_can_record_outputs = {\"hidden_states\": IBertLayer,\u2026",
-    "changed_files": 1,
+    "additions": 302,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title. This PR is the first big step towards removing the `cache_position` everywhere, as they are not needed in general and everything can be inferred from the cache itself. The major changes are the fol\u2026",
+    "changed_files": 23,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44102",
-    "created_at": "2026-02-17T17:21:32Z",
-    "deletions": 154,
+    "comments_count": 7,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44130",
+    "created_at": "2026-02-18T11:58:54Z",
+    "deletions": 886,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44102/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44102",
+    "files_url": "https://github.com/huggingface/transformers/pull/44130/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44130",
+    "labels": [],
+    "merged": true,
+    "number": 44130,
+    "review_comments_count": 16,
+    "state": "closed",
+    "title": "[generate] Completely stop relying on `cache_position` to prepare inputs",
+    "updated_at": "2026-02-20T18:46:19Z"
+  },
+  {
+    "additions": 76,
+    "author": "preetam1407",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? Refactors SpeechT5 to the standardized output tracing interface. - Adds `@capture_outputs` to base encoder/decoder forwards. - Adds `_can_record_outputs` mappings for hidden states and attentions. - Adds `@can_retur\u2026",
+    "changed_files": 1,
+    "cluster_id": "cluster-43979-11",
+    "cluster_ids": [
+      "cluster-43979-11"
+    ],
+    "cluster_role": "member",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44129",
+    "created_at": "2026-02-18T11:24:13Z",
+    "deletions": 222,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44129/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44129",
     "labels": [],
     "merged": false,
-    "number": 44102,
+    "number": 44129,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Refactor ibert output tracing with capture_outputs",
-    "updated_at": "2026-02-18T21:19:35Z"
+    "state": "open",
+    "title": "Refactor SpeechT5 output tracing to standardized output capture",
+    "updated_at": "2026-02-18T11:25:19Z"
   },
   {
-    "additions": 210,
-    "author": "aman-coder03",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? This PR refactors XLM's output tracing to align with the standardized output capturing patterns used across the codebase. ### Key changes: - Refactors transformer blocks into a dedicated `XLMLayer` module to enable\u2026",
+    "additions": 59,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Just makes sure we trigger dev version update",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44101",
-    "created_at": "2026-02-17T17:15:06Z",
-    "deletions": 194,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44128",
+    "created_at": "2026-02-18T10:42:21Z",
+    "deletions": 5,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44101/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44101",
+    "files_url": "https://github.com/huggingface/transformers/pull/44128/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44128",
     "labels": [],
     "merged": false,
-    "number": 44101,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "[XLM] Refactor output tracing to align with capture_outputs standardized architecture",
-    "updated_at": "2026-02-19T08:08:33Z"
+    "number": 44128,
+    "review_comments_count": 4,
+    "state": "closed",
+    "title": "update release workflow",
+    "updated_at": "2026-03-30T13:40:19Z"
   },
   {
     "additions": 3,
-    "author": "qgallouedec",
+    "author": "itazap",
     "author_association": "MEMBER",
-    "body_excerpt": "In https://github.com/huggingface/trl/pull/5112 a user reported that `trl sft --help` fails It's because three inherited args from `TrainingArguments` (`torch_empty_cache_steps`, `gradient_checkpointing` and `use_liger_kernel`)help strings\u2026",
-    "changed_files": 1,
+    "body_excerpt": "when the model_type isn't in `TOKENIZER_MAPPING_NAMES` (ex. \"llama\"), `TOKENIZER_MAPPING_NAMES.get(\"llama\", \"\")` --> \"\". Then we compare `\"\" != \"LlamaTokenizer\"` (the `tokenizer_class` in `tokenizer_config.json`). Since that's true we earl\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44100",
-    "created_at": "2026-02-17T17:10:36Z",
-    "deletions": 3,
+    "cluster_role": null,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44127",
+    "created_at": "2026-02-18T10:41:48Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44100/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44100",
+    "files_url": "https://github.com/huggingface/transformers/pull/44127/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44127",
     "labels": [],
     "merged": true,
-    "number": 44100,
+    "number": 44127,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix percentage formatting in help messages for gradient checkpointing, Liger Kernel, and empty cache steps",
-    "updated_at": "2026-02-20T09:57:51Z"
+    "title": "AutoTokenizer ignores config when model_type is None",
+    "updated_at": "2026-02-18T14:47:52Z"
   },
   {
-    "additions": 2,
-    "author": "qgallouedec",
+    "additions": 17,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "body_excerpt": "# What does this PR do? As per the title. Let's simplify after https://github.com/huggingface/transformers/pull/42848",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44099",
-    "created_at": "2026-02-17T16:45:35Z",
-    "deletions": 2,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44126",
+    "created_at": "2026-02-18T09:58:49Z",
+    "deletions": 40,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44099/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44099",
+    "files_url": "https://github.com/huggingface/transformers/pull/44126/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44126",
     "labels": [],
     "merged": true,
-    "number": 44099,
+    "number": 44126,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Bump dev version",
-    "updated_at": "2026-02-18T10:03:54Z"
+    "title": "Simplify input preparation in generate",
+    "updated_at": "2026-02-18T10:30:48Z"
   },
   {
-    "additions": 125,
-    "author": "aman-coder03",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? This PR refactors ViLT's output handling to align with the standardized patterns used across the codebase. Key changes: - Removes manual `hidden_states`/`attentions` propagation and passes `output_attentions`, `out\u2026",
-    "changed_files": 2,
+    "additions": 8,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/43986",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44098",
-    "created_at": "2026-02-17T16:32:34Z",
-    "deletions": 138,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44125",
+    "created_at": "2026-02-18T09:34:54Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44098/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44098",
+    "files_url": "https://github.com/huggingface/transformers/pull/44125/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44125",
     "labels": [],
-    "merged": false,
-    "number": 44098,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "[ViLT] Refactor output handling to align with standardized patterns",
-    "updated_at": "2026-02-17T16:37:46Z"
+    "merged": true,
+    "number": 44125,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "Raise informative error when loading video processors",
+    "updated_at": "2026-02-20T08:23:35Z"
   },
   {
-    "additions": 12,
-    "author": "Rocketknight1",
-    "author_association": "MEMBER",
-    "body_excerpt": "The `test_keep_in_fp32_modules` issues in #44052 are because the test assumes a model has **either** `_keep_in_fp32_modules` or `_keep_in_fp32_modules_strict` **but not both.** The only model that uses both is `glm_moe_dsa`, so this is the\u2026",
+    "additions": 10,
+    "author": "mariam851",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Description: Adds eval_on_end to TrainingArguments to force evaluation at the end of training, even if the last step doesn't align with eval_steps. Changes: training_args.py: Added eval_on_end field. trainer.py: Added logic to call evaluat\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44097",
-    "created_at": "2026-02-17T15:43:55Z",
-    "deletions": 42,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44124",
+    "created_at": "2026-02-18T08:52:23Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44097/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44097",
+    "files_url": "https://github.com/huggingface/transformers/pull/44124/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44124",
     "labels": [],
-    "merged": true,
-    "number": 44097,
-    "review_comments_count": 2,
+    "merged": false,
+    "number": 44124,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Merge test_keep_in_fp32_modules and test_keep_in_fp32_modules_strict",
-    "updated_at": "2026-02-17T16:23:33Z"
+    "title": "feat: add eval_on_end to Trainer for final evaluation",
+    "updated_at": "2026-02-18T14:14:16Z"
   },
   {
-    "additions": 3,
-    "author": "eustlb",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Tests were written (and pass) on DGX A100, here are the values for our runners.",
+    "additions": 33,
+    "author": "cyyever",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR avoids device sync in training loss accumulation by ```torch.where```. The `is_torch_xla_available` condition is also removed.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44096",
-    "created_at": "2026-02-17T15:14:26Z",
-    "deletions": 3,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44123",
+    "created_at": "2026-02-18T08:22:57Z",
+    "deletions": 22,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44096/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44096",
+    "files_url": "https://github.com/huggingface/transformers/pull/44123/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44123",
     "labels": [],
-    "merged": true,
-    "number": 44096,
+    "merged": false,
+    "number": 44123,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[voxtral-realtime] update runner expected values ",
-    "updated_at": "2026-02-17T15:23:19Z"
+    "state": "open",
+    "title": "Avoid device sync in training loss accumulation",
+    "updated_at": "2026-03-30T07:57:16Z"
   },
   {
-    "additions": 43,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. The check that was added in https://github.com/huggingface/transformers/pull/43768 is wrong, as a missing weight would NOT be reinitialized in some cases! As for the pointers check, it is actually\u2026",
-    "changed_files": 2,
+    "additions": 158,
+    "author": "adityuhkapoor",
+    "author_association": "NONE",
+    "body_excerpt": "# What does this PR do? Adds 4-bit embedding quantization for BitsAndBytes, mirroring TorchAO's existing `include_input_output_embeddings` and `untie_embedding_weights` pattern (PRs #37802, #37905, #37935). Large-vocabulary models (Llama 3\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44095",
-    "created_at": "2026-02-17T14:33:22Z",
-    "deletions": 47,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44122",
+    "created_at": "2026-02-18T06:35:09Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44095/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44095",
-    "labels": [],
-    "merged": true,
-    "number": 44095,
-    "review_comments_count": 2,
+    "files_url": "https://github.com/huggingface/transformers/pull/44122/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44122",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44122,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix loading logic issue",
-    "updated_at": "2026-04-03T04:52:35Z"
+    "title": "Add BnB 4-bit embedding quantization support",
+    "updated_at": "2026-02-18T14:27:25Z"
   },
   {
-    "additions": 24,
-    "author": "ManasVardhan",
+    "additions": 14,
+    "author": "tirth8205",
     "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the `codegen` model to use the `@capture_outputs` and `@can_return_tuple` decorators, replacing manual `output_attentions`/`output_hidden_states`/`return_dict` handling. ### Changes: - Add `_can_record_ou\u2026",
+    "body_excerpt": "Fixes #34920 After applying `normalize()`, images can have negative values. Calling `resize()` on such images fails because it internally converts to PIL, which requires values in [0, 1] or [0, 255]. ### Fix When the image has values outsi\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44094",
-    "created_at": "2026-02-17T14:15:10Z",
-    "deletions": 70,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44120",
+    "created_at": "2026-02-17T23:56:48Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44094/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44094",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44120/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44120",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44094,
+    "number": 44120,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor codegen model to use @capture_outputs and @can_return_tuple decorators",
-    "updated_at": "2026-03-03T00:30:07Z"
+    "title": "fix: allow image_transforms.resize to handle negative values after normalization",
+    "updated_at": "2026-02-18T14:08:54Z"
   },
   {
-    "additions": 28,
-    "author": "ManasVardhan",
+    "additions": 1,
+    "author": "tirth8205",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Refactors the DeBERTa model to use the standardized output collection interface with `@capture_outputs` and `@can_return_tuple` decorators. ### Changes - Add `_can_record_outputs` on `DebertaPreTrainedModel` mapping `hidden_stat\u2026",
+    "body_excerpt": "Fixes #44117 `TOKENIZER_MAPPING_NAMES.get(config_model_type, \"\")` returns `None` when the key exists with value `None`, causing `AttributeError: 'NoneType' object has no attribute 'replace'` when loading models like `google/siglip2-so400m-\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44093",
-    "created_at": "2026-02-17T14:15:07Z",
-    "deletions": 129,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44119",
+    "created_at": "2026-02-17T23:53:20Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44093/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44093",
+    "files_url": "https://github.com/huggingface/transformers/pull/44119/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44119",
     "labels": [],
     "merged": false,
-    "number": 44093,
+    "number": 44119,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor DeBERTa output tracing with @capture_outputs and @can_return_tuple",
-    "updated_at": "2026-03-03T00:30:09Z"
+    "title": "fix: handle None value from TOKENIZER_MAPPING_NAMES.get() in AutoTokenizer",
+    "updated_at": "2026-02-18T14:04:47Z"
   },
   {
-    "additions": 79,
-    "author": "ManasVardhan",
+    "additions": 32,
+    "author": "tirth8205",
     "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the **swinv2** model to use hook-based output capturing via `@capture_outputs` and `@can_return_tuple` decorators, removing the manual threading of `output_attentions`/`output_hidden_states`/`return_dict`\u2026",
+    "body_excerpt": "## Fix Fixes #44079 When a `ModelOutput` dataclass field is initialized as `None`, it is correctly excluded from the OrderedDict keys. However, **subsequently setting that field to a non-None value** via attribute assignment (e.g. `outputs\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44092",
-    "created_at": "2026-02-17T14:15:04Z",
-    "deletions": 159,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44118",
+    "created_at": "2026-02-17T23:31:31Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44092/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44092",
-    "labels": [],
+    "files_url": "https://github.com/huggingface/transformers/pull/44118/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44118",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44092,
+    "number": 44118,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor output tracing for swinv2 model",
-    "updated_at": "2026-03-03T00:30:10Z"
+    "title": "fix: ModelOutput keys not updated when setting previously-None dataclass fields",
+    "updated_at": "2026-02-18T14:18:12Z"
   },
   {
-    "additions": 79,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the `swin` model to use the standardized output collection interface (`@capture_outputs` and `@can_return_tuple` decorators), as described in #43979. ### Changes **SwinPreTrainedModel:** - Added `_can_rec\u2026",
-    "changed_files": 2,
+    "additions": 27,
+    "author": "dtiourine",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Migrate Flaubert to the @capture_outputs and @can_return_tuple decorator pattern for output handling, as part of #43979. # What does this PR do? - Add `_can_record_outputs = {\"attentions\": MultiHeadAttention}` on `FlaubertPreTrainedModel`\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44091",
-    "created_at": "2026-02-17T14:14:56Z",
-    "deletions": 146,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44116",
+    "created_at": "2026-02-17T21:52:13Z",
+    "deletions": 102,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44091/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44091",
+    "files_url": "https://github.com/huggingface/transformers/pull/44116/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44116",
     "labels": [],
     "merged": false,
-    "number": 44091,
+    "number": 44116,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Refactor Swin output tracing with @capture_outputs and @can_return_tuple",
-    "updated_at": "2026-03-03T00:30:11Z"
+    "state": "open",
+    "title": "[WIP] [Flaubert] Refactor output tracing to decorator-based interface",
+    "updated_at": "2026-02-17T21:53:23Z"
   },
   {
-    "additions": 25,
-    "author": "itazap",
+    "additions": 2,
+    "author": "Deep-unlearning",
     "author_association": "MEMBER",
-    "body_excerpt": "bos and eos behaviour should match when updating post processor setting `add_bos_token=True` when `bos_token=None` should silently disables `add_bos_token`. (was already the behavior for `eos_token`)",
+    "body_excerpt": "## Summary - Fix broken `[chat template](./chat_templating)` links in `docs/source/en/tasks/` - `./chat_templating` resolves within `tasks/` (doesn't exist); corrected to `../chat_templating` - Affected files: `tasks/image_text_to_text.md`\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44090",
-    "created_at": "2026-02-17T13:15:07Z",
-    "deletions": 4,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44115",
+    "created_at": "2026-02-17T21:32:55Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44090/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44090",
+    "files_url": "https://github.com/huggingface/transformers/pull/44115/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44115",
     "labels": [],
     "merged": true,
-    "number": 44090,
+    "number": 44115,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Update post proc",
-    "updated_at": "2026-02-18T15:34:18Z"
+    "title": "[docs] fix broken chat_templating links in tasks docs",
+    "updated_at": "2026-02-23T16:27:57Z"
   },
   {
-    "additions": 113,
-    "author": "preetam1407",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes #43979 ## Summary Refactor T5 to the standardized output tracing interface. ## Changes - Added `_can_record_outputs` on T5 encoder/decoder stack subclasses. - Added `@capture_outputs` on the base stack forward. - Added `@can_return_t\u2026",
-    "changed_files": 2,
+    "additions": 716,
+    "author": "23atharvaS",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary This PR migrates the `wav2vec2` family to the standardized output-capturing interface (`@capture_outputs` + `@can_return_tuple`) and includes follow-up compatibility fixes required to make full CI green. ## What changed ### Core\u2026",
+    "changed_files": 19,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44089",
-    "created_at": "2026-02-17T11:37:18Z",
-    "deletions": 294,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44114",
+    "created_at": "2026-02-17T21:17:35Z",
+    "deletions": 1237,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44089/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44089",
+    "files_url": "https://github.com/huggingface/transformers/pull/44114/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44114",
     "labels": [],
     "merged": false,
-    "number": 44089,
+    "number": 44114,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Refactor t5 output tracing",
-    "updated_at": "2026-02-17T13:45:23Z"
+    "state": "open",
+    "title": "Migrate wav2vec2, wav2vec2_conformer, and wav2vec2_bert to standardized output collection decorators",
+    "updated_at": "2026-02-18T20:34:53Z"
   },
   {
-    "additions": 41,
-    "author": "Zephyr-Blessed",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Description Refactors GPT-2 model to use the standardized `@capture_outputs` and `@can_return_tuple` decorators, replacing manual output collection boilerplate. Part of #43979 ## Changes - **`GPT2PreTrainedModel`**: Added `_can_record_o\u2026",
+    "additions": 5,
+    "author": "harshaljanjani",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "### What does this PR do? The following issue was identified and fixed in this PR: \u2192 Updates the stale `test_device_override` in `test_processing_granite_speech.py` to verify that the device param controls where speech inputs are placed, r\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44088",
-    "created_at": "2026-02-17T11:32:42Z",
-    "deletions": 129,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44113",
+    "created_at": "2026-02-17T20:01:32Z",
+    "deletions": 7,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44088/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44088",
+    "files_url": "https://github.com/huggingface/transformers/pull/44113/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44113",
     "labels": [],
-    "merged": false,
-    "number": 44088,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44113,
+    "review_comments_count": 2,
     "state": "closed",
-    "title": "Refactor GPT-2 output tracing with capture_outputs/can_return_tuple",
-    "updated_at": "2026-02-17T11:41:32Z"
+    "title": "fix(testing): Update stale device override test in GraniteSpeech",
+    "updated_at": "2026-02-19T11:24:29Z"
   },
   {
-    "additions": 16,
-    "author": "huyxdang",
+    "additions": 30,
+    "author": "fumadari",
     "author_association": "NONE",
-    "body_excerpt": "### Summary Refactors the Mamba2 model to use the standardized output collection interface as part of #43979. ### Changes * **Standardized Output Mapping**: Added `_can_record_outputs` to `Mamba2PreTrainedModel` mapping `hidden_states` \u2192 `\u2026",
-    "changed_files": 2,
+    "body_excerpt": "## Summary - Part of #43979 \u2014 refactors `poolformer` to use the `capture_outputs`, `can_return_tuple`, and `merge_with_config_defaults` decorators - Simplifies `PoolFormerLayer` to return a single tensor instead of a 1-tuple - Simplifies `\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44087",
-    "created_at": "2026-02-17T11:30:25Z",
-    "deletions": 33,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44111",
+    "created_at": "2026-02-17T19:38:02Z",
+    "deletions": 59,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44087/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44087",
+    "files_url": "https://github.com/huggingface/transformers/pull/44111/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44111",
     "labels": [],
     "merged": false,
-    "number": 44087,
+    "number": 44111,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor Mamba2 to use standardized output tracing",
-    "updated_at": "2026-03-11T02:08:22Z"
+    "title": "refactor(poolformer): use capture_outputs for output tracing",
+    "updated_at": "2026-02-18T21:19:22Z"
   },
   {
-    "additions": 16,
-    "author": "Zephyr-Blessed",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Migrates **MGP-STR** to the standardized output collection interface using `@capture_outputs` and `@can_return_tuple` decorators. ### Changes: - Added `_can_record_outputs = {\"hidden_states\": MgpstrLayer, \"attentio\u2026",
+    "additions": 28,
+    "author": "fumadari",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Part of #43979 \u2014 refactors `tvp` to use the `capture_outputs`, `can_return_tuple`, and `merge_with_config_defaults` decorators - Simplifies `TvpAttention` to always return `(output, attention_probs)` (hooks decide what to capt\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44086",
-    "created_at": "2026-02-17T11:21:22Z",
-    "deletions": 48,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44086/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44086",
-    "labels": [],
-    "merged": false,
-    "number": 44086,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "[MGP-STR] Refactor output tracing to use capture_outputs/can_return_tuple decorators",
-    "updated_at": "2026-02-17T11:22:25Z"
-  },
-  {
-    "additions": 37,
-    "author": "Zephyr-Blessed",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Refactors the RemBERT model to use the new output tracing decorators (`@capture_outputs` and `@can_return_tuple`), replacing manual output collection boilerplate. ### Changes: - Added `@capture_outputs` decorator t\u2026",
-    "changed_files": 1,
-    "cluster_id": "cluster-43979-11",
-    "cluster_ids": [
-      "cluster-43979-11"
-    ],
-    "cluster_role": "member",
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44085",
-    "created_at": "2026-02-17T11:09:55Z",
-    "deletions": 108,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44110",
+    "created_at": "2026-02-17T19:32:55Z",
+    "deletions": 101,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44085/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44085",
+    "files_url": "https://github.com/huggingface/transformers/pull/44110/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44110",
     "labels": [],
     "merged": false,
-    "number": 44085,
+    "number": 44110,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Refactor RemBERT to use output tracing decorators",
-    "updated_at": "2026-02-17T11:10:59Z"
+    "state": "closed",
+    "title": "refactor(tvp): use capture_outputs for output tracing",
+    "updated_at": "2026-02-18T21:19:24Z"
   },
   {
-    "additions": 37,
-    "author": "Zephyr-Blessed",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Refactors the GPT-J model to use the new `capture_outputs` and `can_return_tuple` decorators for output tracing, following the same pattern as #44046 (CodeGen). ### Changes: - Added `@capture_outputs` decorator on\u2026",
-    "changed_files": 1,
+    "additions": 48,
+    "author": "fumadari",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Part of #43979 \u2014 refactors `hgnet_v2` to use the `capture_outputs` and `merge_with_config_defaults` decorators - Simplifies `HGNetV2Encoder` by removing `return_dict` parameter (always returns `BaseModelOutputWithNoAttention`)\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44084",
-    "created_at": "2026-02-17T11:08:48Z",
-    "deletions": 108,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44109",
+    "created_at": "2026-02-17T19:23:03Z",
+    "deletions": 87,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44084/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44084",
+    "files_url": "https://github.com/huggingface/transformers/pull/44109/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44109",
     "labels": [],
     "merged": false,
-    "number": 44084,
+    "number": 44109,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[GPT-J] Refactor output tracing to use capture_outputs/can_return_tuple decorators",
-    "updated_at": "2026-02-17T11:41:38Z"
+    "title": "refactor(hgnet_v2): use capture_outputs for output tracing",
+    "updated_at": "2026-02-18T21:19:25Z"
   },
   {
-    "additions": 1555,
-    "author": "3outeille",
-    "author_association": "MEMBER",
-    "body_excerpt": "- TODO: - fsdp => faire comme tp en mode fsdp_plan manual qui devient l'auto par d\u00e9faut --- This PR introduces **first-class FSDP2 (Fully Sharded Data Parallel v2) support** directly in Transformers, bypassing the need for Accelerate's FSD\u2026",
-    "changed_files": 11,
+    "additions": 33,
+    "author": "fumadari",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Adds `@merge_with_config_defaults` and `@capture_outputs` to both `VitDetModel` and `VitDetBackbone`, removing manual `output_attentions`/`return_dict` resolution - Adds `_can_record_outputs = {\"attentions\": VitDetAttention}`\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44083",
-    "created_at": "2026-02-17T10:57:06Z",
-    "deletions": 120,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44108",
+    "created_at": "2026-02-17T19:15:00Z",
+    "deletions": 82,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44083/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44083",
+    "files_url": "https://github.com/huggingface/transformers/pull/44108/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44108",
     "labels": [],
     "merged": false,
-    "number": 44083,
-    "review_comments_count": 24,
-    "state": "open",
-    "title": "FSDP2 native support in transformers ",
-    "updated_at": "2026-04-14T13:58:30Z"
+    "number": 44108,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "refactor(vitdet): use output tracing decorators",
+    "updated_at": "2026-02-18T21:19:27Z"
   },
   {
-    "additions": 6,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/44077. Indeed, the call is not optional. This is slightly breaking as the defaut used to be False, so fresh model instantiation will now use a different init\u2026",
-    "changed_files": 3,
+    "additions": 40,
+    "author": "fumadari",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Replaces manual `output_hidden_states`/`return_dict` resolution in `MraModel` with `@merge_with_config_defaults` and `@capture_outputs` decorators - Simplifies `MraEncoder` to a plain loop returning a single tensor, removing `\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44082",
-    "created_at": "2026-02-17T10:09:03Z",
-    "deletions": 20,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44107",
+    "created_at": "2026-02-17T19:04:42Z",
+    "deletions": 112,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44082/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44082",
+    "files_url": "https://github.com/huggingface/transformers/pull/44107/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44107",
     "labels": [],
-    "merged": true,
-    "number": 44082,
+    "merged": false,
+    "number": 44107,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix patchtsmixer call to post_init",
-    "updated_at": "2026-02-17T11:05:40Z"
+    "title": "refactor(mra): use output tracing decorators",
+    "updated_at": "2026-02-18T21:19:29Z"
   },
   {
-    "additions": 48,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes #42533 by introducing default flash implementations. cc @vasqu and @cyrilvallez",
-    "changed_files": 6,
+    "additions": 47,
+    "author": "fumadari",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Replace manual `hidden_states`/`attentions` collection in `YosoEncoder` with the `@capture_outputs` decorator and forward hooks - Add `@can_return_tuple` to all 5 wrapper model classes, eliminating manual `return_dict` handlin\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44081",
-    "created_at": "2026-02-17T09:54:01Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44106",
+    "created_at": "2026-02-17T18:59:25Z",
+    "deletions": 132,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44081/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44081",
+    "files_url": "https://github.com/huggingface/transformers/pull/44106/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44106",
     "labels": [],
-    "merged": true,
-    "number": 44081,
-    "review_comments_count": 1,
+    "merged": false,
+    "number": 44106,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "add default flash impl",
-    "updated_at": "2026-02-19T11:29:54Z"
+    "title": "Refactor yoso to use automatic output tracing",
+    "updated_at": "2026-02-18T21:19:30Z"
   },
   {
-    "additions": 22,
-    "author": "tomaarsen",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? * Patch `__setitem__` on `ModelOutput` even if the parameter was previously `None` Fixes #44079, follow-up from #44050. Essentially, it brings behaviour to the expected as described in #44079: > If I 1) initialize a\u2026",
-    "changed_files": 2,
+    "additions": 39,
+    "author": "fumadari",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Replace manual `hidden_states`/`attentions` collection in `LiltEncoder` with the `@capture_outputs` decorator and forward hooks - Add `@can_return_tuple` to all 3 wrapper model classes, eliminating manual `return_dict` handlin\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 4,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44080",
-    "created_at": "2026-02-17T09:53:36Z",
-    "deletions": 8,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44105",
+    "created_at": "2026-02-17T18:54:40Z",
+    "deletions": 127,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44080/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44080",
+    "files_url": "https://github.com/huggingface/transformers/pull/44105/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44105",
     "labels": [],
-    "merged": true,
-    "number": 44080,
+    "merged": false,
+    "number": 44105,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Patch `__setitem__` on `ModelOutput` even if the parameter was previously `None`",
-    "updated_at": "2026-02-20T10:08:38Z"
+    "title": "Refactor lilt to use automatic output tracing",
+    "updated_at": "2026-02-18T21:19:32Z"
   },
   {
-    "additions": 19,
-    "author": "tomaarsen",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? * Set `input_modalities` on various architectures that aren't just text Sentence Transformers would like to rely on `input_modalities` in the future to determine what modalities can be used. However, it's not quite\u2026",
-    "changed_files": 10,
+    "additions": 66,
+    "author": "fumadari",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary - Replace manual `hidden_states`/`attentions`/`cross_attentions` collection in `MegatronBertEncoder` with the `@capture_outputs` decorator and forward hooks - Add `@can_return_tuple` to all 8 wrapper model classes, eliminating m\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44078",
-    "created_at": "2026-02-17T09:15:34Z",
-    "deletions": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44104",
+    "created_at": "2026-02-17T18:43:44Z",
+    "deletions": 207,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44078/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44078",
+    "files_url": "https://github.com/huggingface/transformers/pull/44104/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44104",
     "labels": [],
-    "merged": true,
-    "number": 44078,
-    "review_comments_count": 6,
+    "merged": false,
+    "number": 44104,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[`fix`] Set input_modalities on various architectures that aren't just text",
-    "updated_at": "2026-02-24T10:39:31Z"
+    "title": "Refactor megatron_bert to use automatic output tracing",
+    "updated_at": "2026-02-18T21:19:34Z"
   },
   {
-    "additions": 11,
-    "author": "mmahjoub5",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR refactors the ImageGPT implementation to adopt the new `@capture_outputs` and `@can_return_tuple` decorators for standardized output collection, as part of https://github.com/huggingface/transformers/issues/\u2026",
-    "changed_files": 1,
+    "additions": 53,
+    "author": "engmohamedsalah",
+    "author_association": "NONE",
+    "body_excerpt": "Fixes #44052 Now and then, the indexer ran into trouble switching between masks and cache. Most of the test failures came from these hiccups: - Indexer cache: the old if seq_len > 1: reset cache heuristic broke assisted decoding (multi-tok\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44076",
-    "created_at": "2026-02-17T08:46:55Z",
-    "deletions": 62,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44103",
+    "created_at": "2026-02-17T18:04:48Z",
+    "deletions": 76,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44076/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44076",
+    "files_url": "https://github.com/huggingface/transformers/pull/44103/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44103",
     "labels": [],
     "merged": false,
-    "number": 44076,
+    "number": 44103,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Refectored modeling_imagegpt.py to enable hooks to capture_outputs",
-    "updated_at": "2026-02-18T04:11:40Z"
+    "state": "closed",
+    "title": "Fix glm_moe_dsa",
+    "updated_at": "2026-02-18T19:38:11Z"
   },
   {
-    "additions": 66,
-    "author": "aman-coder03",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? This PR migrates TextNet to the new standardized output tracing system using the `@capture_outputs` and `@can_return_tuple` decorators. It adds `_can_record_outputs`, applies `@capture_outputs` to `TextNetModel.for\u2026",
-    "changed_files": 2,
+    "additions": 42,
+    "author": "fumadari",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Refactors the `ibert` model to use the new `@capture_outputs` and `@can_return_tuple` decorators for output tracing, as part of the meta-issue #43979. **Key changes:** - Added `_can_record_outputs = {\"hidden_states\": IBertLayer,\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44074",
-    "created_at": "2026-02-17T08:23:25Z",
-    "deletions": 52,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44102",
+    "created_at": "2026-02-17T17:21:32Z",
+    "deletions": 154,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44074/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44074",
+    "files_url": "https://github.com/huggingface/transformers/pull/44102/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44102",
     "labels": [],
     "merged": false,
-    "number": 44074,
+    "number": 44102,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[TextNet] Refactor output tracing using capture_outputs decorator",
-    "updated_at": "2026-02-17T11:28:11Z"
+    "state": "closed",
+    "title": "Refactor ibert output tracing with capture_outputs",
+    "updated_at": "2026-02-18T21:19:35Z"
   },
   {
-    "additions": 32,
+    "additions": 210,
     "author": "aman-coder03",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? This PR migrates VisualBert to the new standardized output tracing system using the `@capture_outputs` and `@can_return_tuple` decorators. Specifically, this PR: - Adds `_can_record_outputs` to `VisualBertPreTraine\u2026",
-    "changed_files": 1,
+    "body_excerpt": "## What does this PR do? This PR refactors XLM's output tracing to align with the standardized output capturing patterns used across the codebase. ### Key changes: - Refactors transformer blocks into a dedicated `XLMLayer` module to enable\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44073",
-    "created_at": "2026-02-17T08:16:59Z",
-    "deletions": 38,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44101",
+    "created_at": "2026-02-17T17:15:06Z",
+    "deletions": 194,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44073/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44073",
+    "files_url": "https://github.com/huggingface/transformers/pull/44101/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44101",
     "labels": [],
     "merged": false,
-    "number": 44073,
+    "number": 44101,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[VisualBert] Refactor output tracing using capture_outputs and can_return_tuple decorators",
-    "updated_at": "2026-02-17T11:29:01Z"
+    "title": "[XLM] Refactor output tracing to align with capture_outputs standardized architecture",
+    "updated_at": "2026-02-19T08:08:33Z"
   },
   {
-    "additions": 12,
-    "author": "Siddhartha7340",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# Refactor efficientnet output tracing # What does this PR do? This Pull Request migrates the EfficientNet model to use the standardized @capture_outputs and @can_return_tuple decorators. - Added _can_record_outputs to `EfficientNetPreTrai\u2026",
+    "additions": 3,
+    "author": "qgallouedec",
+    "author_association": "MEMBER",
+    "body_excerpt": "In https://github.com/huggingface/trl/pull/5112 a user reported that `trl sft --help` fails It's because three inherited args from `TrainingArguments` (`torch_empty_cache_steps`, `gradient_checkpointing` and `use_liger_kernel`)help strings\u2026",
     "changed_files": 1,
-    "cluster_id": "cluster-43979-11",
-    "cluster_ids": [
-      "cluster-43979-11"
-    ],
-    "cluster_role": "member",
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44072",
-    "created_at": "2026-02-17T07:42:01Z",
-    "deletions": 38,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44100",
+    "created_at": "2026-02-17T17:10:36Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44072/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44072",
+    "files_url": "https://github.com/huggingface/transformers/pull/44100/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44100",
     "labels": [],
-    "merged": false,
-    "number": 44072,
+    "merged": true,
+    "number": 44100,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "refactor efficientnet output tracing with @capture_outputs and @can_r\u2026",
-    "updated_at": "2026-02-17T07:56:05Z"
+    "state": "closed",
+    "title": "Fix percentage formatting in help messages for gradient checkpointing, Liger Kernel, and empty cache steps",
+    "updated_at": "2026-02-20T09:57:51Z"
   },
   {
-    "additions": 38,
-    "author": "ArivunidhiA",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? Refactors the MPT model to use the new standardized output collection interface as part of #43979. ### Changes: - Added `_can_record_outputs` to `MptPreTrainedModel` mapping `hidden_states` \u2192 `MptBlock` and `attent\u2026",
-    "changed_files": 1,
+    "additions": 2,
+    "author": "qgallouedec",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44071",
-    "created_at": "2026-02-17T07:19:17Z",
-    "deletions": 112,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44099",
+    "created_at": "2026-02-17T16:45:35Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44071/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44071",
+    "files_url": "https://github.com/huggingface/transformers/pull/44099/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44099",
     "labels": [],
-    "merged": false,
-    "number": 44071,
+    "merged": true,
+    "number": 44099,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[Refactor] Migrate MPT to standardized output tracing decorators",
-    "updated_at": "2026-02-17T07:20:17Z"
+    "state": "closed",
+    "title": "Bump dev version",
+    "updated_at": "2026-02-18T10:03:54Z"
   },
   {
-    "additions": 272,
-    "author": "rudybear",
+    "additions": 125,
+    "author": "aman-coder03",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - Add GGUF config mapping, defaults, and tokenizer converter for `qwen3_next` (Qwen3-Coder-Next, hybrid DeltaNet+Attention MoE, 80B total / 3B active) - Add `Qwen3NextTensorProcessor` handling DeltaNet-specific tensor transforms\u2026",
-    "changed_files": 3,
+    "body_excerpt": "## What does this PR do? This PR refactors ViLT's output handling to align with the standardized patterns used across the codebase. Key changes: - Removes manual `hidden_states`/`attentions` propagation and passes `output_attentions`, `out\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44070",
-    "created_at": "2026-02-17T07:18:13Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44098",
+    "created_at": "2026-02-17T16:32:34Z",
+    "deletions": 138,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44070/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44070",
+    "files_url": "https://github.com/huggingface/transformers/pull/44098/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44098",
     "labels": [],
     "merged": false,
-    "number": 44070,
+    "number": 44098,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Add GGUF loading support for Qwen3-Next (qwen3_next) architecture",
-    "updated_at": "2026-02-17T07:21:26Z"
+    "title": "[ViLT] Refactor output handling to align with standardized patterns",
+    "updated_at": "2026-02-17T16:37:46Z"
   },
   {
-    "additions": 26,
-    "author": "cyyever",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR uses ``torch.isfinite`` to simplify conditions, and the CUDA sync calls may also be reduced.",
-    "changed_files": 26,
+    "additions": 12,
+    "author": "Rocketknight1",
+    "author_association": "MEMBER",
+    "body_excerpt": "The `test_keep_in_fp32_modules` issues in #44052 are because the test assumes a model has **either** `_keep_in_fp32_modules` or `_keep_in_fp32_modules_strict` **but not both.** The only model that uses both is `glm_moe_dsa`, so this is the\u2026",
+    "changed_files": 2,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44097",
+    "created_at": "2026-02-17T15:43:55Z",
+    "deletions": 42,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44097/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44097",
+    "labels": [],
+    "merged": true,
+    "number": 44097,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "Merge test_keep_in_fp32_modules and test_keep_in_fp32_modules_strict",
+    "updated_at": "2026-02-17T16:23:33Z"
+  },
+  {
+    "additions": 3,
+    "author": "eustlb",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Tests were written (and pass) on DGX A100, here are the values for our runners.",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44069",
-    "created_at": "2026-02-17T06:49:38Z",
-    "deletions": 48,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44096",
+    "created_at": "2026-02-17T15:14:26Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44069/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44069",
+    "files_url": "https://github.com/huggingface/transformers/pull/44096/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44096",
     "labels": [],
     "merged": true,
-    "number": 44069,
+    "number": 44096,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Use torch.isfinite",
-    "updated_at": "2026-02-18T01:04:19Z"
+    "title": "[voxtral-realtime] update runner expected values ",
+    "updated_at": "2026-02-17T15:23:19Z"
   },
   {
-    "additions": 42,
-    "author": "mtthw13",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Replaces manual `output_attentions`/`output_hidden_states`/`return_dict` boilerplate in GPT-Neo with the hook-based decorator system. **Changes:** - Added `_can_record_outputs = {\"hidden_states\": GPTNeoBlock, \"attentions\": GPTNeoAttention}\u2026",
+    "additions": 43,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title. The check that was added in https://github.com/huggingface/transformers/pull/43768 is wrong, as a missing weight would NOT be reinitialized in some cases! As for the pointers check, it is actually\u2026",
     "changed_files": 2,
-    "cluster_id": "cluster-44018-2",
-    "cluster_ids": [
-      "cluster-44018-2"
-    ],
-    "cluster_role": "canonical",
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44068",
-    "created_at": "2026-02-17T06:13:37Z",
-    "deletions": 119,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44095",
+    "created_at": "2026-02-17T14:33:22Z",
+    "deletions": 47,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44068/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44068",
+    "files_url": "https://github.com/huggingface/transformers/pull/44095/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44095",
     "labels": [],
-    "merged": false,
-    "number": 44068,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Refactor GPT-Neo to use `@capture_outputs` and `@can_return_tuple` decorators",
-    "updated_at": "2026-02-18T08:30:32Z"
+    "merged": true,
+    "number": 44095,
+    "review_comments_count": 2,
+    "state": "closed",
+    "title": "Fix loading logic issue",
+    "updated_at": "2026-04-03T04:52:35Z"
   },
   {
-    "additions": 63,
-    "author": "23atharvaS",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? This PR introduces a new argument `eval_on_end` to the `Trainer` class. When enabled, the Trainer automatically runs evaluation at the end of training. This allows users to obtain final evaluation metrics without e\u2026",
-    "changed_files": 3,
+    "additions": 24,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Refactors the `codegen` model to use the `@capture_outputs` and `@can_return_tuple` decorators, replacing manual `output_attentions`/`output_hidden_states`/`return_dict` handling. ### Changes: - Add `_can_record_ou\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44067",
-    "created_at": "2026-02-17T05:25:26Z",
-    "deletions": 1,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44094",
+    "created_at": "2026-02-17T14:15:10Z",
+    "deletions": 70,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44067/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44067",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44094/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44094",
+    "labels": [],
     "merged": false,
-    "number": 44067,
+    "number": 44094,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Add `eval_on_end` argument to Trainer for final evaluation after training",
-    "updated_at": "2026-02-17T13:32:34Z"
+    "title": "Refactor codegen model to use @capture_outputs and @can_return_tuple decorators",
+    "updated_at": "2026-03-03T00:30:07Z"
   },
   {
-    "additions": 35,
-    "author": "Jay-IIT",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Migrate GPT-J from manual boilerplate output collection to the new decorator-based output tracing system: - Add `_can_record_outputs` to `GPTJPreTrainedModel` - Add `@capture_outputs` and `@merge_with_config_defaults` to `GPTJModel.forward\u2026",
-    "changed_files": 2,
-    "cluster_id": "cluster-43979-11",
-    "cluster_ids": [
-      "cluster-43979-11"
-    ],
-    "cluster_role": "member",
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44066",
-    "created_at": "2026-02-17T05:12:11Z",
-    "deletions": 107,
+    "additions": 28,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Refactors the DeBERTa model to use the standardized output collection interface with `@capture_outputs` and `@can_return_tuple` decorators. ### Changes - Add `_can_record_outputs` on `DebertaPreTrainedModel` mapping `hidden_stat\u2026",
+    "changed_files": 1,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44093",
+    "created_at": "2026-02-17T14:15:07Z",
+    "deletions": 129,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44066/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44066",
+    "files_url": "https://github.com/huggingface/transformers/pull/44093/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44093",
     "labels": [],
     "merged": false,
-    "number": 44066,
+    "number": 44093,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Refactor GPT-J to use standardized output tracing (#43979)",
-    "updated_at": "2026-02-18T18:44:28Z"
+    "state": "closed",
+    "title": "Refactor DeBERTa output tracing with @capture_outputs and @can_return_tuple",
+    "updated_at": "2026-03-03T00:30:09Z"
   },
   {
-    "additions": 21,
-    "author": "tysoncung",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Refactors the CTRL model to use the standardized output collection interface as part of #43979. ## Changes - Added `_can_record_outputs` to `CTRLPreTrainedModel` mapping `hidden_states` \u2192 `EncoderLayer` and `attentions` \u2192 `Multi\u2026",
-    "changed_files": 1,
+    "additions": 79,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Refactors the **swinv2** model to use hook-based output capturing via `@capture_outputs` and `@can_return_tuple` decorators, removing the manual threading of `output_attentions`/`output_hidden_states`/`return_dict`\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44065",
-    "created_at": "2026-02-17T02:03:57Z",
-    "deletions": 76,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44092",
+    "created_at": "2026-02-17T14:15:04Z",
+    "deletions": 159,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44065/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44065",
+    "files_url": "https://github.com/huggingface/transformers/pull/44092/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44092",
     "labels": [],
     "merged": false,
-    "number": 44065,
+    "number": 44092,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor CTRL model output tracing with capture_outputs and can_return_tuple",
-    "updated_at": "2026-02-25T00:49:18Z"
+    "title": "Refactor output tracing for swinv2 model",
+    "updated_at": "2026-03-03T00:30:10Z"
   },
   {
-    "additions": 57,
-    "author": "mariam851",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Description: Adds eval_on_end to TrainingArguments to trigger a final evaluation automatically after training finishes. Key Changes: TrainingArguments: Added eval_on_end boolean flag. Trainer.train: Logic to call evaluate() and merge metri\u2026",
+    "additions": 79,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Refactors the `swin` model to use the standardized output collection interface (`@capture_outputs` and `@can_return_tuple` decorators), as described in #43979. ### Changes **SwinPreTrainedModel:** - Added `_can_rec\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 0,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44064",
-    "created_at": "2026-02-17T01:10:31Z",
-    "deletions": 16,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44091",
+    "created_at": "2026-02-17T14:14:56Z",
+    "deletions": 146,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44064/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44064",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44091/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44091",
+    "labels": [],
     "merged": false,
-    "number": 44064,
+    "number": 44091,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "feat: implement eval_on_end to trigger evaluation after training",
-    "updated_at": "2026-02-17T13:32:40Z"
+    "title": "Refactor Swin output tracing with @capture_outputs and @can_return_tuple",
+    "updated_at": "2026-03-03T00:30:11Z"
   },
   {
-    "additions": 229,
-    "author": "AutumnAurelium",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This brings the Arcee AFMoE architecture in line with other MoE models' implementation patterns since v5. It also adds integration testing using Trinity Nano. ## Before submitting - [ ] This PR fixes a typo or impro\u2026",
-    "changed_files": 5,
+    "additions": 25,
+    "author": "itazap",
+    "author_association": "MEMBER",
+    "body_excerpt": "bos and eos behaviour should match when updating post processor setting `add_bos_token=True` when `bos_token=None` should silently disables `add_bos_token`. (was already the behavior for `eos_token`)",
+    "changed_files": 2,
     "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 5,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44063",
-    "created_at": "2026-02-17T01:07:13Z",
-    "deletions": 150,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44090",
+    "created_at": "2026-02-17T13:15:07Z",
+    "deletions": 4,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44063/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44063",
+    "files_url": "https://github.com/huggingface/transformers/pull/44090/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44090",
     "labels": [],
     "merged": true,
-    "number": 44063,
-    "review_comments_count": 6,
+    "number": 44090,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Update AFMoE architecture to use v5-style MoE impl",
-    "updated_at": "2026-03-19T14:00:46Z"
+    "title": "Update post proc",
+    "updated_at": "2026-02-18T15:34:18Z"
   },
   {
-    "additions": 2,
-    "author": "tarekziade",
-    "author_association": "MEMBER",
-    "body_excerpt": "Reproduced locally with ``` pytest -q -m generate --random-order-bucket=none --flake-finder --flake-runs=200 tests/models/kosmos2/test_modeling_kosmos2.py -k test_assisted_decoding_matches_greedy_search ``` Root cause: - prepare_config_and\u2026",
-    "changed_files": 1,
+    "additions": 113,
+    "author": "preetam1407",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Fixes #43979 ## Summary Refactor T5 to the standardized output tracing interface. ## Changes - Added `_can_record_outputs` on T5 encoder/decoder stack subclasses. - Added `@capture_outputs` on the base stack forward. - Added `@can_return_t\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44061",
-    "created_at": "2026-02-16T22:08:48Z",
-    "deletions": 0,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44089",
+    "created_at": "2026-02-17T11:37:18Z",
+    "deletions": 294,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44061/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44061",
+    "files_url": "https://github.com/huggingface/transformers/pull/44089/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44089",
     "labels": [],
-    "merged": true,
-    "number": 44061,
+    "merged": false,
+    "number": 44089,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix: flaky `Kosmos2ModelTest` test",
-    "updated_at": "2026-02-18T14:23:30Z"
+    "title": "Refactor t5 output tracing",
+    "updated_at": "2026-02-17T13:45:23Z"
   },
   {
-    "additions": 44,
-    "author": "lakprigan",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "## Summary Migrates GPT2 to the standardized output collection interface as part of #43979. - Added `_can_record_outputs` to `GPT2PreTrainedModel` (including `cross_attentions` via `OutputRecorder` targeting the `crossattention` submodule)\u2026",
+    "additions": 41,
+    "author": "Zephyr-Blessed",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Description Refactors GPT-2 model to use the standardized `@capture_outputs` and `@can_return_tuple` decorators, replacing manual output collection boilerplate. Part of #43979 ## Changes - **`GPT2PreTrainedModel`**: Added `_can_record_o\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44059",
-    "created_at": "2026-02-16T20:14:30Z",
-    "deletions": 133,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44088",
+    "created_at": "2026-02-17T11:32:42Z",
+    "deletions": 129,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44059/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44059",
+    "files_url": "https://github.com/huggingface/transformers/pull/44088/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44088",
     "labels": [],
     "merged": false,
-    "number": 44059,
+    "number": 44088,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "[GPT2] Refactor output tracing to use capture_outputs/can_return_tuple decorators",
-    "updated_at": "2026-02-25T17:47:45Z"
+    "state": "closed",
+    "title": "Refactor GPT-2 output tracing with capture_outputs/can_return_tuple",
+    "updated_at": "2026-02-17T11:41:32Z"
   },
   {
-    "additions": 122,
-    "author": "engmohamedsalah",
+    "additions": 16,
+    "author": "huyxdang",
     "author_association": "NONE",
-    "body_excerpt": "## Summary Fixes #44052 \u2014 resolves 10 of 11 skipped tests for the `glm_moe_dsa` model. **Root causes fixed:** - **DSA indexer mask shape mismatch**: The attention mask was not properly normalized to 4D before being passed to the indexer an\u2026",
-    "changed_files": 3,
+    "body_excerpt": "### Summary Refactors the Mamba2 model to use the standardized output collection interface as part of #43979. ### Changes * **Standardized Output Mapping**: Added `_can_record_outputs` to `Mamba2PreTrainedModel` mapping `hidden_states` \u2192 `\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44058",
-    "created_at": "2026-02-16T19:24:30Z",
-    "deletions": 84,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44087",
+    "created_at": "2026-02-17T11:30:25Z",
+    "deletions": 33,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44058/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44058",
-    "labels": [
-      "Code agent slop"
-    ],
+    "files_url": "https://github.com/huggingface/transformers/pull/44087/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44087",
+    "labels": [],
     "merged": false,
-    "number": 44058,
+    "number": 44087,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix skipped tests for glm_moe_dsa model",
-    "updated_at": "2026-02-17T17:23:03Z"
+    "title": "Refactor Mamba2 to use standardized output tracing",
+    "updated_at": "2026-03-11T02:08:22Z"
   },
   {
-    "additions": 0,
-    "author": "mariam851",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR addresses memory efficiency issues in the Qwen2Moe implementation (reported in #43856). Users experienced Out-of-Memory (OOM) errors during quantization and inference, particularly with large reserved memory (e.g., 27GB on H100) th\u2026",
-    "changed_files": 0,
+    "additions": 16,
+    "author": "Zephyr-Blessed",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Migrates **MGP-STR** to the standardized output collection interface using `@capture_outputs` and `@can_return_tuple` decorators. ### Changes: - Added `_can_record_outputs = {\"hidden_states\": MgpstrLayer, \"attentio\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44057",
-    "created_at": "2026-02-16T18:35:01Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44086",
+    "created_at": "2026-02-17T11:21:22Z",
+    "deletions": 48,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44057/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44057",
+    "files_url": "https://github.com/huggingface/transformers/pull/44086/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44086",
     "labels": [],
     "merged": false,
-    "number": 44057,
+    "number": 44086,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "fix(qwen3_moe): optimize memory and fix OOM in MoE layers",
-    "updated_at": "2026-02-16T21:47:41Z"
+    "state": "open",
+    "title": "[MGP-STR] Refactor output tracing to use capture_outputs/can_return_tuple decorators",
+    "updated_at": "2026-02-17T11:22:25Z"
   },
   {
-    "additions": 50,
-    "author": "aman-coder03",
+    "additions": 37,
+    "author": "Zephyr-Blessed",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## What does this PR do? This PR migrates MPNet to the new standardized output tracing system using the `@capture_outputs` decorator. Specifically, this PR: - Applies `@capture_outputs` to `MPNetModel.forward` - Removes manual accumulation\u2026",
+    "body_excerpt": "## What does this PR do? Refactors the RemBERT model to use the new output tracing decorators (`@capture_outputs` and `@can_return_tuple`), replacing manual output collection boilerplate. ### Changes: - Added `@capture_outputs` decorator t\u2026",
     "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
+    "cluster_id": "cluster-43979-11",
+    "cluster_ids": [
+      "cluster-43979-11"
+    ],
+    "cluster_role": "member",
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44056",
-    "created_at": "2026-02-16T18:27:05Z",
-    "deletions": 14,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44085",
+    "created_at": "2026-02-17T11:09:55Z",
+    "deletions": 108,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44056/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44056",
+    "files_url": "https://github.com/huggingface/transformers/pull/44085/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44085",
     "labels": [],
     "merged": false,
-    "number": 44056,
+    "number": 44085,
     "review_comments_count": 0,
     "state": "open",
-    "title": "[MPNet] Refactor output tracing using capture_outputs decorator",
-    "updated_at": "2026-02-17T11:23:12Z"
+    "title": "Refactor RemBERT to use output tracing decorators",
+    "updated_at": "2026-02-17T11:10:59Z"
   },
   {
-    "additions": 5,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do?",
+    "additions": 37,
+    "author": "Zephyr-Blessed",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Refactors the GPT-J model to use the new `capture_outputs` and `can_return_tuple` decorators for output tracing, following the same pattern as #44046 (CodeGen). ### Changes: - Added `@capture_outputs` decorator on\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44055",
-    "created_at": "2026-02-16T18:26:43Z",
-    "deletions": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44084",
+    "created_at": "2026-02-17T11:08:48Z",
+    "deletions": 108,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44055/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44055",
+    "files_url": "https://github.com/huggingface/transformers/pull/44084/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44084",
     "labels": [],
-    "merged": true,
-    "number": 44055,
+    "merged": false,
+    "number": 44084,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix unprotected torch import",
-    "updated_at": "2026-02-16T18:43:01Z"
+    "title": "[GPT-J] Refactor output tracing to use capture_outputs/can_return_tuple decorators",
+    "updated_at": "2026-02-17T11:41:38Z"
   },
   {
-    "additions": 346,
-    "author": "ArthurZucker",
+    "additions": 1555,
+    "author": "3outeille",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Add flash MLA interface. - It does not work I get a segfault - we don't leverage the paged cache so it's not as efficient as that I reckon. ```bash Fetching 6 files: 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2026",
-    "changed_files": 10,
+    "body_excerpt": "- TODO: - fsdp => faire comme tp en mode fsdp_plan manual qui devient l'auto par d\u00e9faut --- This PR introduces **first-class FSDP2 (Fully Sharded Data Parallel v2) support** directly in Transformers, bypassing the need for Accelerate's FSD\u2026",
+    "changed_files": 11,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44054",
-    "created_at": "2026-02-16T18:07:14Z",
-    "deletions": 93,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44054/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44054",
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44083",
+    "created_at": "2026-02-17T10:57:06Z",
+    "deletions": 120,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44083/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44083",
     "labels": [],
     "merged": false,
-    "number": 44054,
-    "review_comments_count": 0,
+    "number": 44083,
+    "review_comments_count": 24,
     "state": "open",
-    "title": "Flash mla interface",
-    "updated_at": "2026-02-20T11:14:39Z"
+    "title": "FSDP2 native support in transformers ",
+    "updated_at": "2026-04-14T13:58:30Z"
   },
   {
-    "additions": 2,
+    "additions": 6,
     "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do?",
-    "changed_files": 1,
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/44077. Indeed, the call is not optional. This is slightly breaking as the defaut used to be False, so fresh model instantiation will now use a different init\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44053",
-    "created_at": "2026-02-16T17:59:48Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44082",
+    "created_at": "2026-02-17T10:09:03Z",
+    "deletions": 20,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44053/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44053",
+    "files_url": "https://github.com/huggingface/transformers/pull/44082/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44082",
+    "labels": [],
+    "merged": true,
+    "number": 44082,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Fix patchtsmixer call to post_init",
+    "updated_at": "2026-02-17T11:05:40Z"
+  },
+  {
+    "additions": 48,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes #42533 by introducing default flash implementations. cc @vasqu and @cyrilvallez",
+    "changed_files": 6,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44081",
+    "created_at": "2026-02-17T09:54:01Z",
+    "deletions": 0,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44081/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44081",
     "labels": [],
     "merged": true,
-    "number": 44053,
-    "review_comments_count": 0,
+    "number": 44081,
+    "review_comments_count": 1,
     "state": "closed",
-    "title": "Fix peft conversion typo",
-    "updated_at": "2026-02-17T11:12:19Z"
+    "title": "add default flash impl",
+    "updated_at": "2026-02-19T11:29:54Z"
   },
   {
-    "additions": 2,
+    "additions": 22,
     "author": "tomaarsen",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? * Fix incorrect forward type hint for Gemma3n ## Details The type hint didn't match the actual returned class: https://github.com/huggingface/transformers/blob/349e00c1a367ce263624e525038250625dcf20c7/src/transforme\u2026",
+    "body_excerpt": "# What does this PR do? * Patch `__setitem__` on `ModelOutput` even if the parameter was previously `None` Fixes #44079, follow-up from #44050. Essentially, it brings behaviour to the expected as described in #44079: > If I 1) initialize a\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44051",
-    "created_at": "2026-02-16T17:26:24Z",
-    "deletions": 2,
+    "comments_count": 4,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44080",
+    "created_at": "2026-02-17T09:53:36Z",
+    "deletions": 8,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44051/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44051",
+    "files_url": "https://github.com/huggingface/transformers/pull/44080/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44080",
     "labels": [],
     "merged": true,
-    "number": 44051,
+    "number": 44080,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[`chore`] Fix incorrect forward type hint for Gemma3n",
-    "updated_at": "2026-02-20T09:08:07Z"
+    "title": "Patch `__setitem__` on `ModelOutput` even if the parameter was previously `None`",
+    "updated_at": "2026-02-20T10:08:38Z"
   },
   {
-    "additions": 15,
+    "additions": 19,
     "author": "tomaarsen",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? * Patch `get_text_features` for ChineseCLIP ### Details The `get_text_features` assumes that the `text_model` returns a `BaseModelOutputWithPooling`, just like is done with many other models. Currently, the `get_tex\u2026",
-    "changed_files": 7,
+    "body_excerpt": "# What does this PR do? * Set `input_modalities` on various architectures that aren't just text Sentence Transformers would like to rely on `input_modalities` in the future to determine what modalities can be used. However, it's not quite\u2026",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44050",
-    "created_at": "2026-02-16T17:23:31Z",
-    "deletions": 19,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44078",
+    "created_at": "2026-02-17T09:15:34Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44050/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44050",
+    "files_url": "https://github.com/huggingface/transformers/pull/44078/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44078",
     "labels": [],
-    "merged": false,
-    "number": 44050,
-    "review_comments_count": 4,
+    "merged": true,
+    "number": 44078,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "[`fix`] Patch `get_text_features` for ChineseCLIP",
-    "updated_at": "2026-02-17T09:55:17Z"
+    "title": "[`fix`] Set input_modalities on various architectures that aren't just text",
+    "updated_at": "2026-02-24T10:39:31Z"
   },
   {
-    "additions": 59,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the `fnet` model to use the new `@capture_outputs` and `@can_return_tuple` decorators, as part of #43979. ### Changes: - Added `_can_record_outputs = {\"hidden_states\": FNetLayer}` to `FNetPreTrainedModel`\u2026",
+    "additions": 11,
+    "author": "mmahjoub5",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR refactors the ImageGPT implementation to adopt the new `@capture_outputs` and `@can_return_tuple` decorators for standardized output collection, as part of https://github.com/huggingface/transformers/issues/\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44049",
-    "created_at": "2026-02-16T17:19:04Z",
-    "deletions": 112,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44076",
+    "created_at": "2026-02-17T08:46:55Z",
+    "deletions": 62,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44049/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44049",
+    "files_url": "https://github.com/huggingface/transformers/pull/44076/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44076",
     "labels": [],
     "merged": false,
-    "number": 44049,
+    "number": 44076,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Refactor fnet model to use @capture_outputs and @can_return_tuple decorators",
-    "updated_at": "2026-03-03T00:30:13Z"
+    "state": "open",
+    "title": "Refectored modeling_imagegpt.py to enable hooks to capture_outputs",
+    "updated_at": "2026-02-18T04:11:40Z"
   },
   {
-    "additions": 4,
-    "author": "tomaarsen",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? * Fix up `__repr__` whitespace/brackets ## Reproducer ```python from transformers import AutoTokenizer, PreTrainedTokenizerBase # __repr__ via PreTrainedTokenizerBase tokenizer = AutoTokenizer.from_pretrained(\"bert-\u2026",
+    "additions": 66,
+    "author": "aman-coder03",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? This PR migrates TextNet to the new standardized output tracing system using the `@capture_outputs` and `@can_return_tuple` decorators. It adds `_can_record_outputs`, applies `@capture_outputs` to `TextNetModel.for\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44048",
-    "created_at": "2026-02-16T17:18:10Z",
-    "deletions": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44074",
+    "created_at": "2026-02-17T08:23:25Z",
+    "deletions": 52,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44048/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44048",
+    "files_url": "https://github.com/huggingface/transformers/pull/44074/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44074",
     "labels": [],
-    "merged": true,
-    "number": 44048,
+    "merged": false,
+    "number": 44074,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "[`simple`] Fix up `__repr__` whitespace/brackets",
-    "updated_at": "2026-02-20T10:03:34Z"
+    "state": "open",
+    "title": "[TextNet] Refactor output tracing using capture_outputs decorator",
+    "updated_at": "2026-02-17T11:28:11Z"
   },
   {
-    "additions": 35,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the `bloom` model to use the new `@capture_outputs` and `@can_return_tuple` decorators, as part of the effort in #43979. ### Changes: - Add `_can_record_outputs` dict to `BloomPreTrainedModel` mapping `hi\u2026",
+    "additions": 32,
+    "author": "aman-coder03",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? This PR migrates VisualBert to the new standardized output tracing system using the `@capture_outputs` and `@can_return_tuple` decorators. Specifically, this PR: - Adds `_can_record_outputs` to `VisualBertPreTraine\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44047",
-    "created_at": "2026-02-16T17:15:25Z",
-    "deletions": 104,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44073",
+    "created_at": "2026-02-17T08:16:59Z",
+    "deletions": 38,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44047/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44047",
+    "files_url": "https://github.com/huggingface/transformers/pull/44073/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44073",
     "labels": [],
     "merged": false,
-    "number": 44047,
+    "number": 44073,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Refactor bloom model to use @capture_outputs and @can_return_tuple decorators",
-    "updated_at": "2026-03-03T00:30:14Z"
+    "state": "open",
+    "title": "[VisualBert] Refactor output tracing using capture_outputs and can_return_tuple decorators",
+    "updated_at": "2026-02-17T11:29:01Z"
   },
   {
-    "additions": 24,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the `codegen` model to use the `@capture_outputs` and `@can_return_tuple` decorators, replacing manual `output_attentions`/`output_hidden_states`/`return_dict` handling. ### Changes: - Add `_can_record_ou\u2026",
+    "additions": 12,
+    "author": "Siddhartha7340",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# Refactor efficientnet output tracing # What does this PR do? This Pull Request migrates the EfficientNet model to use the standardized @capture_outputs and @can_return_tuple decorators. - Added _can_record_outputs to `EfficientNetPreTrai\u2026",
+    "changed_files": 1,
+    "cluster_id": "cluster-43979-11",
+    "cluster_ids": [
+      "cluster-43979-11"
+    ],
+    "cluster_role": "member",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44072",
+    "created_at": "2026-02-17T07:42:01Z",
+    "deletions": 38,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44072/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44072",
+    "labels": [],
+    "merged": false,
+    "number": 44072,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "refactor efficientnet output tracing with @capture_outputs and @can_r\u2026",
+    "updated_at": "2026-02-17T07:56:05Z"
+  },
+  {
+    "additions": 38,
+    "author": "ArivunidhiA",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? Refactors the MPT model to use the new standardized output collection interface as part of #43979. ### Changes: - Added `_can_record_outputs` to `MptPreTrainedModel` mapping `hidden_states` \u2192 `MptBlock` and `attent\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44071",
+    "created_at": "2026-02-17T07:19:17Z",
+    "deletions": 112,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44071/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44071",
+    "labels": [],
+    "merged": false,
+    "number": 44071,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "[Refactor] Migrate MPT to standardized output tracing decorators",
+    "updated_at": "2026-02-17T07:20:17Z"
+  },
+  {
+    "additions": 272,
+    "author": "rudybear",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## Summary - Add GGUF config mapping, defaults, and tokenizer converter for `qwen3_next` (Qwen3-Coder-Next, hybrid DeltaNet+Attention MoE, 80B total / 3B active) - Add `Qwen3NextTensorProcessor` handling DeltaNet-specific tensor transforms\u2026",
+    "changed_files": 3,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44046",
-    "created_at": "2026-02-16T17:07:38Z",
-    "deletions": 70,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44070",
+    "created_at": "2026-02-17T07:18:13Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44046/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44046",
+    "files_url": "https://github.com/huggingface/transformers/pull/44070/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44070",
+    "labels": [],
+    "merged": false,
+    "number": 44070,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "Add GGUF loading support for Qwen3-Next (qwen3_next) architecture",
+    "updated_at": "2026-04-16T11:12:13Z"
+  },
+  {
+    "additions": 26,
+    "author": "cyyever",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR uses ``torch.isfinite`` to simplify conditions, and the CUDA sync calls may also be reduced.",
+    "changed_files": 26,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44069",
+    "created_at": "2026-02-17T06:49:38Z",
+    "deletions": 48,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44069/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44069",
+    "labels": [],
+    "merged": true,
+    "number": 44069,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Use torch.isfinite",
+    "updated_at": "2026-02-18T01:04:19Z"
+  },
+  {
+    "additions": 42,
+    "author": "mtthw13",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "Replaces manual `output_attentions`/`output_hidden_states`/`return_dict` boilerplate in GPT-Neo with the hook-based decorator system. **Changes:** - Added `_can_record_outputs = {\"hidden_states\": GPTNeoBlock, \"attentions\": GPTNeoAttention}\u2026",
+    "changed_files": 2,
+    "cluster_id": "cluster-44018-2",
+    "cluster_ids": [
+      "cluster-44018-2"
+    ],
+    "cluster_role": "canonical",
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44068",
+    "created_at": "2026-02-17T06:13:37Z",
+    "deletions": 119,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44068/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44068",
     "labels": [],
     "merged": false,
-    "number": 44046,
+    "number": 44068,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Refactor codegen model to use @capture_outputs and @can_return_tuple decorators",
-    "updated_at": "2026-02-17T14:15:23Z"
+    "state": "open",
+    "title": "Refactor GPT-Neo to use `@capture_outputs` and `@can_return_tuple` decorators",
+    "updated_at": "2026-02-18T08:30:32Z"
   },
   {
-    "additions": 456215,
-    "author": "ArthurZucker",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 4939,
+    "additions": 63,
+    "author": "23atharvaS",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? This PR introduces a new argument `eval_on_end` to the `Trainer` class. When enabled, the Trainer automatically runs evaluation at the end of training. This allows users to obtain final evaluation metrics without e\u2026",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44045",
-    "created_at": "2026-02-16T17:01:41Z",
-    "deletions": 591028,
-    "draft": true,
-    "files_url": "https://github.com/huggingface/transformers/pull/44045/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44045",
-    "labels": [],
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44067",
+    "created_at": "2026-02-17T05:25:26Z",
+    "deletions": 1,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44067/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44067",
+    "labels": [
+      "Code agent slop"
+    ],
     "merged": false,
-    "number": 44045,
+    "number": 44067,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Flash-mla-interface",
-    "updated_at": "2026-02-16T17:11:51Z"
+    "title": "Add `eval_on_end` argument to Trainer for final evaluation after training",
+    "updated_at": "2026-02-17T13:32:34Z"
   },
   {
-    "additions": 49,
-    "author": "rwtarpit",
+    "additions": 35,
+    "author": "Jay-IIT",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
+    "body_excerpt": "Migrate GPT-J from manual boilerplate output collection to the new decorator-based output tracing system: - Add `_can_record_outputs` to `GPTJPreTrainedModel` - Add `@capture_outputs` and `@merge_with_config_defaults` to `GPTJModel.forward\u2026",
+    "changed_files": 2,
     "cluster_id": "cluster-43979-11",
     "cluster_ids": [
       "cluster-43979-11"
     ],
     "cluster_role": "member",
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44044",
-    "created_at": "2026-02-16T16:43:19Z",
-    "deletions": 112,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44066",
+    "created_at": "2026-02-17T05:12:11Z",
+    "deletions": 107,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44044/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44044",
+    "files_url": "https://github.com/huggingface/transformers/pull/44066/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44066",
     "labels": [],
     "merged": false,
-    "number": 44044,
+    "number": 44066,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Refactor DeBERTa's output tracing interface",
-    "updated_at": "2026-02-16T18:57:29Z"
+    "title": "Refactor GPT-J to use standardized output tracing (#43979)",
+    "updated_at": "2026-02-18T18:44:28Z"
   },
   {
-    "additions": 170,
-    "author": "IlyasMoutawwakil",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 31,
+    "additions": 21,
+    "author": "tysoncung",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "## Summary Refactors the CTRL model to use the standardized output collection interface as part of #43979. ## Changes - Added `_can_record_outputs` to `CTRLPreTrainedModel` mapping `hidden_states` \u2192 `EncoderLayer` and `attentions` \u2192 `Multi\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44043",
-    "created_at": "2026-02-16T16:23:57Z",
-    "deletions": 162,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44065",
+    "created_at": "2026-02-17T02:03:57Z",
+    "deletions": 76,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44043/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44043",
+    "files_url": "https://github.com/huggingface/transformers/pull/44065/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44065",
     "labels": [],
-    "merged": true,
-    "number": 44043,
-    "review_comments_count": 15,
+    "merged": false,
+    "number": 44065,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "`grouped_mm` fallback",
-    "updated_at": "2026-02-23T13:58:09Z"
+    "title": "Refactor CTRL model output tracing with capture_outputs and can_return_tuple",
+    "updated_at": "2026-02-25T00:49:18Z"
   },
   {
-    "additions": 1,
-    "author": "Rocketknight1",
-    "author_association": "MEMBER",
-    "body_excerpt": "Joao is regrettably no longer with us :saluting_face: so we should really stop getting users to ping him! This PR makes @cyrilvallez responsible for `generate` issues outside of VLMs.",
-    "changed_files": 1,
+    "additions": 57,
+    "author": "mariam851",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "Description: Adds eval_on_end to TrainingArguments to trigger a final evaluation automatically after training finishes. Key Changes: TrainingArguments: Added eval_on_end boolean flag. Trainer.train: Logic to call evaluate() and merge metri\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44042",
-    "created_at": "2026-02-16T16:00:36Z",
-    "deletions": 1,
+    "comments_count": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44064",
+    "created_at": "2026-02-17T01:10:31Z",
+    "deletions": 16,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44042/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44042",
-    "labels": [],
-    "merged": true,
-    "number": 44042,
+    "files_url": "https://github.com/huggingface/transformers/pull/44064/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44064",
+    "labels": [
+      "Code agent slop"
+    ],
+    "merged": false,
+    "number": 44064,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Update assignee for generate in bug report template",
-    "updated_at": "2026-02-16T16:09:19Z"
+    "title": "feat: implement eval_on_end to trigger evaluation after training",
+    "updated_at": "2026-02-17T13:32:40Z"
   },
   {
-    "additions": 469,
-    "author": "winglian",
-    "author_association": "COLLABORATOR",
-    "body_excerpt": "# What does this PR do? Alternate PR to #43985 to be a reorder only PR. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, s\u2026",
-    "changed_files": 4,
+    "additions": 229,
+    "author": "AutumnAurelium",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This brings the Arcee AFMoE architecture in line with other MoE models' implementation patterns since v5. It also adds integration testing using Trinity Nano. ## Before submitting - [ ] This PR fixes a typo or impro\u2026",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 9,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44041",
-    "created_at": "2026-02-16T15:40:41Z",
-    "deletions": 457,
+    "comments_count": 5,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44063",
+    "created_at": "2026-02-17T01:07:13Z",
+    "deletions": 150,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44041/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44041",
+    "files_url": "https://github.com/huggingface/transformers/pull/44063/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44063",
     "labels": [],
     "merged": true,
-    "number": 44041,
-    "review_comments_count": 14,
+    "number": 44063,
+    "review_comments_count": 6,
     "state": "closed",
-    "title": "refactor _inner_training_loop to smaller methods",
-    "updated_at": "2026-02-23T16:52:09Z"
+    "title": "Update AFMoE architecture to use v5-style MoE impl",
+    "updated_at": "2026-03-19T14:00:46Z"
   },
   {
-    "additions": 366,
-    "author": "zucchini-nlp",
+    "additions": 2,
+    "author": "tarekziade",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/44008 and re-enables tests",
-    "changed_files": 5,
+    "body_excerpt": "Reproduced locally with ``` pytest -q -m generate --random-order-bucket=none --flake-finder --flake-runs=200 tests/models/kosmos2/test_modeling_kosmos2.py -k test_assisted_decoding_matches_greedy_search ``` Root cause: - prepare_config_and\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44040",
-    "created_at": "2026-02-16T12:43:28Z",
-    "deletions": 230,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44061",
+    "created_at": "2026-02-16T22:08:48Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44040/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44040",
+    "files_url": "https://github.com/huggingface/transformers/pull/44061/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44061",
     "labels": [],
     "merged": true,
-    "number": 44040,
-    "review_comments_count": 14,
+    "number": 44061,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Fix gemma3n `get_audio_features`",
-    "updated_at": "2026-02-19T12:50:00Z"
+    "title": "Fix: flaky `Kosmos2ModelTest` test",
+    "updated_at": "2026-02-18T14:23:30Z"
   },
   {
-    "additions": 47,
-    "author": "itzyesse99-lgtm",
-    "author_association": "NONE",
-    "body_excerpt": "```diff diff --git a/transformers/modeling_utils.py b/transformers/modeling_utils.py index 1234567..8901234 100644 --- a/transformers/modeling_utils.py +++ b/transformers/modeling_utils.py @@ -10,6 +10,7 @@ from transformers import PreTrai\u2026",
+    "additions": 44,
+    "author": "lakprigan",
+    "author_association": "FIRST_TIMER",
+    "body_excerpt": "## Summary Migrates GPT2 to the standardized output collection interface as part of #43979. - Added `_can_record_outputs` to `GPT2PreTrainedModel` (including `cross_attentions` via `OutputRecorder` targeting the `crossattention` submodule)\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44059",
+    "created_at": "2026-02-16T20:14:30Z",
+    "deletions": 133,
+    "draft": false,
+    "files_url": "https://github.com/huggingface/transformers/pull/44059/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44059",
+    "labels": [],
+    "merged": false,
+    "number": 44059,
+    "review_comments_count": 0,
+    "state": "open",
+    "title": "[GPT2] Refactor output tracing to use capture_outputs/can_return_tuple decorators",
+    "updated_at": "2026-02-25T17:47:45Z"
+  },
+  {
+    "additions": 122,
+    "author": "engmohamedsalah",
+    "author_association": "NONE",
+    "body_excerpt": "## Summary Fixes #44052 \u2014 resolves 10 of 11 skipped tests for the `glm_moe_dsa` model. **Root causes fixed:** - **DSA indexer mask shape mismatch**: The attention mask was not properly normalized to 4D before being passed to the indexer an\u2026",
+    "changed_files": 3,
+    "cluster_id": null,
+    "cluster_ids": [],
+    "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44039",
-    "created_at": "2026-02-16T12:01:26Z",
-    "deletions": 0,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44058",
+    "created_at": "2026-02-16T19:24:30Z",
+    "deletions": 84,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44039/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44039",
+    "files_url": "https://github.com/huggingface/transformers/pull/44058/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44058",
     "labels": [
       "Code agent slop"
     ],
     "merged": false,
-    "number": 44039,
+    "number": 44058,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "AI Fix for #43979",
-    "updated_at": "2026-03-14T12:34:32Z"
+    "title": "Fix skipped tests for glm_moe_dsa model",
+    "updated_at": "2026-02-17T17:23:03Z"
   },
   {
-    "additions": 23,
-    "author": "Cyrilvallez",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? As per the title. Supersedes https://github.com/huggingface/transformers/pull/43913",
-    "changed_files": 3,
+    "additions": 0,
+    "author": "mariam851",
+    "author_association": "CONTRIBUTOR",
+    "body_excerpt": "This PR addresses memory efficiency issues in the Qwen2Moe implementation (reported in #43856). Users experienced Out-of-Memory (OOM) errors during quantization and inference, particularly with large reserved memory (e.g., 27GB on H100) th\u2026",
+    "changed_files": 0,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44037",
-    "created_at": "2026-02-16T11:02:12Z",
-    "deletions": 10,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44057",
+    "created_at": "2026-02-16T18:35:01Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44037/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44037",
+    "files_url": "https://github.com/huggingface/transformers/pull/44057/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44057",
     "labels": [],
-    "merged": true,
-    "number": 44037,
-    "review_comments_count": 3,
+    "merged": false,
+    "number": 44057,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "Add a dim check mechanism in Transpose and fix qwen3_vl_moe weight mapping",
-    "updated_at": "2026-02-16T16:01:12Z"
+    "title": "fix(qwen3_moe): optimize memory and fix OOM in MoE layers",
+    "updated_at": "2026-02-16T21:47:41Z"
   },
   {
-    "additions": 0,
-    "author": "ydshieh",
-    "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? So the following logic added in a previous PR #44033 could take effect ```python # `include_all` is `True` when the CI is running on a pull request, so it treats all failing tests # in the current CI run as \"new fai\u2026",
+    "additions": 50,
+    "author": "aman-coder03",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "## What does this PR do? This PR migrates MPNet to the new standardized output tracing system using the `@capture_outputs` decorator. Specifically, this PR: - Applies `@capture_outputs` to `MPNetModel.forward` - Removes manual accumulation\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44036",
-    "created_at": "2026-02-16T10:14:54Z",
-    "deletions": 12,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44056",
+    "created_at": "2026-02-16T18:27:05Z",
+    "deletions": 14,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44036/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44036",
+    "files_url": "https://github.com/huggingface/transformers/pull/44056/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44056",
     "labels": [],
-    "merged": true,
-    "number": 44036,
+    "merged": false,
+    "number": 44056,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Remove `other_workflow_run_ids` for `issue_comment` in `utils/notification_service.py`",
-    "updated_at": "2026-02-16T10:24:07Z"
+    "state": "open",
+    "title": "[MPNet] Refactor output tracing using capture_outputs decorator",
+    "updated_at": "2026-02-17T11:23:12Z"
   },
   {
-    "additions": 25,
-    "author": "ArthurZucker",
+    "additions": 5,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? We already brought it back with: ```python if clean_up_tokenization_spaces: # Call custom cleanup method if it exists (e.g., for CLVP's [SPACE] token replacement) if hasattr(self, \"clean_up_tokenization\") and callab\u2026",
-    "changed_files": 7,
+    "body_excerpt": "# What does this PR do?",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44035",
-    "created_at": "2026-02-16T09:49:28Z",
-    "deletions": 112,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44055",
+    "created_at": "2026-02-16T18:26:43Z",
+    "deletions": 3,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44035/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44035",
+    "files_url": "https://github.com/huggingface/transformers/pull/44055/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44055",
     "labels": [],
     "merged": true,
-    "number": 44035,
+    "number": 44055,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "bring back our demons: clean_up_tokenization_spaces",
-    "updated_at": "2026-02-20T14:50:18Z"
+    "title": "Fix unprotected torch import",
+    "updated_at": "2026-02-16T18:43:01Z"
   },
   {
-    "additions": 18,
-    "author": "ydshieh",
+    "additions": 346,
+    "author": "ArthurZucker",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 2,
+    "body_excerpt": "# What does this PR do? Add flash MLA interface. - It does not work I get a segfault - we don't leverage the paged cache so it's not as efficient as that I reckon. ```bash Fetching 6 files: 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2026",
+    "changed_files": 10,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 8,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44034",
-    "created_at": "2026-02-16T08:04:20Z",
-    "deletions": 0,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44034/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44034",
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44054",
+    "created_at": "2026-02-16T18:07:14Z",
+    "deletions": 93,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44054/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44054",
     "labels": [],
     "merged": false,
-    "number": 44034,
+    "number": 44054,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "don't merge check workflow",
-    "updated_at": "2026-02-16T10:52:50Z"
+    "state": "open",
+    "title": "Flash mla interface",
+    "updated_at": "2026-02-20T11:14:39Z"
   },
   {
-    "additions": 143,
-    "author": "ydshieh",
+    "additions": 2,
+    "author": "Cyrilvallez",
     "author_association": "MEMBER",
-    "body_excerpt": "# What does this PR do? Improve new failing test analysis for PR comment CI",
-    "changed_files": 3,
+    "body_excerpt": "# What does this PR do?",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44033",
-    "created_at": "2026-02-16T07:30:33Z",
-    "deletions": 49,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44053",
+    "created_at": "2026-02-16T17:59:48Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44033/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44033",
+    "files_url": "https://github.com/huggingface/transformers/pull/44053/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44053",
     "labels": [],
     "merged": true,
-    "number": 44033,
+    "number": 44053,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Improve new failing test analysis for PR comment CI",
-    "updated_at": "2026-02-16T08:02:16Z"
+    "title": "Fix peft conversion typo",
+    "updated_at": "2026-02-17T11:12:19Z"
   },
   {
-    "additions": 3,
-    "author": "JJJYmmm",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? Update FP8 expert replacement to use `model.config.text_config` when available (VLMs), falling back to model.config if it's text-only models.",
-    "changed_files": 1,
+    "additions": 2,
+    "author": "tomaarsen",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? * Fix incorrect forward type hint for Gemma3n ## Details The type hint didn't match the actual returned class: https://github.com/huggingface/transformers/blob/349e00c1a367ce263624e525038250625dcf20c7/src/transforme\u2026",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44032",
-    "created_at": "2026-02-16T06:02:28Z",
-    "deletions": 1,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44051",
+    "created_at": "2026-02-16T17:26:24Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44032/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44032",
+    "files_url": "https://github.com/huggingface/transformers/pull/44051/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44051",
     "labels": [],
     "merged": true,
-    "number": 44032,
-    "review_comments_count": 2,
+    "number": 44051,
+    "review_comments_count": 0,
     "state": "closed",
-    "title": "[Misc][vlms] Use text_config when initializing the fine-grained FP8Expert",
-    "updated_at": "2026-02-19T10:28:31Z"
+    "title": "[`chore`] Fix incorrect forward type hint for Gemma3n",
+    "updated_at": "2026-02-20T09:08:07Z"
   },
   {
-    "additions": 11,
-    "author": "omkar-334",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the`dpr` model as per #43979 cc @molbap <img width=\"853\" height=\"323\" alt=\"Screenshot 2026-02-16 at 9 13 30 AM\" src=\"https://github.com/user-attachments/assets/d658f1d0-75e8-4eac-8a12-9aeddf194dde\" />",
-    "changed_files": 1,
+    "additions": 15,
+    "author": "tomaarsen",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? * Patch `get_text_features` for ChineseCLIP ### Details The `get_text_features` assumes that the `text_model` returns a `BaseModelOutputWithPooling`, just like is done with many other models. Currently, the `get_tex\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44030",
-    "created_at": "2026-02-16T03:44:19Z",
-    "deletions": 58,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44050",
+    "created_at": "2026-02-16T17:23:31Z",
+    "deletions": 19,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44030/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44030",
+    "files_url": "https://github.com/huggingface/transformers/pull/44050/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44050",
     "labels": [],
     "merged": false,
-    "number": 44030,
-    "review_comments_count": 1,
-    "state": "open",
-    "title": "refactor output tracing in `dpr`",
-    "updated_at": "2026-02-17T07:46:00Z"
+    "number": 44050,
+    "review_comments_count": 4,
+    "state": "closed",
+    "title": "[`fix`] Patch `get_text_features` for ChineseCLIP",
+    "updated_at": "2026-02-17T09:55:17Z"
   },
   {
-    "additions": 21,
-    "author": "omkar-334",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the`rwkv` model as per #43979 cc @molbap <img width=\"856\" height=\"333\" alt=\"Screenshot 2026-02-16 at 9 06 34 AM\" src=\"https://github.com/user-attachments/assets/9c8c5d41-ffbd-45f6-8b9b-1429bcb14543\" />",
+    "additions": 59,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Refactors the `fnet` model to use the new `@capture_outputs` and `@can_return_tuple` decorators, as part of #43979. ### Changes: - Added `_can_record_outputs = {\"hidden_states\": FNetLayer}` to `FNetPreTrainedModel`\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44029",
-    "created_at": "2026-02-16T03:37:13Z",
-    "deletions": 55,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44049",
+    "created_at": "2026-02-16T17:19:04Z",
+    "deletions": 112,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44029/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44029",
+    "files_url": "https://github.com/huggingface/transformers/pull/44049/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44049",
     "labels": [],
     "merged": false,
-    "number": 44029,
+    "number": 44049,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "refactor output tracing in `rwkv`",
-    "updated_at": "2026-02-17T07:47:02Z"
+    "state": "closed",
+    "title": "Refactor fnet model to use @capture_outputs and @can_return_tuple decorators",
+    "updated_at": "2026-03-03T00:30:13Z"
   },
   {
-    "additions": 13,
-    "author": "omkar-334",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the`superpoint` model as per #43979 cc @molbap <img width=\"857\" height=\"334\" alt=\"Screenshot 2026-02-16 at 8 53 43 AM\" src=\"https://github.com/user-attachments/assets/17781b76-743b-4b38-923a-8db3b94ccd01\" />",
+    "additions": 4,
+    "author": "tomaarsen",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? * Fix up `__repr__` whitespace/brackets ## Reproducer ```python from transformers import AutoTokenizer, PreTrainedTokenizerBase # __repr__ via PreTrainedTokenizerBase tokenizer = AutoTokenizer.from_pretrained(\"bert-\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44028",
-    "created_at": "2026-02-16T03:25:14Z",
-    "deletions": 46,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44048",
+    "created_at": "2026-02-16T17:18:10Z",
+    "deletions": 2,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44028/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44028",
+    "files_url": "https://github.com/huggingface/transformers/pull/44048/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44048",
     "labels": [],
-    "merged": false,
-    "number": 44028,
+    "merged": true,
+    "number": 44048,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "refactor output tracing for `superpoint`",
-    "updated_at": "2026-02-17T07:46:06Z"
+    "state": "closed",
+    "title": "[`simple`] Fix up `__repr__` whitespace/brackets",
+    "updated_at": "2026-02-20T10:03:34Z"
   },
   {
-    "additions": 6,
-    "author": "omkar-334",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the `speech_encoder_decoder` model as per #43979 cc @molbap <img width=\"852\" height=\"335\" alt=\"Screenshot 2026-02-16 at 8 44 05 AM\" src=\"https://github.com/user-attachments/assets/ee25c72b-b995-403c-b47b-3e9cbae0d2cc\" />",
+    "additions": 35,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Refactors the `bloom` model to use the new `@capture_outputs` and `@can_return_tuple` decorators, as part of the effort in #43979. ### Changes: - Add `_can_record_outputs` dict to `BloomPreTrainedModel` mapping `hi\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44027",
-    "created_at": "2026-02-16T03:14:41Z",
-    "deletions": 22,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44047",
+    "created_at": "2026-02-16T17:15:25Z",
+    "deletions": 104,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44027/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44027",
+    "files_url": "https://github.com/huggingface/transformers/pull/44047/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44047",
     "labels": [],
     "merged": false,
-    "number": 44027,
+    "number": 44047,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "refactor output tracing in `speech_encoder_decoder`",
-    "updated_at": "2026-02-17T09:04:35Z"
+    "state": "closed",
+    "title": "Refactor bloom model to use @capture_outputs and @can_return_tuple decorators",
+    "updated_at": "2026-03-03T00:30:14Z"
   },
   {
-    "additions": 12,
-    "author": "omkar-334",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the`vision_encoder_decoder` model as per #43979 cc @molbap <img width=\"849\" height=\"333\" alt=\"Screenshot 2026-02-16 at 8 28 20 AM\" src=\"https://github.com/user-attachments/assets/9f511a17-947b-46ed-82a8-8bb9bb103f15\" />",
+    "additions": 24,
+    "author": "ManasVardhan",
+    "author_association": "NONE",
+    "body_excerpt": "## What does this PR do? Refactors the `codegen` model to use the `@capture_outputs` and `@can_return_tuple` decorators, replacing manual `output_attentions`/`output_hidden_states`/`return_dict` handling. ### Changes: - Add `_can_record_ou\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44026",
-    "created_at": "2026-02-16T02:59:14Z",
-    "deletions": 22,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44046",
+    "created_at": "2026-02-16T17:07:38Z",
+    "deletions": 70,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44026/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44026",
+    "files_url": "https://github.com/huggingface/transformers/pull/44046/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44046",
     "labels": [],
     "merged": false,
-    "number": 44026,
+    "number": 44046,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "refactor output tracing for `vision_encoder_decoder`",
-    "updated_at": "2026-02-17T09:05:22Z"
+    "state": "closed",
+    "title": "Refactor codegen model to use @capture_outputs and @can_return_tuple decorators",
+    "updated_at": "2026-02-17T14:15:23Z"
   },
   {
-    "additions": 7,
-    "author": "omkar-334",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the `depth_anything` model as per #43979 cc @molbap <img width=\"840\" height=\"330\" alt=\"Screenshot 2026-02-16 at 8 25 01 AM\" src=\"https://github.com/user-attachments/assets/fe7770be-70cb-4343-accb-7407c6bbb4f8\" />",
-    "changed_files": 2,
+    "additions": 456215,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 4939,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44025",
-    "created_at": "2026-02-16T02:56:17Z",
-    "deletions": 23,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44025/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44025",
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44045",
+    "created_at": "2026-02-16T17:01:41Z",
+    "deletions": 591028,
+    "draft": true,
+    "files_url": "https://github.com/huggingface/transformers/pull/44045/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44045",
     "labels": [],
     "merged": false,
-    "number": 44025,
+    "number": 44045,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "refactor output tracing for `depth_anything`",
-    "updated_at": "2026-02-17T07:46:31Z"
+    "state": "closed",
+    "title": "Flash-mla-interface",
+    "updated_at": "2026-02-16T17:11:51Z"
   },
   {
-    "additions": 15,
-    "author": "mmahjoub5",
+    "additions": 49,
+    "author": "rwtarpit",
     "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR refactors the FocalNet implementation to adopt the new `@capture_outputs` and `@can_return_tuple` decorators for standardized output collection, as part of https://github.com/huggingface/transformers/issues/\u2026",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
     "changed_files": 1,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
+    "cluster_id": "cluster-43979-11",
+    "cluster_ids": [
+      "cluster-43979-11"
+    ],
+    "cluster_role": "member",
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44024",
-    "created_at": "2026-02-15T23:48:12Z",
-    "deletions": 60,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44044",
+    "created_at": "2026-02-16T16:43:19Z",
+    "deletions": 112,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44024/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44024",
+    "files_url": "https://github.com/huggingface/transformers/pull/44044/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44044",
     "labels": [],
     "merged": false,
-    "number": 44024,
+    "number": 44044,
     "review_comments_count": 0,
     "state": "open",
-    "title": "Focalnet standardized outputs",
-    "updated_at": "2026-02-17T08:47:48Z"
+    "title": "Refactor DeBERTa's output tracing interface",
+    "updated_at": "2026-02-16T18:57:29Z"
   },
   {
-    "additions": 32,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the Nystromformer model to use the new `@capture_outputs` and `@can_return_tuple` decorators for standardized output collection, as part of #43979. ### Changes: - Add `_can_record_outputs` on `Nystromform\u2026",
-    "changed_files": 1,
+    "additions": 170,
+    "author": "IlyasMoutawwakil",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
+    "changed_files": 31,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44023",
-    "created_at": "2026-02-15T21:53:48Z",
-    "deletions": 122,
+    "comments_count": 2,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44043",
+    "created_at": "2026-02-16T16:23:57Z",
+    "deletions": 162,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44023/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44023",
+    "files_url": "https://github.com/huggingface/transformers/pull/44043/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44043",
     "labels": [],
-    "merged": false,
-    "number": 44023,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44043,
+    "review_comments_count": 15,
     "state": "closed",
-    "title": "Refactor Nystromformer output tracing with @capture_outputs and @can_return_tuple decorators",
-    "updated_at": "2026-03-03T00:30:15Z"
+    "title": "`grouped_mm` fallback",
+    "updated_at": "2026-02-23T13:58:09Z"
   },
   {
-    "additions": 57,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the ConvBERT model to use the new `@capture_outputs` and `@can_return_tuple` decorators for standardized output collection, as part of #43979. ### Changes: - Move `ConvBertPreTrainedModel` after layer def\u2026",
+    "additions": 1,
+    "author": "Rocketknight1",
+    "author_association": "MEMBER",
+    "body_excerpt": "Joao is regrettably no longer with us :saluting_face: so we should really stop getting users to ping him! This PR makes @cyrilvallez responsible for `generate` issues outside of VLMs.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44022",
-    "created_at": "2026-02-15T21:49:57Z",
-    "deletions": 152,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44042",
+    "created_at": "2026-02-16T16:00:36Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44022/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44022",
+    "files_url": "https://github.com/huggingface/transformers/pull/44042/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44042",
     "labels": [],
-    "merged": false,
-    "number": 44022,
+    "merged": true,
+    "number": 44042,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor ConvBERT output tracing with @capture_outputs and @can_return_tuple decorators",
-    "updated_at": "2026-03-03T00:30:17Z"
+    "title": "Update assignee for generate in bug report template",
+    "updated_at": "2026-02-16T16:09:19Z"
   },
   {
-    "additions": 22,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Fixes #43906 (related to #38071) ### Problem When using `pipeline('text-generation')` with batched inference on Qwen3 (and other models where `pad_token_id == bos_token_id`), a spurious warning is emitted: > A deco\u2026",
-    "changed_files": 3,
+    "additions": 469,
+    "author": "winglian",
+    "author_association": "COLLABORATOR",
+    "body_excerpt": "# What does this PR do? Alternate PR to #43985 to be a reorder only PR. <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, s\u2026",
+    "changed_files": 4,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44021",
-    "created_at": "2026-02-15T21:45:58Z",
-    "deletions": 13,
+    "comments_count": 9,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44041",
+    "created_at": "2026-02-16T15:40:41Z",
+    "deletions": 457,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44021/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44021",
+    "files_url": "https://github.com/huggingface/transformers/pull/44041/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44041",
     "labels": [],
     "merged": true,
-    "number": 44021,
-    "review_comments_count": 0,
+    "number": 44041,
+    "review_comments_count": 14,
     "state": "closed",
-    "title": "Fix false positive right-padding warning for decoder-only models in pipeline",
-    "updated_at": "2026-02-17T10:41:32Z"
+    "title": "refactor _inner_training_loop to smaller methods",
+    "updated_at": "2026-02-23T16:52:09Z"
   },
   {
-    "additions": 28,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## Summary Refactors the DeBERTa model to use the standardized output collection interface with `@capture_outputs` and `@can_return_tuple` decorators. ### Changes - Add `_can_record_outputs` on `DebertaPreTrainedModel` mapping `hidden_stat\u2026",
-    "changed_files": 1,
+    "additions": 366,
+    "author": "zucchini-nlp",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Fixes https://github.com/huggingface/transformers/issues/44008 and re-enables tests",
+    "changed_files": 5,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44020",
-    "created_at": "2026-02-15T21:39:17Z",
-    "deletions": 129,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44040",
+    "created_at": "2026-02-16T12:43:28Z",
+    "deletions": 230,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44020/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44020",
+    "files_url": "https://github.com/huggingface/transformers/pull/44040/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44040",
     "labels": [],
-    "merged": false,
-    "number": 44020,
-    "review_comments_count": 0,
+    "merged": true,
+    "number": 44040,
+    "review_comments_count": 14,
     "state": "closed",
-    "title": "Refactor DeBERTa output tracing with @capture_outputs and @can_return_tuple",
-    "updated_at": "2026-02-17T14:15:21Z"
+    "title": "Fix gemma3n `get_audio_features`",
+    "updated_at": "2026-02-19T12:50:00Z"
   },
   {
-    "additions": 17,
-    "author": "Sid-V5",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "Refactored the `resnet` model to use the standardized output tracing decorators (`@capture_outputs` and `@can_return_tuple`) as part of the migration ### Changes | File | Change | |------|--------| | `modeling_resnet.py` | Migrated to `@ca\u2026",
+    "additions": 47,
+    "author": "itzyesse99-lgtm",
+    "author_association": "NONE",
+    "body_excerpt": "```diff diff --git a/transformers/modeling_utils.py b/transformers/modeling_utils.py index 1234567..8901234 100644 --- a/transformers/modeling_utils.py +++ b/transformers/modeling_utils.py @@ -10,6 +10,7 @@ from transformers import PreTrai\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44019",
-    "created_at": "2026-02-15T19:53:19Z",
-    "deletions": 62,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44039",
+    "created_at": "2026-02-16T12:01:26Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44019/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44019",
-    "labels": [],
-    "merged": false,
-    "number": 44019,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Refactor `resnet` to use `@capture_outputs` / `@can_return_tuple` output tracing",
-    "updated_at": "2026-02-15T20:01:23Z"
-  },
-  {
-    "additions": 41,
-    "author": "yashbora9",
-    "author_association": "FIRST_TIMER",
-    "body_excerpt": "## Summary - Migrates `gpt_neo` to the standardized output collection interface as part of #43979 - Adds `@capture_outputs` decorator on `GPTNeoModel.forward` (base model) - Adds `@can_return_tuple` decorator on all wrapper model forwards\u2026",
-    "changed_files": 2,
-    "cluster_id": "cluster-44018-2",
-    "cluster_ids": [
-      "cluster-44018-2"
+    "files_url": "https://github.com/huggingface/transformers/pull/44039/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44039",
+    "labels": [
+      "Code agent slop"
     ],
-    "cluster_role": "member",
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44018",
-    "created_at": "2026-02-15T19:35:06Z",
-    "deletions": 109,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44018/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44018",
-    "labels": [],
     "merged": false,
-    "number": 44018,
+    "number": 44039,
     "review_comments_count": 0,
-    "state": "open",
-    "title": "Refactor GPT-Neo output tracing to use capture_outputs/can_return_tuple",
-    "updated_at": "2026-02-16T20:33:37Z"
+    "state": "closed",
+    "title": "AI Fix for #43979",
+    "updated_at": "2026-03-14T12:34:32Z"
   },
   {
-    "additions": 13,
-    "author": "nexiouscaliver",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "This PR refactors \\`SegformersPreTrainedModel\\` and \\`SegformersForImageClassification\\` to use standardized \\`@capture_outputs\\` and \\`@can_return_tuple\\` decorators for automatic output collection. ### Changes 1. **Imported \\`@capture_ou\u2026",
-    "changed_files": 1,
+    "additions": 23,
+    "author": "Cyrilvallez",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? As per the title. Supersedes https://github.com/huggingface/transformers/pull/43913",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44017",
-    "created_at": "2026-02-15T19:27:22Z",
-    "deletions": 6,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44037",
+    "created_at": "2026-02-16T11:02:12Z",
+    "deletions": 10,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44017/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44017",
+    "files_url": "https://github.com/huggingface/transformers/pull/44037/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44037",
     "labels": [],
-    "merged": false,
-    "number": 44017,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Refactor output tracing in segformers (#43979)",
-    "updated_at": "2026-02-20T16:51:42Z"
+    "merged": true,
+    "number": 44037,
+    "review_comments_count": 3,
+    "state": "closed",
+    "title": "Add a dim check mechanism in Transpose and fix qwen3_vl_moe weight mapping",
+    "updated_at": "2026-02-16T16:01:12Z"
   },
   {
-    "additions": 95,
-    "author": "akashadsare",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "This PR migrates GPT-2 and its derivatives (GPTBigCode, Decision Transformer) to the new standardized output collection interface using the [@capture_outputs](vscode-file://vscode-app/usr/share/code/resources/app/out/vs/code/electron-brows\u2026",
-    "changed_files": 3,
+    "additions": 0,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? So the following logic added in a previous PR #44033 could take effect ```python # `include_all` is `True` when the CI is running on a pull request, so it treats all failing tests # in the current CI run as \"new fai\u2026",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44015",
-    "created_at": "2026-02-15T18:07:11Z",
-    "deletions": 231,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44036",
+    "created_at": "2026-02-16T10:14:54Z",
+    "deletions": 12,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44015/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44015",
+    "files_url": "https://github.com/huggingface/transformers/pull/44036/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44036",
     "labels": [],
-    "merged": false,
-    "number": 44015,
-    "review_comments_count": 4,
-    "state": "open",
-    "title": "Refactor GPT2-based models to standardized output collection interface",
-    "updated_at": "2026-02-15T18:13:56Z"
+    "merged": true,
+    "number": 44036,
+    "review_comments_count": 0,
+    "state": "closed",
+    "title": "Remove `other_workflow_run_ids` for `issue_comment` in `utils/notification_service.py`",
+    "updated_at": "2026-02-16T10:24:07Z"
   },
   {
-    "additions": 45,
-    "author": "weiguangli-io",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "## Summary Fixes #43992 by preventing a false missing-key report for `UMT5EncoderModel` when `encoder.embed_tokens.weight` is tied to `shared.weight`. `UMT5EncoderModel` already declares tied weights, but loading checkpoints that only carr\u2026",
-    "changed_files": 2,
+    "additions": 25,
+    "author": "ArthurZucker",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? We already brought it back with: ```python if clean_up_tokenization_spaces: # Call custom cleanup method if it exists (e.g., for CLVP's [SPACE] token replacement) if hasattr(self, \"clean_up_tokenization\") and callab\u2026",
+    "changed_files": 7,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44014",
-    "created_at": "2026-02-15T15:17:22Z",
-    "deletions": 0,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44035",
+    "created_at": "2026-02-16T09:49:28Z",
+    "deletions": 112,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44014/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44014",
+    "files_url": "https://github.com/huggingface/transformers/pull/44035/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44035",
     "labels": [],
-    "merged": false,
-    "number": 44014,
+    "merged": true,
+    "number": 44035,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "[UMT5] Ignore tied encoder embedding missing-key warning",
-    "updated_at": "2026-02-16T13:40:21Z"
+    "title": "bring back our demons: clean_up_tokenization_spaces",
+    "updated_at": "2026-02-20T14:50:18Z"
   },
   {
-    "additions": 10,
-    "author": "gabrielfruet",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "additions": 18,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
     "body_excerpt": "# What does this PR do? <!-- Congratulations! You've made it this far! You're not quite done yet though. Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflec\u2026",
-    "changed_files": 1,
-    "cluster_id": "cluster-43979-11",
-    "cluster_ids": [
-      "cluster-43979-11"
-    ],
-    "cluster_role": "member",
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44013",
-    "created_at": "2026-02-15T13:49:53Z",
-    "deletions": 43,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44013/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44013",
-    "labels": [],
-    "merged": false,
-    "number": 44013,
-    "review_comments_count": 0,
-    "state": "open",
-    "title": "Ouptut tracing: Standardizing MobileNetv2",
-    "updated_at": "2026-02-15T13:50:59Z"
-  },
-  {
-    "additions": 79,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the **swinv2** model to use hook-based output capturing via `@capture_outputs` and `@can_return_tuple` decorators, removing the manual threading of `output_attentions`/`output_hidden_states`/`return_dict`\u2026",
     "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44012",
-    "created_at": "2026-02-15T11:20:17Z",
-    "deletions": 159,
+    "comments_count": 8,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44034",
+    "created_at": "2026-02-16T08:04:20Z",
+    "deletions": 0,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44012/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44012",
+    "files_url": "https://github.com/huggingface/transformers/pull/44034/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44034",
     "labels": [],
     "merged": false,
-    "number": 44012,
+    "number": 44034,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor output tracing for swinv2 model",
-    "updated_at": "2026-02-17T14:15:19Z"
+    "title": "don't merge check workflow",
+    "updated_at": "2026-02-16T10:52:50Z"
   },
   {
-    "additions": 79,
-    "author": "ManasVardhan",
-    "author_association": "NONE",
-    "body_excerpt": "## What does this PR do? Refactors the `swin` model to use the standardized output collection interface (`@capture_outputs` and `@can_return_tuple` decorators), as described in #43979. ### Changes **SwinPreTrainedModel:** - Added `_can_rec\u2026",
-    "changed_files": 2,
+    "additions": 143,
+    "author": "ydshieh",
+    "author_association": "MEMBER",
+    "body_excerpt": "# What does this PR do? Improve new failing test analysis for PR comment CI",
+    "changed_files": 3,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44011",
-    "created_at": "2026-02-15T11:11:02Z",
-    "deletions": 146,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44033",
+    "created_at": "2026-02-16T07:30:33Z",
+    "deletions": 49,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44011/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44011",
+    "files_url": "https://github.com/huggingface/transformers/pull/44033/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44033",
     "labels": [],
-    "merged": false,
-    "number": 44011,
+    "merged": true,
+    "number": 44033,
     "review_comments_count": 0,
     "state": "closed",
-    "title": "Refactor Swin output tracing with @capture_outputs and @can_return_tuple",
-    "updated_at": "2026-02-17T14:15:17Z"
+    "title": "Improve new failing test analysis for PR comment CI",
+    "updated_at": "2026-02-16T08:02:16Z"
   },
   {
-    "additions": 41,
-    "author": "preetam1407",
+    "additions": 3,
+    "author": "JJJYmmm",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "#43979. Refactors SqueezeBert to the standardized output collection interface: - Adds `_can_record_outputs` in `SqueezeBertPreTrainedModel` - Adds `@capture_outputs` on `SqueezeBertModel.forward` - Adds `@can_return_tuple` on task model fo\u2026",
+    "body_excerpt": "# What does this PR do? Update FP8 expert replacement to use `model.config.text_config` when available (VLMs), falling back to model.config if it's text-only models.",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44010",
-    "created_at": "2026-02-15T09:40:09Z",
-    "deletions": 139,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44032",
+    "created_at": "2026-02-16T06:02:28Z",
+    "deletions": 1,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44010/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44010",
+    "files_url": "https://github.com/huggingface/transformers/pull/44032/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44032",
     "labels": [],
-    "merged": false,
-    "number": 44010,
+    "merged": true,
+    "number": 44032,
     "review_comments_count": 2,
-    "state": "open",
-    "title": "[SqueezeBert] Migrate to standardized output collection decorators",
-    "updated_at": "2026-04-13T08:20:09Z"
+    "state": "closed",
+    "title": "[Misc][vlms] Use text_config when initializing the fine-grained FP8Expert",
+    "updated_at": "2026-02-19T10:28:31Z"
   },
   {
-    "additions": 1,
-    "author": "mariam851",
+    "additions": 11,
+    "author": "omkar-334",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "Fixes #43976 Updated the documentation to reflect the actual Python requirement (3.10+) as defined in setup.py. Changes: Updated README.md .",
+    "body_excerpt": "This PR refactors the`dpr` model as per #43979 cc @molbap <img width=\"853\" height=\"323\" alt=\"Screenshot 2026-02-16 at 9 13 30 AM\" src=\"https://github.com/user-attachments/assets/d658f1d0-75e8-4eac-8a12-9aeddf194dde\" />",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44009",
-    "created_at": "2026-02-15T08:51:26Z",
-    "deletions": 1,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44009/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44009",
-    "labels": [],
-    "merged": true,
-    "number": 44009,
-    "review_comments_count": 0,
-    "state": "closed",
-    "title": "update python requirement to 3.10+ to match codebase",
-    "updated_at": "2026-02-16T13:46:56Z"
-  },
-  {
-    "additions": 26,
-    "author": "pdwi2020",
-    "author_association": "FIRST_TIME_CONTRIBUTOR",
-    "body_excerpt": "## Summary - refactor `ResNetModel` to use `@capture_outputs` for hidden-state collection - register `_can_record_outputs` on `ResNetPreTrainedModel` with `ResNetStage` - switch `ResNetForImageClassification` and `ResNetBackbone` to `@can_\u2026",
-    "changed_files": 3,
-    "cluster_id": "cluster-43979-11",
-    "cluster_ids": [
-      "cluster-43979-11"
-    ],
-    "cluster_role": "canonical",
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44007",
-    "created_at": "2026-02-15T07:26:52Z",
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44030",
+    "created_at": "2026-02-16T03:44:19Z",
     "deletions": 58,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44007/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44007",
+    "files_url": "https://github.com/huggingface/transformers/pull/44030/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44030",
     "labels": [],
     "merged": false,
-    "number": 44007,
-    "review_comments_count": 0,
+    "number": 44030,
+    "review_comments_count": 1,
     "state": "open",
-    "title": "[ResNet] Refactor output tracing to decorator-based interface",
-    "updated_at": "2026-02-19T15:49:49Z"
+    "title": "refactor output tracing in `dpr`",
+    "updated_at": "2026-02-17T07:46:00Z"
   },
   {
-    "additions": 8,
-    "author": "cyyever",
+    "additions": 21,
+    "author": "omkar-334",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR uses torch.xlogy for better numerical handling.",
-    "changed_files": 8,
+    "body_excerpt": "This PR refactors the`rwkv` model as per #43979 cc @molbap <img width=\"856\" height=\"333\" alt=\"Screenshot 2026-02-16 at 9 06 34 AM\" src=\"https://github.com/user-attachments/assets/9c8c5d41-ffbd-45f6-8b9b-1429bcb14543\" />",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44006",
-    "created_at": "2026-02-15T04:07:50Z",
-    "deletions": 8,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44029",
+    "created_at": "2026-02-16T03:37:13Z",
+    "deletions": 55,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44006/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44006",
+    "files_url": "https://github.com/huggingface/transformers/pull/44029/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44029",
     "labels": [],
-    "merged": true,
-    "number": 44006,
+    "merged": false,
+    "number": 44029,
     "review_comments_count": 0,
-    "state": "closed",
-    "title": "Use torch.xlogy ",
-    "updated_at": "2026-02-17T00:42:54Z"
-  },
-  {
-    "additions": 224,
-    "author": "cyyever",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "# What does this PR do? This PR transfers grid_thw to a python list at the beginning of some functions to reduce later CUDA sync calls. Therefore, several sync calls are merged into one call.",
-    "changed_files": 16,
-    "cluster_id": null,
-    "cluster_ids": [],
-    "cluster_role": null,
-    "comments_count": 3,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44005",
-    "created_at": "2026-02-15T02:34:55Z",
-    "deletions": 254,
-    "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44005/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44005",
-    "labels": [],
-    "merged": true,
-    "number": 44005,
-    "review_comments_count": 1,
-    "state": "closed",
-    "title": "Reduce reduce CUDA sync",
-    "updated_at": "2026-02-17T01:00:52Z"
+    "state": "open",
+    "title": "refactor output tracing in `rwkv`",
+    "updated_at": "2026-02-17T07:47:02Z"
   },
   {
-    "additions": 21,
+    "additions": 13,
     "author": "omkar-334",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the `codegen` model as per #43979 cc @molbap <img width=\"843\" height=\"445\" alt=\"Screenshot 2026-02-15 at 5 24 52 AM\" src=\"https://github.com/user-attachments/assets/d5aeb711-96a7-4fd8-af7b-0aeac23eeeb1\" /> 2 tests are bei\u2026",
-    "changed_files": 1,
+    "body_excerpt": "This PR refactors the`superpoint` model as per #43979 cc @molbap <img width=\"857\" height=\"334\" alt=\"Screenshot 2026-02-16 at 8 53 43 AM\" src=\"https://github.com/user-attachments/assets/17781b76-743b-4b38-923a-8db3b94ccd01\" />",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44004",
-    "created_at": "2026-02-14T23:56:18Z",
-    "deletions": 62,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44028",
+    "created_at": "2026-02-16T03:25:14Z",
+    "deletions": 46,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44004/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44004",
+    "files_url": "https://github.com/huggingface/transformers/pull/44028/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44028",
     "labels": [],
     "merged": false,
-    "number": 44004,
+    "number": 44028,
     "review_comments_count": 0,
     "state": "open",
-    "title": "refactor output tracing for `codegen`",
-    "updated_at": "2026-02-17T08:56:07Z"
+    "title": "refactor output tracing for `superpoint`",
+    "updated_at": "2026-02-17T07:46:06Z"
   },
   {
-    "additions": 37,
+    "additions": 6,
     "author": "omkar-334",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the `mamba` model as per #43979 cc @molbap <img width=\"859\" height=\"427\" alt=\"Screenshot 2026-02-15 at 5 12 43 AM\" src=\"https://github.com/user-attachments/assets/f23bb675-a9a3-4e21-a6c5-9804910301b4\" /> Note - Only 46 te\u2026",
-    "changed_files": 2,
+    "body_excerpt": "This PR refactors the `speech_encoder_decoder` model as per #43979 cc @molbap <img width=\"852\" height=\"335\" alt=\"Screenshot 2026-02-16 at 8 44 05 AM\" src=\"https://github.com/user-attachments/assets/ee25c72b-b995-403c-b47b-3e9cbae0d2cc\" />",
+    "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44003",
-    "created_at": "2026-02-14T23:46:10Z",
-    "deletions": 68,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44027",
+    "created_at": "2026-02-16T03:14:41Z",
+    "deletions": 22,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44003/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44003",
+    "files_url": "https://github.com/huggingface/transformers/pull/44027/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44027",
     "labels": [],
     "merged": false,
-    "number": 44003,
+    "number": 44027,
     "review_comments_count": 0,
     "state": "open",
-    "title": "refactor output tracing in `mamba`",
-    "updated_at": "2026-02-17T07:40:50Z"
+    "title": "refactor output tracing in `speech_encoder_decoder`",
+    "updated_at": "2026-02-17T09:04:35Z"
   },
   {
-    "additions": 7,
+    "additions": 12,
     "author": "omkar-334",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the `upernet` model as per #43979 cc @molbap <img width=\"856\" height=\"457\" alt=\"Screenshot 2026-02-15 at 4 51 03 AM\" src=\"https://github.com/user-attachments/assets/5dc478d7-d708-4296-a86b-c3bb252d0325\" />",
+    "body_excerpt": "This PR refactors the`vision_encoder_decoder` model as per #43979 cc @molbap <img width=\"849\" height=\"333\" alt=\"Screenshot 2026-02-16 at 8 28 20 AM\" src=\"https://github.com/user-attachments/assets/9f511a17-947b-46ed-82a8-8bb9bb103f15\" />",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 2,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44002",
-    "created_at": "2026-02-14T23:21:45Z",
-    "deletions": 20,
+    "comments_count": 1,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44026",
+    "created_at": "2026-02-16T02:59:14Z",
+    "deletions": 22,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44002/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44002",
+    "files_url": "https://github.com/huggingface/transformers/pull/44026/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44026",
     "labels": [],
     "merged": false,
-    "number": 44002,
+    "number": 44026,
     "review_comments_count": 0,
     "state": "open",
-    "title": "refactor output tracing in `upernet`",
-    "updated_at": "2026-02-17T08:55:16Z"
+    "title": "refactor output tracing for `vision_encoder_decoder`",
+    "updated_at": "2026-02-17T09:05:22Z"
   },
   {
-    "additions": 3,
+    "additions": 7,
     "author": "omkar-334",
     "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the`univnet` model as per #43979 cc @molbap <img width=\"848\" height=\"462\" alt=\"Screenshot 2026-02-15 at 4 19 00 AM\" src=\"https://github.com/user-attachments/assets/75848429-b9ff-49b3-a028-645aa67fc2ad\" />",
-    "changed_files": 1,
+    "body_excerpt": "This PR refactors the `depth_anything` model as per #43979 cc @molbap <img width=\"840\" height=\"330\" alt=\"Screenshot 2026-02-16 at 8 25 01 AM\" src=\"https://github.com/user-attachments/assets/fe7770be-70cb-4343-accb-7407c6bbb4f8\" />",
+    "changed_files": 2,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
-    "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44001",
-    "created_at": "2026-02-14T22:50:39Z",
-    "deletions": 9,
+    "comments_count": 3,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44025",
+    "created_at": "2026-02-16T02:56:17Z",
+    "deletions": 23,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44001/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44001",
+    "files_url": "https://github.com/huggingface/transformers/pull/44025/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44025",
     "labels": [],
     "merged": false,
-    "number": 44001,
+    "number": 44025,
     "review_comments_count": 0,
     "state": "open",
-    "title": "refactor output tracing in `univnet`",
-    "updated_at": "2026-02-14T23:22:13Z"
+    "title": "refactor output tracing for `depth_anything`",
+    "updated_at": "2026-02-17T07:46:31Z"
   },
   {
-    "additions": 8,
-    "author": "omkar-334",
-    "author_association": "CONTRIBUTOR",
-    "body_excerpt": "This PR refactors the `vision_text_dual_encoder` model issue as per #43979 cc @molbap <img width=\"876\" height=\"292\" alt=\"Screenshot 2026-02-15 at 4 09 07 AM\" src=\"https://github.com/user-attachments/assets/11147a56-993b-4abc-b07a-ec739a53d\u2026",
+    "additions": 15,
+    "author": "mmahjoub5",
+    "author_association": "FIRST_TIME_CONTRIBUTOR",
+    "body_excerpt": "# What does this PR do? This PR refactors the FocalNet implementation to adopt the new `@capture_outputs` and `@can_return_tuple` decorators for standardized output collection, as part of https://github.com/huggingface/transformers/issues/\u2026",
     "changed_files": 1,
     "cluster_id": null,
     "cluster_ids": [],
     "cluster_role": null,
     "comments_count": 1,
-    "conversation_url": "https://github.com/huggingface/transformers/pull/44000",
-    "created_at": "2026-02-14T22:44:14Z",
-    "deletions": 21,
+    "conversation_url": "https://github.com/huggingface/transformers/pull/44024",
+    "created_at": "2026-02-15T23:48:12Z",
+    "deletions": 60,
     "draft": false,
-    "files_url": "https://github.com/huggingface/transformers/pull/44000/files",
-    "html_url": "https://github.com/huggingface/transformers/pull/44000",
+    "files_url": "https://github.com/huggingface/transformers/pull/44024/files",
+    "html_url": "https://github.com/huggingface/transformers/pull/44024",
     "labels": [],
     "merged": false,
-    "number": 44000,
+    "number": 44024,
     "review_comments_count": 0,
     "state": "open",
-    "title": "refactor output tracing in `vision_text_dual_encoder`",
-    "updated_at": "2026-02-17T07:46:33Z"
+    "title": "Focalnet standardized outputs",
+    "updated_at": "2026-02-17T08:47:48Z"
   }
 ]