| COMMIT |
1.00 |
fix(i18n): replace broken relative links to awesome-transfor |
|
Commit message contains explicit AI assi |
2026-03-23 |
| COMMIT |
1.00 |
Update AFMoE architecture to use v5-style MoE impl (#44063) |
|
Commit message contains explicit AI assi |
2026-03-19 |
| COMMIT |
1.00 |
Sdpa for owlvit (#42136) |
|
Commit message contains explicit AI assi |
2026-03-17 |
| COMMIT |
1.00 |
:rotating_light: Validate config attributes (#41250) |
|
Commit message contains explicit AI assi |
2026-03-16 |
| PR |
1.00 |
Add doc page for capturing outputs |
|
PR body explicitly mentions AI collabora |
2026-03-23 |
| PR |
0.35 |
🚨 Refactor ViT to updated standards |
|
Phrase 'This PR aims at...' slightly AI- |
2025-10-17 |
| PR |
0.20 |
Fix variable shadowing in pipeline example and typo in BART |
|
Slightly formal but includes specific ja |
2026-03-22 |
| PR |
0.20 |
Internalise the NomicBERT model |
|
Mainly technical content and references. |
2025-12-29 |
| PR |
0.18 |
Fix GIL=0 segfault and Add GIL=0 compat for regex paths |
|
Technical, includes Python version refer |
2025-10-03 |
| PR |
0.15 |
fix: set `clean_up_tokenization_spaces=False` in Llama 3 tok |
|
Technical explanation and issue referenc |
2026-03-21 |
| PR |
0.15 |
refactor: improved the cli server module code organization |
|
Standard technical language, no AI phras |
2026-03-20 |
| PR |
0.15 |
fix load_best_model_checkpoint_at_end do not load the best m |
|
Straightforward but slightly formal; mos |
2026-03-10 |
| PR |
0.15 |
fix(testing): Fix PaliGemma 2 and PaddleOCR-VL test failures |
|
Lists failures with terse, technical phr |
2026-03-16 |
| PR |
0.12 |
Fix Mllama torch.compile failure caused by new attention mas |
|
Technical discussion with detailed conte |
2026-03-19 |
| PR |
0.12 |
Fix llama4 bnb mode |
|
Domain jargon, concise explanation, no A |
2026-03-11 |
| COMMIT |
0.10 |
Add big angry code agent warnings! (#44890) |
|
Commit messages use domain language and |
2026-03-22 |
| PR |
0.10 |
refactor: mlinter as its own package |
|
Uses domain terms, bullet points, and in |
2026-03-23 |
| PR |
0.10 |
Add Music Flamingo |
|
Technical content with some detail, no c |
2026-01-27 |
| PR |
0.10 |
Correct docstrings for `from_pretrained` (url input deprecat |
|
Domain-specific change, brief and to the |
2026-03-23 |
| PR |
0.10 |
Fix: Update optimization.py |
|
Terse, typical commit/PR phrasing, no AI |
2026-03-21 |
| PR |
0.10 |
[CB] Add an option to return logprobs |
|
Concise, uses domain terms; not overly f |
2026-03-18 |
| PR |
0.10 |
fix(i18n): replace broken relative links to awesome-transfor |
|
Descriptive, technical, not overly forma |
2026-03-21 |
| PR |
0.10 |
Fix: Pass scheduler_specific_kwargs to inverse_sqrt schedule |
|
Technical fix, uses reference to issue, |
2026-03-22 |
| PR |
0.10 |
Fix backward compatibility for full path imports of Fast Ima |
|
Direct, specific to import error, natura |
2026-03-22 |
| PR |
0.10 |
fix: pop output_* flags from kwargs in capture_outputs to pr |
|
Terse style, references issue, specific |
2026-03-22 |
| PR |
0.10 |
fix(gpt-neox): preserve rotary_pct across save/load cycle |
|
Describes config details directly, no ge |
2026-03-21 |
| PR |
0.10 |
fix(deberta-v2): move "Copied from" comments above @torch.ji |
|
Technical jargon and a succinct, informa |
2026-03-21 |
| PR |
0.10 |
[Deepspeed Inference] HF Integration |
|
Informal, focused explanation of domain |
2021-11-17 |
| PR |
0.10 |
chore(typing): added rule 11 |
|
Succinct, with domain-specific shorthand |
2026-03-19 |
| PR |
0.10 |
Dynamic weight conversion is recursive |
|
Refers to specific PR and technical rati |
2026-02-26 |
| PR |
0.10 |
Remove unnecessary expand_as in get_placeholder_mask across |
|
Technical change list and terse PR summa |
2026-03-21 |
| PR |
0.10 |
fix(models): Fix Perceiver interpolate_pos_encoding interpol |
|
Direct reference to commit and bug fix, |
2026-03-20 |
| PR |
0.10 |
fix: propagate num_labels/id2label to text_config in Qwen3_5 |
|
Short and technical with specific config |
2026-03-22 |
| PR |
0.10 |
fix: prevent IndexError in Whisper word timestamp decode |
|
Describes an edge-case bug with clear, c |
2026-03-20 |
| PR |
0.10 |
fix: Whisper word timestamp OOB access on trailing replaceme |
|
Concise bug description with code refere |
2026-03-20 |
| PR |
0.10 |
Add inference time layer fusion optimisations via `PreTraine |
|
Straightforward technical title, no AI p |
2026-03-23 |
| PR |
0.10 |
Add big angry code agent warnings! |
|
Natural domain tone and some typos; refe |
2026-03-20 |
| PR |
0.10 |
fix(testing): Fix Kyutai Speech-To-Text, LLaVA-OneVision, an |
|
Uses domain-specific terms and structure |
2026-03-14 |
| PR |
0.10 |
[vllm + v5 fix] handle TokenizersBackend fallback properly f |
|
Human-like informal tone and specific me |
2026-02-24 |
| PR |
0.10 |
fix: improve processor loading performance by avoiding redun |
|
Domain-specific detail, technical focus, |
2026-03-22 |
| PR |
0.10 |
Allow `mm_token_type` be non-padded lists |
|
Highly technical, natural tone, not over |
2026-03-10 |
| PR |
0.10 |
fix(camembert): add tie_word_embeddings=True to CamembertCon |
|
Detailed domain-specific explanation; no |
2026-03-22 |
| PR |
0.10 |
Fix flash attention crash with 3D position_ids (Qwen3.5) |
|
Technical, very specific; not formal or |
2026-03-21 |
| PR |
0.10 |
fix: handle ragged batch inputs in Qwen2_5_VLProcessor mm_to |
|
Natural technical phrasing, description |
2026-03-21 |
| PR |
0.10 |
Add /v1/completions endpoint (OpenAI legacy completions API) |
|
Free-text uses concise, technical style; |
2026-03-10 |
| PR |
0.10 |
docs(pipelines): remove outdated question-answering example |
|
Terse, technical update with domain refe |
2026-03-17 |
| PR |
0.10 |
Add static FP8 expert support |
|
Domain-specific jargon, typos; informal |
2026-03-20 |
| PR |
0.10 |
Fix failing `T5ModelIntegrationTest` |
|
Direct reference to test logs; concise a |
2026-03-22 |
| PR |
0.10 |
LwDetrImageLoss: Fix dtype casting to prevent crash when usi |
|
Domain-specific terms, informal, no AI s |
2026-03-20 |
| PR |
0.10 |
model: Add DEIMv2 to Transformers |
|
Slightly more structured, but references |
2026-02-27 |
| PR |
0.10 |
Ensure final evaluation runs with step-based evaluation stra |
|
Direct technical description, uses domai |
2026-02-19 |
| PR |
0.10 |
Add qwen3 tts |
|
Specific technical additions, structure, |
2026-03-07 |
| PR |
0.10 |
fix: Add MXFP4 MoE/attention backward kernels |
|
Slightly formal but uses domain terms an |
2026-02-05 |
| PR |
0.10 |
[Misc] add enable_thinking to template kwargs |
|
Contains domain-specific abbreviations, |
2026-03-18 |
| PR |
0.10 |
Fix core dumped when `NemotronH` is torch compiled |
|
Uses real-world test output and casual l |
2026-03-19 |
| PR |
0.10 |
Fix several based models' pipeline parallel support |
|
Domain jargon, uses abbreviation, inform |
2026-03-14 |
| COMMIT |
0.05 |
Fix unexpected `position_ids` keys when loading OwlViT model |
|
Uses domain language and concise technic |
2026-03-18 |
| COMMIT |
0.05 |
feat(integration): Add KubeflowCallback to enable automatic |
|
Standard signed-off commits; technical a |
2026-03-18 |
| COMMIT |
0.05 |
Centralize AI agent templates in `.ai` (#44489) |
|
Varsity of edits, casual phrases like 't |
2026-03-18 |
| PR |
0.05 |
[docs] peft |
|
Content is technical, uses abbreviations |
2026-03-18 |
| PR |
0.05 |
Dequant fix |
|
Template content only, no meaningful fre |
2026-03-18 |
| PR |
0.05 |
Fix failing `Qwen3OmniModelIntegrationTests` |
|
Direct bugfix details and linked referen |
2026-03-19 |
| PR |
0.05 |
fix: avoid unconditional model_info call in _patch_mistral_r |
|
Terse, technical, and direct fixes; not |
2026-03-22 |
| PR |
0.05 |
Fix MobileNet v1/v2 image processor default interpolation to |
|
Domain-specific vocabulary and clear mot |
2026-01-16 |
| PR |
0.05 |
Fix `_set_model_specific_special_tokens` to accept list-form |
|
Concise, domain-specific explanation wit |
2026-03-17 |
| PR |
0.05 |
[DeepSpeed] Fix evaluate()/predict() before train() |
|
Uses domain jargon and technical breakdo |
2026-03-20 |
| PR |
0.05 |
[docs] model cards |
|
Casual style, domain-specific references |
2026-03-18 |
| PR |
0.05 |
Add backward compatibility for direct imports from legacy `i |
|
Concise technical explanation typical fo |
2026-03-20 |
| PR |
0.05 |
Switch FP8 per tensor quant to use `torch._scaled_mm` |
|
Direct, minimal prose with technical ter |
2026-03-19 |
| PR |
0.05 |
add `StaticLayer.crop()` to match `DynamicLayer` API |
|
Short, direct, and domain-specific with |
2026-03-20 |
| PR |
0.05 |
Add THD support in ESM |
|
Uses terse technical explanation, not Ch |
2026-02-19 |
| PR |
0.05 |
fix: ensure prediction_step returns tensor for logits, not t |
|
Brief and technical, fits human contribu |
2026-03-20 |
| PR |
0.05 |
refactor: unify QA calls |
|
Uses concise technical bullets typical o |
2026-03-20 |
| PR |
0.05 |
[PoC] HF exporters |
|
Casual, refers to specific PRs, and uses |
2025-11-03 |
| PR |
0.05 |
Fix how PreTrainedModel checks annotations on Python 3.14+ |
|
Explains specific Python version and PEP |
2026-02-20 |
| PR |
0.05 |
Proposal to add Qwen3-ASR support [WIP] |
|
Casual, domain-specific; uses brief stat |
2026-02-08 |
| PR |
0.05 |
fix config type |
|
Minimal, technical, with inline Python; |
2026-03-20 |
| PR |
0.05 |
[Trainer] add MoERouterHealthCallback Callback |
|
Brief, direct technical description with |
2026-03-20 |
| PR |
0.03 |
Bump kernels version dependency to avoid crashes |
|
Informal tone, uses emoji, bug-driven, n |
2026-03-20 |
| PR |
0.03 |
[refactor] Serving into proper modules |
|
Colloquial explanations and POC referenc |
2026-03-17 |
| PR |
0.02 |
[docs] continuous batching |
|
Direct update list and informal phrasing |
2026-03-20 |
| PR |
0.02 |
[`FA4`] Add kernels fallback |
|
Very terse, domain-specific, no signs of |
2026-03-17 |
| COMMIT |
0.00 |
Fix failing job `Update Transformers metadata` after #43514 |
|
Terse commit messages, minimal free-text |
2026-03-23 |
| COMMIT |
0.00 |
Clearer type hints and fix rope validation in configs (#4494 |
|
Casual phrasing, typos, domain-specific |
2026-03-23 |
| COMMIT |
0.00 |
Correct docstrings for `from_pretrained` (url input deprecat |
|
Technical, short, no ChatGPT markers, hu |
2026-03-23 |
| COMMIT |
0.00 |
Fix backward compatibility for full path imports of Fast Ima |
|
Technical changelog, informal tone, huma |
2026-03-23 |
| COMMIT |
0.00 |
chore(typing): added rule 11 (#44865) |
|
Informal commit titles, domain jargon, n |
2026-03-23 |
| COMMIT |
0.00 |
fix: improve processor loading performance by avoiding redun |
|
Structured technical changes, natural ph |
2026-03-23 |
| COMMIT |
0.00 |
fix(camembert): add tie_word_embeddings=True to CamembertCon |
|
Detailed technical context, some typos, |
2026-03-23 |
| COMMIT |
0.00 |
Support SizeDict import in get_size_dict (#44903) |
|
Short, direct, typical commit phrasing, |
2026-03-23 |
| COMMIT |
0.00 |
fix `processing_utils.py`: avoid deepcopying tokenizer in `P |
|
Concise, minimal, domain-specific, human |
2026-03-23 |
| COMMIT |
0.00 |
fix: set `clean_up_tokenization_spaces=False` in Llama 3 tok |
|
Clear technical explanation, not overly |
2026-03-23 |
| COMMIT |
0.00 |
[docs] model cards (#44837) |
|
Extremely terse and informal; signals hu |
2026-03-20 |
| COMMIT |
0.00 |
[Model] Add UVDoc Model Support (#43385) |
|
Fragmented, minimal commit message style |
2026-03-20 |
| COMMIT |
0.00 |
Add backward compatibility for direct imports from legacy `i |
|
Brief, domain-specific phrasing, no AI s |
2026-03-20 |
| COMMIT |
0.00 |
[`FA4`] Add kernels fallback (#44797) |
|
Informal, technical, and concise message |
2026-03-20 |
| COMMIT |
0.00 |
Bump kernels version dependency to avoid crashes (#44887) |
|
Very terse commit messages with co-autho |
2026-03-20 |
| COMMIT |
0.00 |
[Model] Add SLANeXt Model Support (#43707) |
|
Informal, many quick fixes, joking ('it |
2026-03-20 |
| COMMIT |
0.00 |
Fix core dumped when `NemotronH` is torch compiled (#44854) |
|
Commit messages are terse with typical h |
2026-03-20 |
| COMMIT |
0.00 |
Fix several based models' pipeline parallel support (#44699) |
|
Pragmatic one-line descriptions and stan |
2026-03-20 |
| COMMIT |
0.00 |
fix(testing): Fix PaliGemma 2 and PaddleOCR-VL test failures |
|
Concise technical message; style is typi |
2026-03-20 |
| COMMIT |
0.00 |
Fix dtype guessing from state dict (#44883) |
|
Very short, domain-specific commit title |
2026-03-20 |
| COMMIT |
0.00 |
Add missing dunder methods to `SizeDict` (#44884) |
|
Standard minimal commit summary; no AI h |
2026-03-20 |
| COMMIT |
0.00 |
Fix VL model rope_deltas batch size mismatch in online RL tr |
|
Short, technical, human-style summary an |
2026-03-20 |
| COMMIT |
0.00 |
Fix `layer_types` type hint for `AFMoE` and `Llama4` (#44874 |
|
Standard type hint update, signed by use |
2026-03-20 |
| COMMIT |
0.00 |
Align lfm2 cache to other mamba caches (#44866) |
|
Minimal, direct messages with informal c |
2026-03-20 |
| COMMIT |
0.00 |
Fix nemotron config docstrings (#44878) |
|
Terse domain description, matches human |
2026-03-20 |
| COMMIT |
0.00 |
Fix nemotron_h modular (#44876) |
|
Extremely minimal, rushed style; typical |
2026-03-20 |
| COMMIT |
0.00 |
feat: added cache to the model linter (#44790) |
|
Terse commit messages; no AI traits. |
2026-03-20 |
| COMMIT |
0.00 |
[Model] Add PP-Chart2Table Model Support (#43767) |
|
Minimal messages, human style, no AI hal |
2026-03-19 |
| COMMIT |
0.00 |
[Mistral] Fix query scaling for Mistral4 and Ministral3 (#44 |
|
Extremely brief message, typical human s |
2026-03-19 |
| COMMIT |
0.00 |
Propagate the model loading from transformers serve to chat |
|
Normal human commit structure and tone. |
2026-03-19 |
| COMMIT |
0.00 |
Update some type hints (#44851) |
|
Short, informal commit messages with hum |
2026-03-19 |
| COMMIT |
0.00 |
enable tp for benchmark (#43750) |
|
Informal tone, short commits, human-writ |
2026-03-19 |
| COMMIT |
0.00 |
Fix glm dsa (#44564) |
|
Single-word commit log, human-written. |
2026-03-19 |
| COMMIT |
0.00 |
🚨🚨 Refactor Image Processors to support different backends ( |
|
Short updates, human workflow on large P |
2026-03-19 |
| COMMIT |
0.00 |
[generate] Never use `cache_position` anymore in generation |
|
Human iterative commit process, no forma |
2026-03-19 |
| COMMIT |
0.00 |
Fix KeyError in convert_to_native_format for dict vocab (#44 |
|
Technical explanation, informal, human-w |
2026-03-19 |
| COMMIT |
0.00 |
fix: XLNet: relative_positional_encoding computes on CPU eve |
|
Concise commit messages with clear domai |
2026-03-19 |
| COMMIT |
0.00 |
Fix annotations reader for python 3.14 in `PreTrainedModel` |
|
Brief messages with specific version tar |
2026-03-19 |
| COMMIT |
0.00 |
[CB] Better parametrization for compile (#44578) |
|
Casual language, informal notes, and min |
2026-03-19 |
| COMMIT |
0.00 |
Fix `KeyError` when patching mistral regex (#43376) |
|
Succinct, technical commit logs; include |
2026-03-19 |
| COMMIT |
0.00 |
Correct code block formatting in weightconverter.md (#44839) |
|
Straightforward edit description typical |
2026-03-19 |
| COMMIT |
0.00 |
deepseek_v2, deepseek_v3, and modernbert fix for having inco |
|
Informal PR structure and terse notes su |
2026-03-18 |
| COMMIT |
0.00 |
[Model] Add PP-OCRv5_server_rec and PP-OCRv5_mobile_rec mod |
|
Sequence of brief, domain-specific commi |
2026-03-18 |
| COMMIT |
0.00 |
Add `Jina-Embeddings-V3` Model (#44251) |
|
Modular commit breakdown with terse and |
2026-03-18 |
| COMMIT |
0.00 |
feat(ci): added a network debug report (#44636) |
|
Changelog includes domain jargon and inf |
2026-03-18 |
| COMMIT |
0.00 |
Add GreedyLR adaptive learning rate scheduler (#44271) |
|
Detailed, technical changelog with human |
2026-03-18 |
| COMMIT |
0.00 |
Update more modular examples (#44834) |
|
One-word human-typical commit message an |
2026-03-18 |
| COMMIT |
0.00 |
Fix and re-run modular converter on examples (#44833) |
|
Short, informal commit messages with typ |
2026-03-18 |
| COMMIT |
0.00 |
Remove cache_position in more models (4 and last one) (#4482 |
|
Terse, informal, and non-AI phrasing lik |
2026-03-18 |
| COMMIT |
0.00 |
Fix loading issue in Sam3 (#44831) |
|
Minimal human-typical 'fix loading issue |
2026-03-18 |
| COMMIT |
0.00 |
Add GGUF support for MiniMax-M2.1 model (#44526) |
|
No free-text, only PR title; human typic |
2026-03-18 |
| COMMIT |
0.00 |
support xxxFast alias in v5 tokenizers (#44766) |
|
Domain-typical short test/dev commit mes |
2026-03-18 |
| COMMIT |
0.00 |
Remove cache_position in more models (3) (#44759) |
|
Natural, informal, and technical commit |
2026-03-18 |
| COMMIT |
0.00 |
Fix `supports_{tp/pp}_plan` (#44696) |
|
Commit uses informal, terse messages wit |
2026-03-18 |
| COMMIT |
0.00 |
[CI] Temporarily skip Mistral4 tests as they almost all fail |
|
Extremely minimal message; classic human |
2026-03-18 |
| COMMIT |
0.00 |
update flex attention to use `return_aux` instead of `return |
|
Contains typos and informal language, ty |
2026-03-18 |
| COMMIT |
0.00 |
[Gemma] Update conversion scripts for Transformers v5 Comapt |
|
Direct, domain-specific commit messages; |
2026-03-18 |
| COMMIT |
0.00 |
fix bug embedding_size mismatch with hidden_size in electra |
|
Commit message is terse, with a typical |
2026-03-18 |
| COMMIT |
0.00 |
Fix pegasus conversion (#44571) |
|
Brief, technical, and mentions force mer |
2026-03-18 |
| COMMIT |
0.00 |
Fix repo-check bot (#44812) |
|
Single-word, informal message; clearly h |
2026-03-18 |
| COMMIT |
0.00 |
[docs] is_causal feature (#44777) |
|
Extremely terse, no AI markers, human co |
2026-03-17 |
| COMMIT |
0.00 |
docs(tasks): remove references to removed question-answering |
|
Detailed but natural explanation with do |
2026-03-17 |
| COMMIT |
0.00 |
Fix configs with `@strict` (#44770) |
|
Informal, expressive language; clear sig |
2026-03-17 |
| COMMIT |
0.00 |
[AMD CI] Fix test failures across important models (#44632) |
|
Commit messages are terse, use abbreviat |
2026-03-17 |
| COMMIT |
0.00 |
Move VLM conversions to the main mapping (#44627) |
|
Short, informal commit messages and huma |
2026-03-17 |
| COMMIT |
0.00 |
Fix config loading issues (type issues) (#44789) |
|
All messages are single word 'fix' or eq |
2026-03-17 |
| COMMIT |
0.00 |
Remove `is_causal` from `EuroBertConfig` (#44774) |
|
Very brief informal message; no AI style |
2026-03-17 |
| COMMIT |
0.00 |
model-linter: Added rule 10 (#44761) |
|
Terse summary; no signs of AI tone or ph |
2026-03-17 |
| COMMIT |
0.00 |
[fix] mistral 4 docs (#44776) |
|
Single word commit; no evidence of AI st |
2026-03-16 |
| COMMIT |
0.00 |
Add Mistral 4 (#44760) |
|
Uses typical human commit structure, wit |
2026-03-16 |
| COMMIT |
0.00 |
Fix: Eurobert model was missing @strict decorator and invali |
|
Contains technical explanation with doma |
2026-03-16 |
| COMMIT |
0.00 |
fix: sig lip import (#44764) |
|
Short, practical summary; not AI-like. |
2026-03-16 |
| COMMIT |
0.00 |
Disable async loading when quantizing on the fly (#44576) |
|
Informal style and suggestions; normal h |
2026-03-16 |
| COMMIT |
0.00 |
Bump torchao >=0.15 and fix quantization CI (#44604) |
|
Concise commit messages with domain-spec |
2026-03-16 |
| COMMIT |
0.00 |
Fix tensor indexing crash in serve generate_response KV cach |
|
Technical explanation with direct style |
2026-03-16 |
| COMMIT |
0.00 |
[MistralCommonBackend] Upgrade mistral-common to v1.10.0 (#4 |
|
Standard PR format with technical conten |
2026-03-16 |
| COMMIT |
0.00 |
Fix `mlcd` auto config/model/mapping issues (#44730) |
|
Short, informal commit messages with dom |
2026-03-16 |
| COMMIT |
0.00 |
Fix bug and add XPU Expectations for qwen2 and jamba tests ( |
|
Technical content with repeated signed-o |
2026-03-16 |
| COMMIT |
0.00 |
Add model lerobot PI0 to transformers (#44160) |
|
Informal commit style, domain abbreviati |
2026-03-16 |
| COMMIT |
0.00 |
[medasr] doc update (#44633) |
|
Simple doc update with direct co-authors |
2026-03-16 |
| COMMIT |
0.00 |
Idefics3 without cache fix (#44607) |
|
Technical fixes with direct notes and ex |
2026-03-16 |
| COMMIT |
0.00 |
Add XPU Expectations for vibe voice acoustic tokenizer tests |
|
Domain-specific content, formatted and s |
2026-03-16 |
| COMMIT |
0.00 |
Fix transformers serve's 422 unprocessable entity (#44620) |
|
Direct revert and terse technical descri |
2026-03-16 |
| COMMIT |
0.00 |
Fix missing / incorrect `config` class in some model class d |
|
Terse commit messages and domain-specifi |
2026-03-15 |
| COMMIT |
0.00 |
Update Nvidia CI docker file to use torch 2.10 (#44712) |
|
Direct, technical changelog with minimal |
2026-03-14 |
| COMMIT |
0.00 |
[`FA`] Fix fa detection (#44703) |
|
Short, domain-specific phrasing and fix |
2026-03-14 |
| COMMIT |
0.00 |
Fix `set_encoder` (#44698) |
|
Minimal message with domain context and |
2026-03-14 |
| COMMIT |
0.00 |
[docs] cb config (#44675) |
|
Extremely brief, informal commit message |
2026-03-13 |
| COMMIT |
0.00 |
Fix more model tester missing `parent` issue (#44685) |
|
Single-word message indicates typical hu |
2026-03-13 |
| COMMIT |
0.00 |
:rotating_light: [`FA4`] Initial support (#42435) |
|
Numerous terse, technical commit lines a |
2026-03-13 |
| COMMIT |
0.00 |
Add register method for `ParallelInterface` (#44640) |
|
Domain term 'feat' and concise summary; |
2026-03-13 |
| COMMIT |
0.00 |
[CB] [Bug] Fix crashes when running without cuda (#44673) |
|
Technical, non-formal phrasing and bulle |
2026-03-13 |
| PR |
0.00 |
Fix failing job `Update Transformers metadata` after #43514 |
|
Terse, domain-specific, and not overly f |
2026-03-23 |
| PR |
0.00 |
Continuous batching thread safety |
|
Free-text is technical, specific, uses s |
2026-03-22 |
| PR |
0.00 |
Fix AutoProcessor.from_pretrained silently dropping hub kwar |
|
Bug report uses technical language, abbr |
2026-03-14 |
| PR |
0.00 |
Support Modular (!!) + Configs in `check_auto_docstrings` |
|
Parenthetical asides, casual tone, and t |
2026-03-17 |
| PR |
0.00 |
Clearer type hints and fix rope validation in configs |
|
Extremely brief, relying on a user tag, |
2026-03-23 |
| PR |
0.00 |
Fix AutoImageProcessor.from_pretrained failing on URL input |
|
References tracker IDs and recent refact |
2026-03-20 |
| PR |
0.00 |
fix tie_weights skipping logic is not tied to model thread s |
|
Colloquial tone, technical detail, and u |
2026-03-23 |
| PR |
0.00 |
Allow arbitrary template kwargs in processors |
|
Informal tone, domain-specific reference |
2026-03-20 |
| PR |
0.00 |
🚨🚧 FeatureExtractor → AudioProcessor |
|
Very terse, only ticket ref; no signs of |
2026-03-02 |
| PR |
0.00 |
Fix Mistral4 tests |
|
Extremely brief, only context reference; |
2026-03-18 |
| PR |
0.00 |
fix(granite_speech): convert int to float for multiplier fie |
|
Description is terse, technical, typical |
2026-03-21 |
| PR |
0.00 |
Support SizeDict import in get_size_dict |
|
Concise, domain-specific phrasing with s |
2026-03-21 |
| PR |
0.00 |
fix `processing_utils.py`: avoid deepcopying tokenizer in `P |
|
Technical discussion; abrupt stop but st |
2026-03-20 |
| PR |
0.00 |
fix: use shape index access in compute_3d_position_ids for Q |
|
Technical, straight to the problem; clea |
2026-03-22 |
| PR |
0.00 |
DeepGEMM |
|
No content; only the template is present |
2026-03-18 |
| PR |
0.00 |
fix: split MXFP4 dependency checks for specific error messag |
|
Technical bullet points, error case focu |
2026-03-22 |
| PR |
0.00 |
Add VidEoMT |
|
Domain jargon, concise, relevant links; |
2026-02-25 |
| PR |
0.00 |
[MOE] MoE routing capture and replay support |
|
Technical, lists, domain abbreviations, |
2026-03-22 |
| PR |
0.00 |
[docs] training on specific hardware |
|
Informal tone, bullet points, typical hu |
2026-03-17 |
| PR |
0.00 |
fix: skip `clean_up_tokenization` for BPE tokenizers in `Pre |
|
Direct, technical, describes a code-spec |
2026-03-21 |
| PR |
0.00 |
Fix missing post_processor in DebertaV2Tokenizer causing no |
|
No free-text content present; only templ |
2026-03-10 |
| PR |
0.00 |
incorrect model list update |
|
Brief, informal and minimal; clearly hum |
2026-03-20 |
| PR |
0.00 |
RagTokenizer: add encode and patch_token(_id) forwarding |
|
Technical, concise; filled template with |
2026-01-24 |
| PR |
0.00 |
[Model] Add UVDoc Model Support |
|
No user-written content present; only te |
2026-01-21 |
| PR |
0.00 |
[Model] Add SLANeXt Model Support |
|
No author free text; only template and c |
2026-02-03 |
| PR |
0.00 |
Remove explicit cuda stream in nemotron_h |
|
Terse, domain-specific, and informal; no |
2026-03-20 |
