| COMMIT |
1.00 |
Fix `maybe_autocast` crashing on meta device tensors (#44984 |
|
Commit message contains explicit AI assi |
2026-03-25 |
| COMMIT |
1.00 |
LwDetrImageLoss: Fix dtype casting to prevent crash when usi |
|
Commit message contains explicit AI assi |
2026-03-24 |
| 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 |
| PR |
1.00 |
Add cuda compatibility check for using `grouped_mm` |
|
PR body explicitly mentions AI collabora |
2026-03-25 |
| PR |
1.00 |
Add doc page for capturing outputs |
|
PR body explicitly mentions AI collabora |
2026-03-23 |
| PR |
1.00 |
Add sarvam model |
|
PR body explicitly mentions AI collabora |
2026-03-25 |
| PR |
1.00 |
Add sarvam model |
|
PR body explicitly mentions AI collabora |
2026-03-25 |
| PR |
1.00 |
feat: Add router_logits override to enable Routing Replay fo |
|
PR body explicitly mentions AI collabora |
2026-03-23 |
| PR |
1.00 |
Fix CPU 16 bytes alignment issue using equivalent fallback |
|
PR body explicitly mentions AI collabora |
2026-03-24 |
| PR |
1.00 |
fixed import error with PILImageResampling |
|
PR body explicitly mentions AI collabora |
2026-03-23 |
| PR |
1.00 |
Fix: Add correct return behaviour when output_hidden_states= |
|
PR body explicitly mentions AI collabora |
2026-03-23 |
| PR |
1.00 |
try |
|
PR body explicitly mentions AI collabora |
2026-03-24 |
| PR |
0.55 |
🚨 Refactor ViT to updated standards |
|
"This PR aims at..." phrasing slightly C |
2025-10-17 |
| PR |
0.20 |
Fix MobileNet v1/v2 image processor default interpolation to |
|
Concise technical summary, normal engine |
2026-01-16 |
| PR |
0.20 |
Fix AutoProcessor.from_pretrained silently dropping hub kwar |
|
Clear bug explanation in technical style |
2026-03-14 |
| PR |
0.20 |
FSDP2 native support in transformers |
|
Technical content with domain-specific t |
2026-02-17 |
| PR |
0.20 |
[PoC] HF exporters |
|
Some informal phrasing, references other |
2025-11-03 |
| PR |
0.20 |
Fix Mllama torch.compile failure caused by new attention mas |
|
Technical explanation, domain-specific, |
2026-03-19 |
| PR |
0.15 |
docs(yoso): fix and stabilize doctest for YOSO |
|
Some formality, but human-style headings |
2026-01-17 |
| PR |
0.12 |
docs: add energy efficiency considerations to bitsandbytes q |
|
Brief practical section added; lacks AI |
2026-03-03 |
| COMMIT |
0.10 |
Fix missing post_processor in DebertaV2Tokenizer causing no |
|
Detailed technical context, natural phra |
2026-03-24 |
| COMMIT |
0.10 |
Add big angry code agent warnings! (#44890) |
|
Commit messages use domain language and |
2026-03-22 |
| PR |
0.10 |
Enable multiprocessing in glue datasets |
|
Slightly formal but concise, domain-spec |
2020-06-06 |
| PR |
0.10 |
Internalise the NomicBERT model |
|
Structured explanation, some formality, |
2025-12-29 |
| PR |
0.10 |
Embedding VLMs don't need a head |
|
Brief free-text and domain-specific phra |
2026-03-25 |
| PR |
0.10 |
Add Music Flamingo |
|
Direct, technical content with human-wri |
2026-01-27 |
| PR |
0.10 |
Add THD support in ESM |
|
Technical, informal tone; specific model |
2026-02-19 |
| PR |
0.10 |
[Cache] Native mamba & hybrid cache |
|
Domain-specific content with informal to |
2026-03-23 |
| PR |
0.10 |
Add VidEoMT |
|
Technical description, references, and i |
2026-02-25 |
| PR |
0.10 |
Add deepseek 3.2 exp |
|
Domain-specific code block and direct la |
2025-10-01 |
| PR |
0.10 |
ci: add anti-slop action |
|
Short, technical changes, human-like str |
2026-03-19 |
| PR |
0.10 |
Ensure final evaluation runs with step-based evaluation stra |
|
Technical explanation with domain jargon |
2026-02-19 |
| PR |
0.10 |
update release workflow |
|
Informal tone, uses 'Just', domain-speci |
2026-02-18 |
| PR |
0.10 |
Add qwen3 tts |
|
Technical content, domain-specific langu |
2026-03-07 |
| PR |
0.10 |
Allow arbitrary template kwargs in processors |
|
Casual style, references to code logic a |
2026-03-20 |
| PR |
0.09 |
Add StyleTTS 2 |
|
Technical addition and code refs; casual |
2025-01-20 |
| COMMIT |
0.07 |
Fix: Update optimization.py (#44909) |
|
Technical explanation and changelog; no |
2026-03-24 |
| PR |
0.07 |
fix(camembert): add tie_word_embeddings=True to CamembertCon |
|
Describes regression and technical fix c |
2026-03-22 |
| COMMIT |
0.05 |
Fix tie_word_embedding issues with `Qwen2VL` (#44976) |
|
Commit messages are terse and domain-spe |
2026-03-24 |
| COMMIT |
0.05 |
Support Modular (!!) + Configs in `check_auto_docstrings` (# |
|
Brief, technical changelog; no AI text h |
2026-03-24 |
| 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 |
Trainer: set skip_logits for loss-only eval when liger enabl |
|
Terse, domain-specific explanation, typi |
2026-03-25 |
| PR |
0.05 |
Fix failing `Qwen3OmniModelIntegrationTests` |
|
Free-text is brief and informal with a p |
2026-03-19 |
| PR |
0.05 |
fix(testing): Fix Parakeet, Evolla, Pi0, and Phi-3 test fail |
|
Casual, grouped test fixes; human tone a |
2026-03-25 |
| PR |
0.05 |
Fix type hint for `attention_chunk_size` in `Llama4TextConfi |
|
Concise, domain-specific explanation, la |
2026-03-25 |
| PR |
0.05 |
fix: handle absent sys.modules entry in modeling_utils |
|
Uses domain jargon and code snippets, co |
2026-03-24 |
| PR |
0.05 |
Fix `maybe_autocast` crashing on meta device tensors |
|
Domain-specific issue summary; clear and |
2026-03-25 |
| PR |
0.05 |
fix: preserve rotary_pct across save/load cycle in GPTNeoX c |
|
Technical, concise summary; lacks AI-for |
2026-03-25 |
| PR |
0.05 |
fix: remove Copied from comments between @torch.jit.script a |
|
PR content is concise, specific, typo pr |
2026-03-25 |
| PR |
0.05 |
fix: implement Mxfp4Dequantize.reverse_op for save_pretraine |
|
Terse, technical description, with jargo |
2026-03-25 |
| PR |
0.05 |
bug-fix: do not assume torch.cuda is available when setting |
|
Brief, direct bug description; filled wi |
2026-03-24 |
| PR |
0.05 |
fix(models): Fix Perceiver interpolate_pos_encoding interpol |
|
Concise issue reference and technical fi |
2026-03-20 |
| PR |
0.05 |
Fix: pass kwargs to cross_entropy in fixed_cross_entropy |
|
Direct issue reference and bug explanati |
2026-01-13 |
| PR |
0.05 |
Fix tie_word_embedding issues with `Qwen2VL` |
|
Bullet points, model-specific jargon, te |
2026-03-24 |
| PR |
0.05 |
Add SAM3-LiteText |
|
Domain-focused, includes citation, short |
2026-02-27 |
| PR |
0.05 |
[CB] [Minor] Simplify test suite |
|
Technically focused, uses shorthand and |
2026-03-19 |
| COMMIT |
0.02 |
refactor: mlinter as its own package (#44939) |
|
Informal, domain-specific commit message |
2026-03-24 |
| COMMIT |
0.01 |
[CB] [Minor] Simplify test suite (#44858) |
|
Minimal, terse commit messages; highly h |
2026-03-24 |
| COMMIT |
0.01 |
Allow arbitrary template kwargs in processors (#44881) |
|
Commit messages are brief and informal; |
2026-03-24 |
| COMMIT |
0.01 |
incorrect model list update (#44880) |
|
Terse, casual commit history; human-writ |
2026-03-24 |
| COMMIT |
0.01 |
[CB] Add an option to return logprobs (#44835) |
|
Brief, informal commit messages; human s |
2026-03-23 |
| COMMIT |
0.01 |
[docs] peft (#44804) |
|
Very minimal and informal; no signs of A |
2026-03-23 |
| COMMIT |
0.01 |
Continuous batching thread safety (#44924) |
|
Informal commit log, technical focus; ty |
2026-03-23 |
| COMMIT |
0.01 |
Add static FP8 expert support (#44895) |
|
Highly terse, typical human commit patte |
2026-03-23 |
| COMMIT |
0.00 |
Fix type hint for `attention_chunk_size` in `Llama4TextConfi |
|
Brief, domain-specific commit, no AI sig |
2026-03-25 |
| COMMIT |
0.00 |
Fix AutoProcessor.from_pretrained silently dropping hub kwar |
|
Detailed tech explanation, domain jargon |
2026-03-25 |
| COMMIT |
0.00 |
Add VidEoMT (#44285) |
|
Commit history is terse and technical, h |
2026-03-25 |
| COMMIT |
0.00 |
fix: remove Copied from comments between @torch.jit.script a |
|
Concise, domain-specific explanation, cl |
2026-03-25 |
| COMMIT |
0.00 |
More small vllm fixes (#44990) |
|
Commit messages are terse and informal; |
2026-03-25 |
| COMMIT |
0.00 |
fix(models): Fix Perceiver interpolate_pos_encoding interpol |
|
Commit messages use domain jargon and in |
2026-03-25 |
| COMMIT |
0.00 |
Allow `mm_token_type` be non-padded lists (#44563) |
|
Commit log is brief and contains human-l |
2026-03-25 |
| COMMIT |
0.00 |
Fix CPU 16 bytes alignment issue using equivalent fallback ( |
|
Commit messages are terse, technical, an |
2026-03-25 |
| COMMIT |
0.00 |
refactor: unify QA calls (#44879) |
|
Commit messages filled with informal ton |
2026-03-25 |
| COMMIT |
0.00 |
[ `vllm x v5`] nit (#44971) |
|
Very terse nits and technical jargon, ty |
2026-03-24 |
| COMMIT |
0.00 |
[AMD CI] Gemma3/Gemma3n Expectations (#44972) |
|
Direct, slangy commits and clear domain |
2026-03-24 |
| COMMIT |
0.00 |
Officially launch parse_response (#44674) |
|
— |
2026-03-24 |
| COMMIT |
0.00 |
fix load_best_model_checkpoint_at_end do not load the best m |
|
— |
2026-03-24 |
| COMMIT |
0.00 |
fix: split MXFP4 dependency checks for specific error messag |
|
— |
2026-03-24 |
| COMMIT |
0.00 |
Fix failing `T5ModelIntegrationTest` (#44934) |
|
— |
2026-03-24 |
| COMMIT |
0.00 |
Config kwargs (#44953) |
|
— |
2026-03-24 |
| COMMIT |
0.00 |
Fix variable shadowing in pipeline example and typo in BART |
|
Commit message is terse and domain-speci |
2026-03-23 |
| 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 |
| PR |
0.00 |
Refactor CLIP-like models |
|
Casual tone, domain abbreviations, human |
2026-03-04 |
| PR |
0.00 |
Module Fusion API |
|
No free-text content; only section/title |
2026-03-24 |
| PR |
0.00 |
fix series of failed test case for janus model |
|
Minimal, terse description; no AI writin |
2026-03-16 |
| PR |
0.00 |
from_pretrained distributed refactor (FSDP2 + TP) |
|
Bulleted todo list and technical terms; |
2026-03-25 |
| PR |
0.00 |
Refactor core_model_loading to support FSDP shard-on-read lo |
|
Title only, no text to evaluate for AI s |
2026-03-24 |
| PR |
0.00 |
refactoring: speedup static checks |
|
PR content is technical, concise, and co |
2026-03-25 |
| PR |
0.00 |
🚨 Distributed training API |
|
PR shows domain-specific code and goal; |
2026-03-25 |
| PR |
0.00 |
Dynamic weight conversion is recursive |
|
PR description uses domain references an |
2026-02-26 |
| PR |
0.00 |
Add inference time layer fusion optimisations via `PreTraine |
|
Only a title, no free-text content. |
2026-03-23 |
| PR |
0.00 |
Fix max_seqlen type in vision attention for torch.compile + |
|
— |
2026-03-24 |
| PR |
0.00 |
Add sarvam model |
|
Only a model addition title, no free tex |
2026-03-25 |
| PR |
0.00 |
More small vllm fixes |
|
PR content is minimal and domain-specifi |
2026-03-25 |
| PR |
0.00 |
typing: add rule 14 - checks for tie_word_embeddings presenc |
|
PR content is terse, technical, and uses |
2026-03-25 |
| PR |
0.00 |
Allow `mm_token_type` be non-padded lists |
|
— |
2026-03-10 |
| PR |
0.00 |
perceptron: Isaac-0.1 implementation |
|
— |
2025-09-18 |
| PR |
0.00 |
[kernels] update docker file |
|
— |
2026-02-12 |
| PR |
0.00 |
refactor: unify QA calls |
|
— |
2026-03-20 |
| PR |
0.00 |
[refactor] Serving into proper modules |
|
— |
2026-03-17 |
| PR |
0.00 |
Fix llama4 bnb mode |
|
— |
2026-03-11 |
| PR |
0.00 |
Fix: NotebookProgressCallback crash when evaluating with the |
|
— |
2026-03-23 |
| PR |
0.00 |
[WIP] Add CharacterBERT model |
|
Template reused; no filled free-text con |
2023-10-05 |
| PR |
0.00 |
Parakeet tdt |
|
— |
2026-02-20 |
| PR |
0.00 |
Remove unused parameters and improve add_tensor_parallel_hoo |
|
— |
2026-03-16 |
| PR |
0.00 |
Support Modular (!!) + Configs in `check_auto_docstrings` |
|
— |
2026-03-17 |
| PR |
0.00 |
[ `vllm x v5`] nit |
|
— |
2026-03-24 |
| PR |
0.00 |
fix: rebase main; clean config reads, ImageProcessor backend |
|
Only title, topic-specific, very terse; |
2026-03-24 |
| PR |
0.00 |
LwDetrImageLoss: Fix dtype casting to prevent crash when usi |
|
— |
2026-03-20 |
| PR |
0.00 |
[AMD CI] Gemma3/Gemma3n Expectations |
|
— |
2026-03-24 |
| PR |
0.00 |
Officially launch parse_response |
|
— |
2026-03-13 |
| PR |
0.00 |
fix load_best_model_checkpoint_at_end do not load the best m |
|
— |
2026-03-10 |
| PR |
0.00 |
fix: split MXFP4 dependency checks for specific error messag |
|
— |
2026-03-22 |
| PR |
0.00 |
feat: added cache to the model linter |
|
— |
2026-03-17 |
| PR |
0.00 |
fix tie_weights skipping logic is not tied to model thread s |
|
— |
2026-03-23 |
| PR |
0.00 |
Add `base_model_tp_plan` to `OlmoeConfig` |
|
— |
2026-03-13 |
| PR |
0.00 |
Added Make to the docker and `tomli` to `.[quality]` |
|
— |
2026-03-24 |
| PR |
0.00 |
Fix failing `T5ModelIntegrationTest` |
|
— |
2026-03-22 |
| PR |
0.00 |
fix |
|
— |
2026-01-31 |
| PR |
0.00 |
Dequant fix |
|
— |
2026-03-18 |
| PR |
0.00 |
Fix flash attention crash with 3D position_ids (Qwen3.5) |
|
— |
2026-03-21 |
| PR |
0.00 |
Add static FP8 expert support |
|
— |
2026-03-20 |
| PR |
0.00 |
add HyperClovaX Vision |
|
— |
2026-02-27 |
| PR |
0.00 |
Config kwargs |
|
— |
2026-03-23 |
| PR |
0.00 |
fix: Add MXFP4 MoE/attention backward kernels |
|
— |
2026-02-05 |
| PR |
0.00 |
Fix FA kernel launch needs correct cuda device ctx in multi- |
|
— |
2026-03-24 |
| PR |
0.00 |
model: Add DEIMv2 to Transformers |
|
— |
2026-02-27 |
| PR |
0.00 |
Add RF-DETR |
|
— |
2025-03-21 |
| PR |
0.00 |
DeepGEMM |
|
— |
2026-03-18 |
| PR |
0.00 |
Fix AutoImageProcessor.from_pretrained failing on URL input |
|
— |
2026-03-20 |
| PR |
0.00 |
Fix whisper return language |
|
Template plus minimal free text; all fil |
2025-11-16 |
| PR |
0.00 |
Modular playground |
|
Updates listed with domain-specific, inf |
2026-02-04 |
| PR |
0.00 |
Fix: Update optimization.py |
|
Technical explanation is concise and inf |
2026-03-21 |
| PR |
0.00 |
[MOE] MoE routing capture and replay support |
|
Domain jargon, concise summary, informal |
2026-03-22 |
| PR |
0.00 |
Fix backward compatibility for CommonKwargs in processing_ut |
|
Domain-specific, concise summary, inform |
2026-03-24 |
