| COMMIT |
1.00 |
Add Neuron to auto-compile hardware list (#44757) |
|
Commit message contains explicit AI assi |
2026-04-16 |
| COMMIT |
1.00 |
Fix Kimi-K2.5 tokenizer regression and _patch_mistral_regex |
|
Commit message contains explicit AI assi |
2026-04-13 |
| COMMIT |
1.00 |
fix(serving): resolve rust tokenizer from ProcessorMixin in |
|
Commit message contains explicit AI assi |
2026-04-13 |
| COMMIT |
1.00 |
fix(qwen3_moe): correct return type annotation on Qwen3MoeSp |
|
Explicit mention of Claude Code and 'Bui |
2026-04-13 |
| COMMIT |
1.00 |
docs: fix 5 docstring errors in Gemma3nTextConfig (typos, gr |
|
Mentions 'Built by Rudrendu Paul, develo |
2026-04-13 |
| PR |
1.00 |
SonicMoe |
|
PR body explicitly mentions AI collabora |
2026-04-14 |
| PR |
1.00 |
T5Gemma2: fix `prepare_decoder_input_ids_from_labels` |
|
PR body explicitly mentions AI collabora |
2026-04-19 |
| PR |
1.00 |
Pass packed boundary metadata to Qwen3.5 linear-attention fa |
|
PR body explicitly mentions AI collabora |
2026-03-26 |
| PR |
1.00 |
resize_token_embeddings does not effect to output_embeddings |
|
PR body explicitly mentions AI collabora |
2026-04-08 |
| PR |
1.00 |
Extract dynamic vision/audio tensors into standalone pure fu |
|
PR body explicitly mentions AI collabora |
2026-04-13 |
| PR |
1.00 |
Add EXAONE 4.5 implementations |
|
PR body explicitly mentions AI collabora |
2026-04-16 |
| PR |
1.00 |
Fix NaN in Gemma3/EmbeddingGemma when batching mixed-length |
|
PR body explicitly mentions AI collabora |
2026-04-19 |
| PR |
1.00 |
fix(DSV3): parity between native `DeepseekV3MoE` and remote |
|
PR body explicitly mentions AI collabora |
2026-04-14 |
| PR |
1.00 |
Add Xiaomi MiMo-V2 |
|
PR body explicitly mentions AI collabora |
2026-03-31 |
| PR |
1.00 |
add expert parallelism for gemma-4-26B-A4B-it |
|
PR body explicitly mentions AI collabora |
2026-04-07 |
| PR |
1.00 |
Fix EtaLogitsWarper on fully masked logits |
|
PR body explicitly mentions AI collabora |
2026-04-13 |
| PR |
1.00 |
[Qwen3.5] Fix Qwen3.5 linear attention multi-token cached fo |
|
PR body explicitly mentions AI collabora |
2026-04-19 |
| PR |
0.80 |
Draft commit |
|
'Ai init' and brevity suggest possible A |
2026-04-15 |
| PR |
0.70 |
🚨 Refactor ViT to updated standards |
|
Phrases like 'This PR aims at...' and fo |
2025-10-17 |
| PR |
0.35 |
Proposal to add Qwen3-ASR support [WIP] |
|
Slightly boilerplate but includes domain |
2026-02-08 |
| PR |
0.30 |
audio tester class |
|
Structured but free-text uses natural, b |
2026-04-13 |
| PR |
0.30 |
Fix Mllama torch.compile failure caused by new attention mas |
|
Starts structured, somewhat formal, but |
2026-03-19 |
| PR |
0.25 |
Add IndexCache support for GLM5 DSA |
|
Some formal explanation but contains tec |
2026-04-14 |
| PR |
0.20 |
[Model] Add SLANet Model Support |
|
PR free-text is minimal and casual; huma |
2026-04-20 |
| PR |
0.20 |
Add full GGUF loading support for GPT‑OSS (fixes #43366, sup |
|
Technical, references PRs/issues, and li |
2026-04-18 |
| PR |
0.20 |
[GGUF] Reduce peak RAM usage by casting dequantized tensors |
|
Brief, technical optimization explanatio |
2026-04-12 |
| PR |
0.20 |
Fix CUDA availability check in get_device_properties() |
|
Slightly formal opening but technical su |
2026-04-19 |
| PR |
0.20 |
Fix get_device_properties crash when CUDA is installed but n |
|
Slightly formal greeting but otherwise s |
2026-04-18 |
| PR |
0.20 |
Parakeet tdt |
|
Concise technical explanation, slight fo |
2026-02-20 |
| PR |
0.20 |
[CB] [Major] Add CPU request offloading |
|
Technical summary, direct and clear phra |
2026-04-02 |
| PR |
0.20 |
[docs] model testing |
|
Direct rewrite explanation, domain-focus |
2026-03-31 |
| PR |
0.20 |
Require input_ids for repetition penalty |
|
A bit more formal, but context-specific |
2026-04-13 |
| PR |
0.20 |
Fix Seq2SeqLM ExecuTorch export: add encoder_attention_mask |
|
Structured bug explanations, not generic |
2026-04-20 |
| PR |
0.20 |
[Doc] Fix 'tokenized' -> 'tokenizer' typo in streamer docstr |
|
Brief summary with clear technical focus |
2026-04-18 |
| PR |
0.20 |
Fix `text-to-speech` pipeline crash when generation config c |
|
Technical detail; some formality but no |
2026-03-30 |
| PR |
0.20 |
fix device 0 memory allocate |
|
Domain-specific; concise explanation; no |
2026-01-08 |
| PR |
0.20 |
Fix llama4 bnb mode |
|
Somewhat explanatory, but detailed and t |
2026-03-11 |
| PR |
0.20 |
[Trainer] add MoERouterHealthCallback Callback |
|
Very brief, technical; no AI signals exc |
2026-03-20 |
| PR |
0.20 |
Add AudioFlamingoNext model |
|
Short, technical language; slight formal |
2026-03-18 |
| PR |
0.15 |
Add Videoprism |
|
Concise with domain references; no signs |
2025-08-04 |
| PR |
0.15 |
Add expert parallelism (EP) config support for Qwen3 MoE |
|
Structured, domain-specific config discu |
2026-04-14 |
| COMMIT |
0.10 |
[Doc] Fix 'tokenized' -> 'tokenizer' typo in streamer docstr |
|
Polished but references real authors; cl |
2026-04-20 |
| COMMIT |
0.10 |
[Gemma4] Add docstrings for Per-Layer Embeddings (PLE) pipel |
|
Uses technical detail and casual, non-fo |
2026-04-14 |
| COMMIT |
0.10 |
fix: prevent accelerate from splitting vision encoder by set |
|
Human, domain-focused writing and inform |
2026-04-14 |
| COMMIT |
0.10 |
[Doc] MoE routing capture and replay recipe (#44925) |
|
Informal style with partial sentences an |
2026-04-14 |
| PR |
0.10 |
Add dtype config options for Four Over Six |
|
Brief, domain-specific, and mentions ano |
2026-04-11 |
| PR |
0.10 |
Add /v1/completions endpoint (OpenAI legacy completions API) |
|
Domain-specific, concise, and somewhat t |
2026-03-10 |
| PR |
0.10 |
fix: skip `clean_up_tokenization` for BPE tokenizers in `Pre |
|
Technical and direct with specific jargo |
2026-03-21 |
| PR |
0.10 |
Blockwise mask fn as opt arg in all masking functions |
|
Brief, technical, informal tone; clearly |
2026-04-16 |
| PR |
0.10 |
Improve nested `base_model_prefix` handling in weight conver |
|
Uses domain-specific references and info |
2026-04-13 |
| PR |
0.10 |
Fix: propagate interpolate_pos_encoding through Pixio model |
|
Technical explanation and detail oriente |
2026-04-16 |
| PR |
0.10 |
model: Add DEIMv2 to Transformers |
|
Casual tone, emojis, and bullet points s |
2026-02-27 |
| PR |
0.10 |
[WIP] Major processing refactor |
|
Short, technical, and uses informal expl |
2026-04-17 |
| PR |
0.10 |
feat[vLLM × v5]: Add vLLM compatibility for audio models |
|
Casual bullet-list style and model jargo |
2026-04-08 |
| PR |
0.10 |
[`Conversion Mapping`] Small fixups |
|
Casual, unpolished notes and code contex |
2026-04-16 |
| PR |
0.10 |
[loading] Clean way to add/remove full parts in checkpoint n |
|
'As per the title.' is a terse human idi |
2026-04-15 |
| PR |
0.10 |
:rotating_light: [`Kernels`] Fix kernel function registratio |
|
Slightly formal, but technical and conci |
2026-04-13 |
| PR |
0.10 |
Align gemma3n cache sharing to gemma4 |
|
References PRs, uses human shorthand and |
2026-04-17 |
| PR |
0.10 |
fix: return empty tuple from import_protobuf_decode_error wh |
|
Direct fix, concise and technical explan |
2026-04-16 |
| PR |
0.10 |
throw error when conversion required |
|
Brief, direct language with domain refer |
2026-03-27 |
| PR |
0.10 |
[CB] Changes for long generation |
|
Very minimal free-text, normal shorthand |
2026-04-20 |
| PR |
0.10 |
Fix: modular image processors |
|
Direct reference to PR, informal truncat |
2026-04-17 |
| PR |
0.10 |
Reapply modular to examples |
|
Very terse, human-like brevity without f |
2026-04-20 |
| PR |
0.10 |
chore(qa): split pipeline and add type checking |
|
Concise technical writing, abbreviations |
2026-04-14 |
| PR |
0.10 |
chore: bump doc-builder SHA for PR upload workflow |
|
Short, informal explanation, fits human |
2026-04-15 |
| PR |
0.10 |
xpu output align with cuda in test case |
|
Very direct, technical context, proper n |
2026-04-20 |
| PR |
0.10 |
chore(qa): split out mlinter |
|
Concise phrasing with project-specific r |
2026-04-16 |
| PR |
0.10 |
Fix Zamba2MambaMixer ignoring use_mamba_kernels=False |
|
Direct and terse, with technical detail |
2026-03-19 |
| PR |
0.10 |
[`fix`] Always early return for non-Mistral models in _patch |
|
Contains issue references and brief just |
2026-04-14 |
| PR |
0.10 |
utils: handle flash_attn missing from importlib packages_dis |
|
Technical focus and terse description, n |
2026-04-20 |
| PR |
0.10 |
[nemotron_h] Add support for MLP mixers |
|
Contains domain-specific references, dir |
2026-03-16 |
| PR |
0.10 |
Fix loading logic issue |
|
Casual phrase ('as per the title'), typo |
2026-02-17 |
| PR |
0.10 |
Fix AWQ tests for GPTQModel migration |
|
Contains domain jargon, concise changes, |
2026-03-13 |
| PR |
0.10 |
Fix UMT5EncoderModel embedding weights not being tied after |
|
Technical, concise, and direct issue des |
2026-02-10 |
| PR |
0.10 |
Add qwen3 tts |
|
Concise technical phrasing with domain r |
2026-03-07 |
| PR |
0.10 |
Fix GraniteMoeHybrid _update_mamba_mask crash on attention-o |
|
Brief, focused technical fix; domain jar |
2026-04-19 |
| PR |
0.10 |
[OutputRecorder] re.search on layer_name |
|
Uses domain terms, concise explanations; |
2026-04-19 |
| PR |
0.10 |
Add option to export encoder hidden states for Granite-spee |
|
Technical and direct, uses domain-specif |
2026-03-03 |
| PR |
0.10 |
fix(testing_utils): guard get_device_capability with torch.c |
|
Brief, technical fix and domain referenc |
2026-04-09 |
| PR |
0.10 |
Fix position_ids docstring in modeling_flash_attention_utils |
|
Specific references, natural summary, no |
2026-03-09 |
| PR |
0.05 |
TP refactor for FSDP + TP integration |
|
Terse notes and questions; informal and |
2026-03-26 |
| PR |
0.05 |
qa: re-run modular converter when the script itself is modif |
|
Short, technical summary; no AI style de |
2026-04-20 |
| PR |
0.05 |
Fix conditional check for float formatting |
|
Brief, direct technical explanation; hum |
2026-03-04 |
| PR |
0.05 |
Fix redundant logic in video processing SmolVLM |
|
Very terse, direct fix; typical human PR |
2026-04-06 |
| PR |
0.05 |
fix pin_memory for contiguous batching |
|
Direct technical detail, informal tone, |
2026-03-05 |
| PR |
0.05 |
Enable mxfp4 model on CPU |
|
Terse, implementation-focused, human-sty |
2026-01-27 |
| PR |
0.05 |
Remove `num_frames` in ASR pipeline |
|
Script and technical detail, natural eng |
2026-01-28 |
| PR |
0.05 |
enable cpu paged cache |
|
Technical, includes code and error, info |
2025-12-15 |
| COMMIT |
0.00 |
qa: re-run modular converter when the script itself is modif |
|
Concise, domain-specific commit; informa |
2026-04-20 |
| COMMIT |
0.00 |
[GGUF] Reduce peak RAM usage by casting dequantized tensors |
|
Standard PR/commit with domain details a |
2026-04-20 |
| COMMIT |
0.00 |
Fix CSM `TextToAudioPipeline` missing `` token (#45525) |
|
Terse, human shorthand, dev email signat |
2026-04-20 |
| COMMIT |
0.00 |
Raise 400 on model mismatch when `transformers serve` is pin |
|
Brief, domain-specific terms, minimal ex |
2026-04-20 |
| COMMIT |
0.00 |
[serve] Update tool call to switch to `parse_response` (#454 |
|
Direct, brief language with dev collabor |
2026-04-20 |
| COMMIT |
0.00 |
[`Conversion Mapping`] Small fixups (#45483) |
|
Single-word commit messages, clearly hum |
2026-04-20 |
| COMMIT |
0.00 |
:rotating_light: [`Kernels`] Fix kernel function registratio |
|
Informal, terse, and domain-focused chan |
2026-04-20 |
| COMMIT |
0.00 |
Revert "Fix: modular image processors (#45492)" (#45531) |
|
Standard revert message auto-generated b |
2026-04-20 |
| COMMIT |
0.00 |
fix: return empty tuple from import_protobuf_decode_error wh |
|
Detailed technical bug explanation, info |
2026-04-20 |
| COMMIT |
0.00 |
throw error when conversion required (#45078) |
|
Multiple small, informal commit messages |
2026-04-20 |
| COMMIT |
0.00 |
chore: bump doc-builder SHA for PR upload workflow (#45450) |
|
Standard commit message format, no AI si |
2026-04-20 |
| COMMIT |
0.00 |
xpu output align with cuda in test case (#45526) |
|
Domain-specific terms, signed by a human |
2026-04-20 |
| COMMIT |
0.00 |
chore(qa): split out mlinter (#45475) |
|
Terse, informative, and domain-specific |
2026-04-20 |
| COMMIT |
0.00 |
Fix: modular image processors (#45492) |
|
Single word message, clearly human brevi |
2026-04-20 |
| COMMIT |
0.00 |
[loading] Clean way to add/remove full parts in checkpoint n |
|
Terse, includes typos and informal comme |
2026-04-20 |
| COMMIT |
0.00 |
Fix Zamba2MambaMixer ignoring use_mamba_kernels=False (#4485 |
|
Terse commits, edits, and references; no |
2026-04-20 |
| COMMIT |
0.00 |
revert sha commit pointing to main for transformers_amd_ci_ |
|
Brief, technical, terse revert commit ty |
2026-04-17 |
| COMMIT |
0.00 |
Fix ZeRO-3 from_pretrained: load registered buffers in _load |
|
Technical language, specific errors, dir |
2026-04-17 |
| COMMIT |
0.00 |
Remove redundant condition checks in `get_image_size` method |
|
Terse bullet points, technical, informal |
2026-04-17 |
| COMMIT |
0.00 |
add Qianfan-OCR model definition (#45280) |
|
Informal, iterative fix messages, domain |
2026-04-17 |
| COMMIT |
0.00 |
Add check-auto in repo-consistency and fix sorting (#45481) |
|
Colloquial wording, speculative ('maybe? |
2026-04-17 |
| COMMIT |
0.00 |
Fix typos in src/transformers/utils/output_capturing.py (#45 |
|
Concise typo fix, no free-text, purely a |
2026-04-17 |
| COMMIT |
0.00 |
typing: rule 15 - checks for tie_word_embeddings presence (# |
|
Brief update notes, technical context, c |
2026-04-17 |
| COMMIT |
0.00 |
[CB] Fix capture of max_seqlen (#45323) |
|
Informal commit titles, multiple granula |
2026-04-17 |
| COMMIT |
0.00 |
Fix response api support (#45463) |
|
Commit messages are terse, informal, sho |
2026-04-16 |
| COMMIT |
0.00 |
Minor update (#45484) |
|
Minimal human-written commit; contains C |
2026-04-16 |
| COMMIT |
0.00 |
Allow loading Qwen Thinker 'base' models without generative |
|
Technical explanation, domain language, |
2026-04-16 |
| COMMIT |
0.00 |
[`fix`] Always early return for non-Mistral models in _patch |
|
Human-style summary and messages, includ |
2026-04-16 |
| COMMIT |
0.00 |
Fix spurious position_ids warnings for at least 40 architect |
|
Structured, technical commit explanation |
2026-04-16 |
| COMMIT |
0.00 |
[`fix`] Make Qwen2_5OmniProcessor warning a lot less noisy v |
|
Multiple edits and reverts; informal sty |
2026-04-16 |
| COMMIT |
0.00 |
Dynamic auto mapping (#45018) |
|
Terse, messy progression; highly informa |
2026-04-16 |
| COMMIT |
0.00 |
[serve] Forward `tool_calls`/`tool_call_id` in processor inp |
|
Commit message is concise, technical, an |
2026-04-15 |
| COMMIT |
0.00 |
[docs] vlm addition (#45271) |
|
Short, informal commit messages with min |
2026-04-15 |
| COMMIT |
0.00 |
fix: dont download artifacts from the test hub (#45319) |
|
Commit log has informal, domain-specific |
2026-04-15 |
| COMMIT |
0.00 |
refactor(qa): extend extras so ty can run on server modules |
|
Brief, domain-specific commit message wi |
2026-04-15 |
| COMMIT |
0.00 |
fix(clipseg): fix 2 failing tests (#45403) |
|
Technical, informal bullet points, signe |
2026-04-15 |
| COMMIT |
0.00 |
[docs] @auto_docstring decorator (#45130) |
|
Concise, informal commit messages typica |
2026-04-15 |
| COMMIT |
0.00 |
Fix Sam3Processor missing input_boxes_labels for padded None |
|
Detailed technical explanation using dom |
2026-04-15 |
| COMMIT |
0.00 |
Multimodal serve support (#45220) |
|
Informal, domain-based commit history, h |
2026-04-15 |
| COMMIT |
0.00 |
better grad acc tests (#45434) |
|
Terse commit message with domain abbrevi |
2026-04-15 |
| COMMIT |
0.00 |
avoid wrap 4bit-quantized model into DP (#45407) |
|
Signed by human; no AI indicators in con |
2026-04-15 |
| COMMIT |
0.00 |
Add example for iterative chatting with MLLMs (#45398) |
|
Commit message is terse and has co-autho |
2026-04-15 |
| COMMIT |
0.00 |
Gemma4 resizing per layer inputs (#45324) |
|
Commit message is short, uses domain jar |
2026-04-15 |
| COMMIT |
0.00 |
Add `step3_vl` to `MODELS_WITH_INCORRECT_HUB_TOKENIZER_CLASS |
|
Commit is a standard changelog with clea |
2026-04-15 |
| COMMIT |
0.00 |
Update workflow references to new commit hash (#45442) |
|
Standard commit message; minimal human w |
2026-04-14 |
| COMMIT |
0.00 |
[Doc] Correct checkpoint path in Dinov2 model_docs (#45430) |
|
Commit is concise and fixes a specific t |
2026-04-14 |
| COMMIT |
0.00 |
Fix ty for transformers cli (#45190) |
|
Casual, terse commit messages typical of |
2026-04-14 |
| COMMIT |
0.00 |
fix(models): Resolve regressions in Wav2Vec2PhonemeCTCTokeni |
|
Natural, technical commit flow; includes |
2026-04-14 |
| COMMIT |
0.00 |
Fix Qwen2.5VL temporal grid positions (#45400) |
|
Commit messages are brief, casual, and h |
2026-04-14 |
| COMMIT |
0.00 |
[`fix`] PEFT integration fixes preventing save/load & integr |
|
Technical, concise, with human co-author |
2026-04-14 |
| COMMIT |
0.00 |
Fix the response schema for the gemma4 converter (#45411) |
|
Short, direct PR summary; no AI-style ph |
2026-04-14 |
| COMMIT |
0.00 |
Fix `apply_chat_template` crash on `tool_call` messages with |
|
Terse commit messages with domain terms; |
2026-04-13 |
| COMMIT |
0.00 |
Add SAM3-LiteText (#44320) |
|
Very terse, incremental commit messages; |
2026-04-13 |
| COMMIT |
0.00 |
Fix IndexError with DeepSpeed ZeRO-3 when kernels rotary is |
|
Detailed explanation with domain context |
2026-04-13 |
| COMMIT |
0.00 |
[AMD CI] Fix torch.compile/export failures on AMD CI due to |
|
Brief, informal commit messages; typical |
2026-04-13 |
| COMMIT |
0.00 |
[inference_fusion] convert conv3d patch embed to linear (#45 |
|
Informal commit messages with technical |
2026-04-13 |
| COMMIT |
0.00 |
Fix #45305 + add regression test GAS (#45349) |
|
Terse, informal tone with inline technic |
2026-04-13 |
| COMMIT |
0.00 |
Update `trackio` integration to use Buckets and "freeze" Spa |
|
Mostly placeholder commit messages, no A |
2026-04-13 |
| COMMIT |
0.00 |
Fix: NotebookProgressCallback crash when evaluating with the |
|
Technical bugfixes and test updates in b |
2026-04-13 |
| COMMIT |
0.00 |
Less unnecessary RoPE warnings (#45289) |
|
Terse, domain-specific commit; no AI sig |
2026-04-13 |
| COMMIT |
0.00 |
[`Tokenizers`] Move gpt sw3 tokenizer out (#45404) |
|
Brief, technical fix; no AI phrasing pre |
2026-04-13 |
| COMMIT |
0.00 |
Fix unintended Hub metadata calls from _patch_mistral_regex |
|
Detailed, technical with domain slang; h |
2026-04-13 |
| COMMIT |
0.00 |
Fix MoE routers returning probabilities instead of logits (# |
|
Technical fix, informal style; clearly h |
2026-04-13 |
| COMMIT |
0.00 |
Fix NaN weights on non-rank-0 FSDP processes (#45050) |
|
Short, domain-specific; lacks AI tone. |
2026-04-13 |
| COMMIT |
0.00 |
remove cache file from tree (#45392) |
|
Very terse commit about cache file; huma |
2026-04-13 |
| COMMIT |
0.00 |
[docs] training on specific hardware (#44799) |
|
Commit messages are terse and minimal, t |
2026-04-10 |
| COMMIT |
0.00 |
[docs] zero + sequence parallelism (#44605) |
|
Very brief, domain-specific shorthand; n |
2026-04-10 |
| COMMIT |
0.00 |
Fix vlm weight mappings (#45358) |
|
Informal tone and shorthand imply human |
2026-04-10 |
| COMMIT |
0.00 |
Copy the template resolution logic from the base apply_chat_ |
|
Direct, informal commit log—no AI phrasi |
2026-04-10 |
| COMMIT |
0.00 |
add kwargs to all methods in the CallbackHandler class (#453 |
|
No text beyond the conventional PR title |
2026-04-10 |
| COMMIT |
0.00 |
Close file handler (#45187) |
|
Terse, technical fix with human co-autho |
2026-04-10 |
| COMMIT |
0.00 |
fix: restore mypy type checking for PreTrainedConfig subclas |
|
Technical summary with explicit changelo |
2026-04-10 |
| COMMIT |
0.00 |
`cohere_asr`: fix device issue for `test_model_parallel_beam |
|
Domain-specific fixes, Signed-off-by and |
2026-04-10 |
| COMMIT |
0.00 |
Fix AttributeError in Gemma3ForConditionalGeneration and Gem |
|
Standard patch title and human co-author |
2026-04-10 |
| COMMIT |
0.00 |
fix bug for videomt model device mismatch (#45204) |
|
Domain-specific, includes Signed-off-by; |
2026-04-10 |
| COMMIT |
0.00 |
fix gemma4 gradient accumulation loss and last token incorre |
|
Terse commit messages and domain abbrevi |
2026-04-10 |
| COMMIT |
0.00 |
Logger has `[transformers]` prefix in non-verbose mode (#453 |
|
Very short, casual commit messages sugge |
2026-04-10 |
| COMMIT |
0.00 |
Fix AttributeError in AssistantToTargetTranslator.unmap_inpu |
|
Technical problem explanation, domain de |
2026-04-10 |
| COMMIT |
0.00 |
Fix Qwen2.5-VL temporal RoPE scaling applied to still images |
|
Technical and precise, contains domain-s |
2026-04-10 |
| COMMIT |
0.00 |
musicflamingo: add test support for Intel XPU device (#45212 |
|
Terse, domain-specific, and includes sig |
2026-04-10 |
| COMMIT |
0.00 |
nomic_bert: make the test suitable for general device. (#452 |
|
Minimal, with only template sign-off and |
2026-04-10 |
| COMMIT |
0.00 |
Skip invalid flash-attn tests for `pi0` model (#45011) |
|
Terse, informal commit messages and incl |
2026-04-10 |
| COMMIT |
0.00 |
Add cuda compatibility check for using `grouped_mm` (#45001) |
|
Terse commit subject; technical co-autho |
2026-04-10 |
| COMMIT |
0.00 |
Load adapter with TP (#45155) |
|
Terse, uses technical abbreviations, no |
2026-04-09 |
| COMMIT |
0.00 |
[docs] tp training (#44613) |
|
Very minimal, domain-specific shorthand, |
2026-04-09 |
| COMMIT |
0.00 |
[docs] training performance (#44342) |
|
Short, informal phrases typical of human |
2026-04-09 |
| COMMIT |
0.00 |
[docs] optimizers, hyperparam search, training features (#44 |
|
Informal list format, domain terms, not |
2026-04-09 |
| COMMIT |
0.00 |
Remove unused parameters and improve add_tensor_parallel_hoo |
|
Domain-specific wording, includes co-aut |
2026-04-09 |
| COMMIT |
0.00 |
Use torchvision `decode_image` to load images in the torchv |
|
Succinct, technical, includes co-author, |
2026-04-09 |
| COMMIT |
0.00 |
[gemma4] Fix device map auto (#45347) |
|
Brief, uses domain terminology, not AI g |
2026-04-09 |
| COMMIT |
0.00 |
Refactor CLIP-like models (#44431) |
|
Terse, casual edits, clear human voice t |
2026-04-09 |
| PR |
0.00 |
Drop `content=None` from messages in `apply_chat_template` |
|
Terse, domain-specific language with par |
2026-04-14 |
| PR |
0.00 |
cache_utils: fix QuantizedLayer to correctly propagate reord |
|
Technical, concise; clear domain knowled |
2026-04-19 |
| PR |
0.00 |
Fix AMD CI: rebuild torchvision with libjpeg + refresh expec |
|
Technical changelog, specific; not AI-li |
2026-04-20 |
| PR |
0.00 |
Fix CSM `TextToAudioPipeline` missing `` token |
|
Jargon, context-specific wording, inform |
2026-04-20 |
| PR |
0.00 |
Raise 400 on model mismatch when `transformers serve` is pin |
|
Direct technical description, not overly |
2026-04-14 |
| PR |
0.00 |
Add GGUF support to Gemma4 (31B & 26B-A4B) text |
|
Mostly template text; unique title is hu |
2026-04-07 |
| PR |
0.00 |
[serve] Update tool call to switch to `parse_response` |
|
Some awkward grammar and terseness indic |
2026-04-16 |
| PR |
0.00 |
[Trainer] Add ddp_static_graph option |
|
Technical summary, concise, not overly f |
2026-04-20 |
| PR |
0.00 |
Fix: propagate quantization_config to text sub-config for co |
|
Detailed, technical, with typos/awkward |
2026-04-17 |
| PR |
0.00 |
ALM base model class |
|
Domain-specific discussion, non-AI tone, |
2026-04-20 |
| PR |
0.00 |
fix(x_clip): fix 8 failed test cases |
|
Very concise bugfix title, no AI hallmar |
2026-04-13 |
| PR |
0.00 |
sam3_lite_text: skip flash_attn_2_can_dispatch_composite_mod |
|
Direct mention of reviewer and informal |
2026-04-16 |
| PR |
0.00 |
Fix EP: RouterParallel shape, tp_plan property, grouped_mm s |
|
Detailed real-world context, domain-spec |
2026-04-16 |
| PR |
0.00 |
Revert "Fix: modular image processors (#45492)" |
|
Standard GitHub revert language and info |
2026-04-20 |
| PR |
0.00 |
feat: add Gemma4ForSequenceClassification |
|
Extremely brief, standard technical comm |
2026-04-07 |
| PR |
0.00 |
Remove explicit cuda stream in nemotron_h |
|
Informal, concise; domain-specific; no A |
2026-03-20 |
| PR |
0.00 |
Update cpu torchao usage |
|
Very terse and to the point; no AI signa |
2025-12-22 |
| PR |
0.00 |
Remove ipex/ccl in CPU training doc |
|
Uses list syntax; terse, informal, domai |
2025-12-15 |
| PR |
0.00 |
Fix speccht5_tts pipeline |
|
Minimal description; references issue; n |
2025-12-12 |
| PR |
0.00 |
Fix xpu output check for Ministral3 tests |
|
Simply repeats PR title; no AI traits; h |
2025-12-10 |
| PR |
0.00 |
Compress |
|
Short, informal, clear domain context. |
2025-12-05 |
