| commit |
1.00 |
Lfm2: also thread `seq_idx` through ShortConv.slow_forward ( |
|
Commit message contains explicit AI assi |
2026-06-16 |
| commit |
1.00 |
feat(generation): allow user to keep input tensors on cpu (# |
|
Commit message contains explicit AI assi |
2026-06-15 |
| commit |
1.00 |
Lfm2: thread `seq_idx` through ShortConv for packed/varlen i |
|
Commit message contains explicit AI assi |
2026-06-12 |
| commit |
1.00 |
Remove unnecessary expand_as in get_placeholder_mask across |
|
Commit message contains explicit AI assi |
2026-06-11 |
| commit |
1.00 |
[CI] Implement real security check in PR CI security gate (# |
|
Commit message contains explicit AI assi |
2026-06-11 |
| commit |
1.00 |
[CI] Add 60s delay in security gate for flow observation (#4 |
|
Commit message contains explicit AI assi |
2026-06-11 |
| commit |
1.00 |
[TBC] [CI] Auto-approve PR CI for fork PRs via security gate |
|
Commit message contains explicit AI assi |
2026-06-11 |
| pr |
1.00 |
Add support for Nemotron Omni |
|
PR body explicitly mentions AI collabora |
2026-06-08 |
| pr |
1.00 |
Sapiens2: Remove modeling imports in image processing |
|
PR body explicitly mentions AI collabora |
2026-06-15 |
| pr |
1.00 |
Return logits from semantic segmentation post-process |
|
PR body explicitly mentions AI collabora |
2026-05-22 |
| pr |
1.00 |
[Mistral] Move MistralConverter into integrations/mistral/ p |
|
PR body explicitly mentions AI collabora |
2026-06-12 |
| pr |
1.00 |
Fix packed-sequence mask ignored when a 2D attention_mask is |
|
PR body explicitly mentions AI collabora |
2026-06-14 |
| pr |
1.00 |
Fix docstring formatting issues causing Sphinx autodoc warni |
|
PR body explicitly mentions AI collabora |
2026-06-12 |
| pr |
1.00 |
DSv4 static cache |
|
PR body explicitly mentions AI collabora |
2026-06-03 |
| pr |
1.00 |
Fix use_cache with seq_len > 1 ( #46032) |
|
PR body explicitly mentions AI collabora |
2026-05-19 |
| pr |
0.20 |
Add Susono model |
|
Domain-specific acronyms and direct expl |
2026-06-16 |
| pr |
0.15 |
Add test for additionalProperties=False in recursive_parse |
|
Slightly more formal, but still concise |
2026-06-12 |
| commit |
0.10 |
docs: fix typo in make_list_of_images docstring (#46469) |
|
Slightly formal but consistent with huma |
2026-06-08 |
| pr |
0.10 |
Add TIPSv2 |
|
Concise changelog-style; domain terms pr |
2026-06-02 |
| pr |
0.10 |
[Fix] Serve drops some attributes at serialization |
|
Technical, clear issue reference, domain |
2026-06-16 |
| pr |
0.10 |
[CTRL] Support attn_implementation dispatch |
|
Concise, problem-focused, and normal for |
2026-05-19 |
| pr |
0.10 |
fix: honor static cache max len config |
|
Bullet list, terse, domain-specific, not |
2026-06-05 |
| pr |
0.10 |
feat(pipelines): accept numpy arrays and tensors in ImageCla |
|
Direct, concise, technical language, not |
2026-06-11 |
| pr |
0.10 |
Lazily build the default kernel mapping to decouple `kernels |
|
Technical language, code snippet, and di |
2026-06-16 |
| pr |
0.10 |
fix(masking): cast tensor q_offset/kv_offset to int in flex_ |
|
Domain-specific reference and bugfix sum |
2026-06-16 |
| pr |
0.10 |
Document/type torch.Tensor EvalPrediction fields under batch |
|
Human style with issue reference and dir |
2026-06-16 |
| pr |
0.10 |
fix: define TransformersCompletionCreateParamsStreaming insi |
|
Technical summary and standard git/issue |
2026-06-15 |
| pr |
0.10 |
Fix dynamic RoPE not resetting inv_freq when layer_type is N |
|
Contains specific file, method, and vari |
2026-06-13 |
| pr |
0.10 |
Fix Mistral models which contain both `tokenizer.json` and ` |
|
Direct, context-aware technical explanat |
2026-06-13 |
| pr |
0.10 |
Multi-gpu loading when the whole backbone is tied |
|
Informal, technical, and contextually sp |
2026-06-13 |
| pr |
0.10 |
Smovlm: pad videos up to max frames |
|
Direct, technical, human review with cas |
2026-06-15 |
| pr |
0.10 |
Made SAM3 to return loss in outputs |
|
Domain details, references prior art, an |
2026-06-13 |
| pr |
0.10 |
Fix StaticLayer get_seq_length type hint |
|
Domain-specific, terse summary; no AI ha |
2026-06-07 |
| pr |
0.10 |
[DiffusionGemma] Return router logits and load balancing los |
|
Informal tone and partial sentence end. |
2026-06-14 |
| pr |
0.10 |
Fix several bugs in `cache_implementation=static` |
|
Technical, summary with commit list; no |
2026-06-05 |
| pr |
0.10 |
Fix Wav2Vec2PhonemeCTCTokenizer stale phonemizer_lang state |
|
Direct bugfix explanation referencing sp |
2026-06-16 |
| pr |
0.10 |
Fix byte-level BPE tokenizers reusing SentencePiece classes |
|
Direct technical explanation and real-wo |
2026-06-16 |
| pr |
0.10 |
[CB] Fix offloading |
|
Uses domain language, concise and techni |
2026-06-12 |
| pr |
0.10 |
Fix CodeLlama tokenizer stripping real leading spaces on dec |
|
Uses technical detail and code symbols; |
2026-06-11 |
| pr |
0.10 |
Fix DeepSeek-Coder v1 / DeepSeek-LLM v1 loading as wrong tok |
|
Issue references, technical content, inf |
2026-06-12 |
| pr |
0.10 |
[Qwen3VL] Fix video token placeholder: use self.video_token |
|
Direct technical fix, jargon, and contex |
2026-05-30 |
| pr |
0.10 |
DIstributed branch base |
|
Informal, includes acknowledgement, doma |
2026-05-28 |
| pr |
0.10 |
clip_grad_norm avoid O(global_params) replication |
|
References another PR, domain detail, in |
2026-05-20 |
| pr |
0.10 |
docs(zh): add Chinese translation of kernels.md |
|
Straightforward, project-specific conten |
2026-06-13 |
| pr |
0.10 |
🚨 🚨Bring some dinos to modern standards |
|
Colloquial, includes emoji, project cont |
2026-05-28 |
| pr |
0.10 |
mistral common backend fix |
|
Uses project jargon, highly concise, hum |
2026-06-15 |
| pr |
0.10 |
[pr template] update |
|
Very terse, linked issue, human approval |
2026-06-12 |
| pr |
0.10 |
feat(generation): allow user to keep input tensors on cpu |
|
Domain-language, specific, no AI-style w |
2026-06-12 |
| pr |
0.10 |
[wav2vec2] Support attn_implementation=sdpa dispatch |
|
Technical and concise, with domain-speci |
2026-05-25 |
| pr |
0.10 |
Add new model: Isaac |
|
Reference to another PR, casual tone, an |
2026-04-02 |
| pr |
0.10 |
Fix AttributeError in auto_factory when model_class lacks co |
|
Contains error trace and domain-specific |
2026-06-15 |
| pr |
0.10 |
[model] Support MiniCPM-V 4.0 |
|
Succinct, uses domain reference, human r |
2025-08-05 |
| pr |
0.10 |
fix(generation): honor cache_config.max_cache_len for static |
|
Bullet list and technical implementation |
2026-06-15 |
| pr |
0.10 |
fix(t5): use "decoder's " prefix in error messages when call |
|
References issue, technical context, and |
2026-06-15 |
| pr |
0.10 |
[DiffusionGemma] Support gradient checkpointing |
|
Technical, concise, and references previ |
2026-06-11 |
| pr |
0.10 |
[CB] Slice logits inside the model |
|
Human tone, uses jargon and informal lan |
2026-06-15 |
| pr |
0.10 |
ci: add NO_COLOR=1 to suppress ANSI color codes in CI output |
|
Straightforward technical summary typica |
2026-06-15 |
| pr |
0.10 |
[New Model] Add MiniCPM3 support |
|
Brief domain-specific update, direct and |
2026-04-23 |
| pr |
0.10 |
feat: Add HRM Model |
|
Short, domain-specific content, some tem |
2025-10-01 |
| commit |
0.05 |
Make DiffusionGemma trainable (#46568) |
|
Changelog format, technical terms, conci |
2026-06-11 |
| commit |
0.05 |
docs: 🌐 add Turkish translation for README file (#46312) |
|
Direct, signed contribution, informal hu |
2026-06-11 |
| commit |
0.05 |
Add Parakeet-RNNT (#46331) |
|
Technical jargon, typos, terse commit me |
2026-06-11 |
| commit |
0.05 |
[CI] Catch all shell/process execution issues in security ga |
|
Technical explanation, domain-specific d |
2026-06-11 |
| commit |
0.05 |
Honor a concrete dtype in AutoModel for composite checkpoint |
|
Technical explanation, typos, domain-spe |
2026-06-11 |
| commit |
0.05 |
[CB] [Minor] Add data-parallel to overall script (#46490) |
|
Abbreviated, technical log updates, very |
2026-06-11 |
| commit |
0.05 |
Fix missing Gemma4Processor._compute_audio_num_tokens (#4641 |
|
Contains domain jargon, concise technica |
2026-06-10 |
| pr |
0.05 |
Fix: honor cache_config["max_cache_len"] on the static-cache |
|
Concise domain-specific content with nat |
2026-06-16 |
| pr |
0.05 |
Add GB10/SM121 Hub-kernel path for Qwen3.6 Gated DeltaNet |
|
Uses domain jargon, informal tone, not A |
2026-06-04 |
| pr |
0.05 |
fix(sdpa): add MPS workarounds for bidirectional attention a |
|
Contains domain-specific terms and conci |
2026-06-16 |
| pr |
0.05 |
[docs] clarify initialization module usage |
|
Brief, domain-specific phrasing; no AI h |
2026-06-16 |
| pr |
0.05 |
add conversion scripts for EUPE |
|
Minimal content and review, domain abbre |
2026-06-16 |
| pr |
0.05 |
Laguna: support per-element output gating |
|
Natural domain-specific info, adds techn |
2026-06-16 |
| pr |
0.05 |
[CB] Hide activation footprint by using the CUDA graph pool |
|
Uses technical CUDA domain jargon and su |
2026-05-12 |
| pr |
0.05 |
Fix M-RoPE `inv_freq` device and `meta` → `to_empty` re-init |
|
Specific technical fixes, natural tone, |
2026-05-12 |
| pr |
0.05 |
[docs] ALMModelTest |
|
Brief, minimal doc PR, reviews are infor |
2026-05-11 |
| pr |
0.05 |
Fix dynamic module symlinked cache on trust_remote_code mode |
|
Technical, specific bug context, no AI p |
2026-06-13 |
| pr |
0.05 |
Delete docstring if same as in auto-doc |
|
Terse, technical, topic-specific and inf |
2026-05-29 |
| pr |
0.05 |
Add Nemotron ASR Streaming |
|
Informal, technical references, unpolish |
2026-06-01 |
| pr |
0.05 |
Widen tols for float16/bfloat16 |
|
Problem context, direct language, links; |
2026-05-18 |
| pr |
0.05 |
feat[vLLM x v5]: Expose max_source_positions on VibeVoiceAsr |
|
Uses domain marks, short, technical, and |
2026-06-07 |
| pr |
0.05 |
[PoC] HF exporters |
|
Edits, references, informal structure; l |
2025-11-03 |
| pr |
0.05 |
[docs] require trust_remote_code for custom_generate |
|
Minimal, terse, domain-specific phrasing |
2026-06-15 |
| pr |
0.05 |
Port ESMC and ESMFold2 to Transformers |
|
Personal, informal note and domain refer |
2026-06-04 |
| pr |
0.05 |
[Fix] Bug when loading audio data from an URI |
|
Specific bug and test context; clearly h |
2026-06-15 |
| commit |
0.00 |
add conversion scripts for EUPE (#46691) |
|
Commit message uses terse, domain-specif |
2026-06-16 |
| commit |
0.00 |
[`Blip2`] Fix cross attention reshape (#46695) |
|
Extremely terse, likely human and not AI |
2026-06-16 |
| commit |
0.00 |
[docs] compile level and batch/scheduling limits (#46676) |
|
Minimal, informal notes typical of human |
2026-06-16 |
| commit |
0.00 |
[Fix] Serve drops some attributes at serialization (#46680) |
|
Concise, natural phrasing and domain spe |
2026-06-16 |
| commit |
0.00 |
[blip_2] Support attn_implementation dispatch (#46401) |
|
Brief, informal commit messages indicati |
2026-06-16 |
| commit |
0.00 |
[CTRL] Support attn_implementation dispatch (#46073) |
|
Short, repeated, and refined messages sh |
2026-06-16 |
| commit |
0.00 |
Sapiens2: Remove modeling imports in image processing (#4667 |
|
Single-sentence, direct commit with doma |
2026-06-16 |
| commit |
0.00 |
feat(pipelines): accept numpy arrays and tensors in ImageCla |
|
Human names and terse fixes indicate nat |
2026-06-16 |
| commit |
0.00 |
Smovlm: pad videos up to max frames (#46662) |
|
Extremely terse commit notes, not AI-lik |
2026-06-16 |
| commit |
0.00 |
[Continuous Batching] Snapshot generation outputs without mu |
|
Direct, concise summary; typical human P |
2026-06-16 |
| commit |
0.00 |
Reduce per_page from 100 to 50 in GitHub API calls to avoid |
|
Commit message is terse and technical; n |
2026-06-15 |
| commit |
0.00 |
[docs] keep generation tensors on cpu (#46675) |
|
Very brief, domain-specific, and lacks A |
2026-06-15 |
| commit |
0.00 |
docs(zh): add Chinese translation of kernels.md (#46621) |
|
Standard template use, mentions translat |
2026-06-15 |
| commit |
0.00 |
mistral common backend fix (#46667) |
|
Informal tone, typos, and terse edits; h |
2026-06-15 |
| commit |
0.00 |
[pr template] update (#46606) |
|
Very short, informal, purely descriptive |
2026-06-15 |
| commit |
0.00 |
Fix AttributeError in auto_factory when model_class lacks co |
|
Technical, concise, fits human commit co |
2026-06-15 |
| commit |
0.00 |
[CB] Slice logits inside the model (#46660) |
|
Brief, casual, and technical language ty |
2026-06-15 |
| commit |
0.00 |
ci: add NO_COLOR=1 to suppress ANSI color codes in CI output |
|
Terse, technical, and informal commit st |
2026-06-15 |
| commit |
0.00 |
Fix dynamic RoPE not resetting inv_freq when layer_type is N |
|
Detailed technical reasoning with domain |
2026-06-15 |
| commit |
0.00 |
Better processing tests (#46374) |
|
Short, informal, and human-typical commi |
2026-06-15 |
| commit |
0.00 |
ci: add merge_group trigger to pr-ci-caller.yml (#46668) |
|
Standard commit with co-author, no AI si |
2026-06-15 |
| commit |
0.00 |
skip invalid quant_cache test for nemotron_h (#46368) |
|
Terse, domain-specific, human style |
2026-06-15 |
| commit |
0.00 |
Revert "Disable PR CI workflow for PRs from forked repo. dur |
|
Standard revert commit, no AI signals |
2026-06-15 |
| commit |
0.00 |
[CB] Fix seqlens and use TypedDict (#46593) |
|
Commit message is terse and technical, c |
2026-06-13 |
| commit |
0.00 |
Disable PR CI workflow for PRs from forked repo. during the |
|
Concise, domain-specific, standard commi |
2026-06-12 |
| commit |
0.00 |
Update post release (#46608) |
|
Casual tone, minimal content, no AI sign |
2026-06-12 |
| commit |
0.00 |
Fix `peft` lower bound (#46605) |
|
Domain abbreviation, concise, standard c |
2026-06-12 |
| commit |
0.00 |
Fix docstring formatting issues causing Sphinx autodoc warni |
|
Detailed technical description, natural |
2026-06-12 |
| commit |
0.00 |
Add minimax m3vl (#46600) |
|
Dense domain jargon, informal, detailed |
2026-06-12 |
| commit |
0.00 |
[CI] don't export OTELs within the tests (#46602) |
|
Informal, terse, domain abbreviation, hu |
2026-06-12 |
| commit |
0.00 |
[CI] capture checkers output in OTEL (#46601) |
|
Terse, domain-specific, casual tone |
2026-06-12 |
| commit |
0.00 |
put output_hidden_states into filter_output_hidden_states (# |
|
Technical, informal, multiple edits, hum |
2026-06-12 |
| commit |
0.00 |
a11 for checkers (#46599) |
|
Abbreviation, informal, no AI signals |
2026-06-12 |
| commit |
0.00 |
Fix stop string matching for byte-fragment tokens (#46530) |
|
Technical domain phrase, concise, human |
2026-06-12 |
| commit |
0.00 |
[DiffusionGemma] better docs and links (#46569) |
|
Brief commit log and changelog, domain-s |
2026-06-12 |
| commit |
0.00 |
Require `trust_remote_code` to run a local-directory `custom |
|
In-depth technical explanation, specific |
2026-06-12 |
| commit |
0.00 |
Fix torchaudio version not tied to torch version in docker f |
|
Extremely brief, informal, and includes |
2026-06-12 |
| commit |
0.00 |
PP-OCRv6: update documentation and slow tests (#46576) |
|
Only a terse summary; lacks AI hallmarks |
2026-06-12 |
| commit |
0.00 |
[CI] Enable PR CI for all fork PRs via security gate (#46591 |
|
Minimal free-text content, terse, clear |
2026-06-12 |
| commit |
0.00 |
[CB] [Minor] Add parameter to tune default compile level (#4 |
|
Bullet-point changelog, technical termin |
2026-06-12 |
| commit |
0.00 |
fix-trainer-tests (#46541) |
|
Very terse, fix-only commit, typical hum |
2026-06-11 |
| commit |
0.00 |
[CI] fix and make less flaky (#46543) |
|
Terse commit messages and code-context p |
2026-06-10 |
| commit |
0.00 |
Fix hf_hub_download not placing file in current dir for url_ |
|
Very brief commit summary and fix note; |
2026-06-10 |
| commit |
0.00 |
Only download hub cache for pipelines_torch job (#46544) |
|
Minimal, colloquial commit message with |
2026-06-10 |
| commit |
0.00 |
[docs] torchvision lanczos (#46528) |
|
Very brief, initialization and feedback; |
2026-06-10 |
| commit |
0.00 |
Add deepseek 3.2 exp (#41251) |
|
Detailed technical changelog using domai |
2026-06-10 |
| commit |
0.00 |
GPU go brr (#46540) |
|
Informal slang ('go brr') and casual phr |
2026-06-10 |
| commit |
0.00 |
[docs] 4D attention masks (#46507) |
|
Terse, customary docs commit phrasing; n |
2026-06-10 |
| commit |
0.00 |
Fix the offsets in processing (#46525) |
|
Very minimal ('!'); no sign of AI involv |
2026-06-10 |
| commit |
0.00 |
Fix buggy action sha pin (#46534) |
|
Single-word message and co-author; not A |
2026-06-10 |
| commit |
0.00 |
Fix trailing comma bug in DataCollatorForLanguageModeling ex |
|
Simple, specific bugfix message; lacks A |
2026-06-10 |
| commit |
0.00 |
Fix InternVL models (#46524) |
|
Very terse commit message, standard huma |
2026-06-10 |
| commit |
0.00 |
fix(afmoe): reduce tokens in test_compile_static_cache to av |
|
Brief commit, includes domain details an |
2026-06-09 |
| commit |
0.00 |
[CB] Add a "max_requests_per_batch" parameter (#46434) |
|
Terse checklist-style commit messages wi |
2026-06-09 |
| commit |
0.00 |
Extended & simplified n-to-1 kernel fusion via KernelConfig |
|
Human-like WIP commits, domain terms, it |
2026-06-09 |
| commit |
0.00 |
Speed up modular conversion by reusing process pool (#46438) |
|
Short, informal subject typical of a hum |
2026-06-09 |
| commit |
0.00 |
revamp cv docs and fix rf-detr (#46219) |
|
Domain-specific, many updates and peer-r |
2026-06-09 |
| commit |
0.00 |
Update hub metadata (#46379) |
|
Informal update comments, domain context |
2026-06-09 |
| commit |
0.00 |
extend DeepseekV4FlashIntegrationTest to non-cuda device (#4 |
|
Technical commit, standard co-authoring, |
2026-06-09 |
| commit |
0.00 |
[docs] deepgemm (#46361) |
|
Commit messages are terse, informal, and |
2026-06-08 |
| commit |
0.00 |
Fix `{image/video/audio}_token_ids` in `ProcessorMixin` (#46 |
|
Message is succinct and uses domain-spec |
2026-06-08 |
| commit |
0.00 |
[fix] regression introduced by #45534 (#46456) |
|
Very brief, casual commit fragments; no |
2026-06-08 |
| commit |
0.00 |
[docs] tp for continuous batching (#46019) |
|
Minimal, repetition likely auto-commit; |
2026-06-08 |
| commit |
0.00 |
Use torchvision's native LANCZOS interpolation instead of PI |
|
Short action-based notes; informal, not |
2026-06-08 |
| commit |
0.00 |
skip quant cache tests for deepseek v4 (#46362) |
|
Standard skip/test message with signatur |
2026-06-08 |
| commit |
0.00 |
revisit history parallel beam search tests to avoid unnecess |
|
All lines are informal with a domain-typ |
2026-06-08 |
| commit |
0.00 |
Add debugging info in `pr-ci-caller.yml` (#46505) |
|
Co-authored-by trailer, no AI assistance |
2026-06-08 |
| commit |
0.00 |
Fix tests: 'Cohere2MoeModel' object has no attribute 'hf_dev |
|
Commit log with code references and sign |
2026-06-08 |
| commit |
0.00 |
Triton finegrained fp8/fp4 (#46407) |
|
Informal short functional notes typical |
2026-06-08 |
| commit |
0.00 |
Fix dtype mismatch in NemotronH/Zamba2 Mamba2 CUDA-kernel pa |
|
Brief and technical, uses typical commit |
2026-06-08 |
| commit |
0.00 |
Bump the actions group across 1 directory with 19 updates (# |
|
— |
2026-06-08 |
| commit |
0.00 |
fix(falcon_mamba): recommend `pip install kernels` in fast-p |
|
Clear domain jargon, concise phrasing, n |
2026-06-08 |
| commit |
0.00 |
Log some information in `.github/workflows/pr-ci-post-dashbo |
|
Terse, domain-specific with co-author, n |
2026-06-08 |
| commit |
0.00 |
feat(quantizers): support non-weight param names in TorchAo |
|
Specific technical references, typical e |
2026-06-08 |
| commit |
0.00 |
fix: convert to fp32 first when doing image resize (#46262) |
|
Commit language terse, domain-specific, |
2026-06-08 |
| commit |
0.00 |
add XPU expectation for deepseek_ocr2 model tests (#46492) |
|
Minimal message; directly technical, hum |
2026-06-08 |
| commit |
0.00 |
Fix sapiens2 tests: add XPU device expectations (#46488) |
|
Brief, domain-specific—no AI stylistic p |
2026-06-08 |
| commit |
0.00 |
Add vLLM smoke test to CI (#46383) |
|
Technical, concise language with direct |
2026-06-08 |
| commit |
0.00 |
Fix: interleaved RoPE application for MLA and Support Index |
|
Commit is detailed, technical, and uses |
2026-06-08 |
| commit |
0.00 |
extend deepseek v4 test to xpu (#46366) |
|
Commit messages are terse and informal, |
2026-06-05 |
| commit |
0.00 |
Quantization for small models (#46449) |
|
Terse phrasing and explicit TODOs, typic |
2026-06-05 |
| commit |
0.00 |
Added cosmos3 model (#46146) |
|
Direct, sometimes incomplete messages, s |
2026-06-05 |
| commit |
0.00 |
ci: fix slack report job — use AWS runner and authenticated |
|
Short, informal, fix-oriented; no AI phr |
2026-06-05 |
| commit |
0.00 |
fbgemm_fp8:Keep the current device aligned with the input te |
|
Commit messages are direct and domain-sp |
2026-06-05 |
| commit |
0.00 |
[Modular] Add `no_inherit_decorators` and fixup wrong RoPE r |
|
Informal, patch-focused commit statement |
2026-06-05 |
| pr |
0.00 |
[DO NOT merge!] test migration |
|
Extremely terse test-related PR, typical |
2026-05-20 |
| pr |
0.00 |
[docs] decode fast path |
|
Short, direct, domain-specific content, |
2026-05-11 |
| pr |
0.00 |
Fix inverted is_tracing condition causing FA2 long-context d |
|
Contains technical jargon and natural bu |
2026-06-16 |
| pr |
0.00 |
DiffusionGemma: mask layout and CI |
|
WIP note, technical content, informal ph |
2026-06-15 |
| pr |
0.00 |
fix(cli/serve): import serve handlers lazily so the CLI work |
|
Concise, technical, specific to the repo |
2026-06-07 |
| pr |
0.00 |
[`Blip2`] Fix cross attention reshape |
|
Uses shorthand and technical reference, |
2026-06-16 |
| pr |
0.00 |
wip step 3 7 flash |
|
Extremely terse, minimal, typical of hum |
2026-06-15 |
| pr |
0.00 |
[docs] compile level and batch/scheduling limits |
|
Terse changelog style, domain-specific l |
2026-06-15 |
| pr |
0.00 |
Add Videoprism |
|
Technical with informal notes and review |
2025-08-04 |
| pr |
0.00 |
[fix] enable base64 str audio in load_audio |
|
Very brief, typical for a small fix upda |
2026-06-16 |
| pr |
0.00 |
:rotating_light: [`Kernels`] Sync to latest version |
|
Technical, personalized, direct; not AI- |
2026-05-18 |
| pr |
0.00 |
Add xcodec2 model |
|
Reopening past PR, TODO list is informal |
2026-02-20 |
| pr |
0.00 |
Add Gefen optimizer support to Trainer |
|
Contains domain jargon and abbreviations |
2026-06-16 |
| pr |
0.00 |
Qwen3 ASR and Forced Aligner |
|
Uses checklist, repo-specific info, tech |
2026-02-08 |
| pr |
0.00 |
fix: reorder cache for all cache key names during beam searc |
|
Technical description, domain jargon, in |
2026-06-15 |
| pr |
0.00 |
Update tokenizer mappings to use TokenizersBackend for addit |
|
Informal, uses abbr. and directs to othe |
2026-05-20 |
| pr |
0.00 |
[blip_2] Support attn_implementation dispatch |
|
Extremely terse description, confirmed h |
2026-06-04 |
| pr |
0.00 |
Lfm2: also thread `seq_idx` through ShortConv.slow_forward ( |
|
Technical follow-up, domain abbreviation |
2026-06-13 |
| pr |
0.00 |
Add minimax m3vl |
|
Minimal human-written summary; review ev |
2026-06-12 |
| pr |
0.00 |
Add Molmo2 |
|
No free-text content—likely a placeholde |
2026-01-23 |
| pr |
0.00 |
[Continuous Batching] Snapshot generation outputs without mu |
|
Technical, concise, domain-specific, hum |
2026-06-15 |
| pr |
0.00 |
Tp param level |
|
Bulleted, terse, lots of domain-specific |
2026-05-29 |
| pr |
0.00 |
simplify fsdp |
|
Very terse and minimal, typical human st |
2026-06-15 |
| pr |
0.00 |
separate cpu ofloadding and mixed precision tests into mocki |
|
Short, fragmentary, informal style; no A |
2026-05-26 |
| pr |
0.00 |
sp + ep training / tp + ep inference |
|
Highly technical, bullet points, not AI- |
2026-05-29 |
| pr |
0.00 |
Reduce per_page from 100 to 50 in GitHub API calls to avoid |
|
Technical summary, contains human-like i |
2026-06-15 |
| pr |
0.00 |
Add MiniCPM3 |
|
Detailed, domain-specific, uses human ph |
2025-09-24 |
| pr |
0.00 |
feat: bump safetensors to `0.8.0` |
|
Uses casual tone with domain abbreviatio |
2026-06-09 |
| pr |
0.00 |
feat(serve): add OTEL |
|
Straightforward, domain-specific, no AI- |
2025-09-05 |
| pr |
0.00 |
[docs] keep generation tensors on cpu |
|
Short, direct, human conversational revi |
2026-06-15 |
| pr |
0.00 |
Fix `PhimoeIntegrationTest` |
|
Brief, pragmatic language; human review |
2026-06-10 |
