| commit |
1.00 |
Fix possessive typo in Whisper long-form warning (#48877) |
|
Commit message contains explicit AI assi |
2026-09-18 |
| commit |
1.00 |
Standardise Aria's MoE onto the experts interface (#48907) |
|
Commit message contains explicit AI assi |
2026-09-18 |
| commit |
1.00 |
Clean up some MoE models' integration tests (#48833) |
|
Commit message contains explicit AI assi |
2026-09-17 |
| commit |
1.00 |
Mark test_generate_with_static_cache as flaky for olmo and b |
|
Commit message contains explicit AI assi |
2026-09-17 |
| commit |
1.00 |
[Chat Parsing] Coerce oneOf tool arguments (#48719) |
|
Commit message contains explicit AI assi |
2026-09-17 |
| commit |
1.00 |
Make LongcatFlashConfig self-consistent without building the |
|
Commit message contains explicit AI assi |
2026-09-17 |
| commit |
1.00 |
Fix GLM rope_parameters dict shared mutation across test ins |
|
Commit message contains explicit AI assi |
2026-09-17 |
| commit |
1.00 |
Fix Glm4MoeIntegrationTest: split into 3 tests, switch to GL |
|
Commit message contains explicit AI assi |
2026-09-16 |
| commit |
1.00 |
Fix flaky RfDetr test_save_load (#48869) |
|
Commit message contains explicit AI assi |
2026-09-16 |
| commit |
1.00 |
Fix ESMFold2 ligand iPLDDT weighting: mol_type non-polymer c |
|
Commit message contains explicit AI assi |
2026-09-16 |
| commit |
1.00 |
docs: fix broken #combining-with-fsdp2 anchor in expert_para |
|
Commit message contains explicit AI assi |
2026-09-16 |
| commit |
1.00 |
Detect nested rope_parameters without relying on layer_types |
|
Commit message contains explicit AI assi |
2026-09-15 |
| commit |
1.00 |
Relax static-cache tolerance in test_generate_with_static_ca |
|
Commit message contains explicit AI assi |
2026-09-15 |
| commit |
1.00 |
Pin utf-8 in test_can_init_all_missing_weights source read ( |
|
Commit message contains explicit AI assi |
2026-09-15 |
| commit |
1.00 |
Add integration tests for MuseGlimmerAssistantModel (#48796) |
|
Commit message contains explicit AI assi |
2026-09-14 |
| commit |
1.00 |
Fix hidden state selection on Gemma4 assistant's first prefi |
|
Commit message contains explicit AI assi |
2026-09-14 |
| commit |
1.00 |
Fix Glm4vMoeIntegrationTest: offload_folder + MemoryCleanupM |
|
Commit message contains explicit AI assi |
2026-09-14 |
| commit |
1.00 |
Switch daily CI to torch 2.14 — update expected outputs (#48 |
|
Commit message contains explicit AI assi |
2026-09-13 |
| commit |
1.00 |
[DeepseekV3] OOM cascade root-cause investigation (generator |
|
Commit message contains explicit AI assi |
2026-09-11 |
| commit |
1.00 |
Fix D-FINE / RT-DETR main loss being computed over the denoi |
|
Commit message contains explicit AI assi |
2026-09-11 |
| commit |
1.00 |
Synthetic test assets (#48589) |
|
Commit message contains explicit AI assi |
2026-09-10 |
| commit |
1.00 |
[MusicgenMelody] Fix conditioning silently dropped at genera |
|
Commit message contains explicit AI assi |
2026-09-10 |
| pr |
1.00 |
docs(optimization): standardize 'Return:' to 'Returns:' in s |
|
PR body explicitly mentions AI collabora |
2026-09-20 |
| pr |
1.00 |
[generate] Opt-in normalized repetition penalty |
|
PR body explicitly mentions AI collabora |
2026-09-09 |
| pr |
1.00 |
Add dots3-note Preview model support |
|
PR body explicitly mentions AI collabora |
2026-08-08 |
| pr |
1.00 |
Add LLaVA-OneVision-1.5 model |
|
PR body explicitly mentions AI collabora |
2026-09-17 |
| pr |
1.00 |
Fix stale _added_tokens_encoder entries in cpmant and wav2ve |
|
PR body explicitly mentions AI collabora |
2026-07-20 |
| pr |
1.00 |
Persist the Whisper spelling normalizer across save_pretrain |
|
The description explicitly states that A |
2026-08-20 |
| pr |
1.00 |
Fix Windows path separators breaking relative imports of Hub |
|
The description explicitly says it was g |
2026-09-19 |
| pr |
1.00 |
Fix Windows path handling in repo checkers (check_auto, chec |
|
The description explicitly says it was g |
2026-09-19 |
| pr |
1.00 |
Pin utf-8 in add-new-model-like file IO (Windows non-UTF-8 l |
|
The description explicitly says it was g |
2026-09-19 |
| pr |
1.00 |
Pin utf-8 in test source reads (Windows non-UTF-8 locale fix |
|
Explicitly states it was generated with |
2026-09-19 |
| pr |
1.00 |
Restore the Unicode whitespace set in the GPT-SW3 tokenizer |
|
PR body explicitly mentions AI collabora |
2026-09-17 |
| pr |
1.00 |
Add lm_head tp_plan to tied Gemma multimodal ForConditionalG |
|
PR body explicitly mentions AI collabora |
2026-09-18 |
| pr |
1.00 |
tests: skip the symlinked hub-cache test when symlinks are u |
|
PR body explicitly mentions AI collabora |
2026-09-05 |
| pr |
1.00 |
Fix vibevoice TTS batched audio index |
|
Explicitly states that Claude helped wri |
2026-09-17 |
| pr |
1.00 |
docs: fix dead doc links in i18n READMEs and the xlnet docst |
|
PR body explicitly mentions AI collabora |
2026-09-17 |
| pr |
1.00 |
DOC: Improve documentation for remap_legacy_layer_types func |
|
PR body explicitly mentions AI collabora |
2026-09-09 |
| pr |
1.00 |
docs: fix docstring parameters that do not match signatures |
|
PR body explicitly mentions AI collabora |
2026-09-17 |
| pr |
1.00 |
Finegrained kernels: unified quantization kernels api |
|
PR body explicitly mentions AI collabora |
2026-08-18 |
| pr |
1.00 |
Skip an unnecessary image copy in the torchvision image norm |
|
PR body explicitly mentions AI collabora |
2026-09-17 |
| pr |
1.00 |
Fix possessive typo in Whisper long-form warning |
|
Explicitly states it was made with Curso |
2026-09-16 |
| pr |
1.00 |
Resolve the Hub revision once per load instead of passing a |
|
PR body explicitly mentions AI collabora |
2026-07-28 |
| pr |
1.00 |
Standardise Aria's MoE onto the experts interface |
|
The description explicitly says it was g |
2026-09-17 |
| pr |
1.00 |
[perf] Qwen3.5-9B full fine-tuning with FSDP2 on 8x B200: 0. |
|
PR body explicitly mentions AI collabora |
2026-09-10 |
| pr |
1.00 |
[perf] Qwen3.5: fused RMSNorm kernels from flash-linear-atte |
|
PR body explicitly mentions AI collabora |
2026-09-18 |
| pr |
0.82 |
fix(luke): set post-processor to add |
|
Highly structured prose and reviewer fee |
2026-09-06 |
| commit |
0.80 |
QA: Added a `MemoryCleanupMixin` class for tests (#48681) |
|
The text explicitly refers to an agent a |
2026-09-11 |
| pr |
0.40 |
fix: resolve #48964 - Issue doing PEFT on Embedding Gemma w/ |
|
Generic verification language lacks conc |
2026-09-20 |
| pr |
0.30 |
Add support for BharatGen's Param2MoE model architecture |
|
Some generic promotional wording appears |
2026-06-04 |
| pr |
0.30 |
Fix: Preserve non-persistent buffer values when moving missi |
|
Technical and specific, though unusually |
2026-09-18 |
| pr |
0.28 |
Improve lazy import error messages |
|
Polished explanatory prose and exhaustiv |
2026-09-07 |
| pr |
0.25 |
fix(vibevoice-asr): use integer ceiling division for audio t |
|
Very polished explanation, though the pr |
2026-09-16 |
| pr |
0.25 |
Add BerryLM |
|
Detailed, domain-specific technical pros |
2026-09-17 |
| pr |
0.18 |
Add qwen3 tts |
|
Structured model summary is polished but |
2026-03-07 |
| pr |
0.18 |
Feat: Add Kimi K3 architecture support |
|
Detailed architecture discussion is doma |
2026-09-17 |
| pr |
0.18 |
Use unique request IDs for continuous batching |
|
Detailed security report with concrete r |
2026-09-18 |
| pr |
0.16 |
Fix SentencePieceBackend.convert_tokens_to_string dropping b |
|
Polished technical explanation, but grou |
2026-08-08 |
| commit |
0.08 |
Support for MoE in the GGUF integration (#48529) |
|
Detailed domain terminology is polished |
2026-09-15 |
| pr |
0.08 |
[Model] Add PP-DocLayoutV4 Model Support |
|
Highly specific implementation terminolo |
2026-08-28 |
| pr |
0.08 |
Fix SwitchTransformers Top1 router: raw logits, expert capac |
|
Highly specific debugging details and te |
2026-08-30 |
| pr |
0.08 |
Add heterogeneous model support (per-layer config and modeli |
|
Detailed domain-specific explanation, bu |
2026-04-09 |
| pr |
0.08 |
Add Ovis2.5 model support |
|
Standard model-support summary with repo |
2026-07-30 |
| pr |
0.08 |
FSDP x TP Saving / Loading |
|
Technical detail and grammatical errors |
2026-09-14 |
| pr |
0.08 |
feat(hf_argparser): support dict-typed dataclass fields from |
|
Concise technical explanation with issue |
2026-09-05 |
| pr |
0.08 |
[`MLA`] Add native FA4 support |
|
Terse, domain-specific description with |
2026-09-16 |
| pr |
0.08 |
FSDP x TP Saving / Loading, Planner / Writer based solution |
|
Detailed but strongly domain-specific en |
2026-09-15 |
| pr |
0.08 |
Fix quantized cache |
|
Informal phrasing, typos, and detailed d |
2026-09-11 |
| pr |
0.06 |
[HunYuanVL] Stop rewriting the static NTK-alpha RoPE frequen |
|
Highly specific technical diagnosis and |
2026-09-20 |
| pr |
0.06 |
Fix special token erasure in align_special_tokens |
|
Detailed bug analysis and architecture-s |
2026-09-07 |
| pr |
0.06 |
Align special tokens on the text config |
|
Specific repository behavior and concise |
2026-09-16 |
| pr |
0.06 |
feat: use safetensors prefetch api |
|
Brief, highly specific performance note |
2026-09-18 |
| commit |
0.05 |
Keep the expert-parallel sentinel slots out of the router gr |
|
Specific implementation details and info |
2026-09-11 |
| pr |
0.05 |
Fix module aliases breaking unittest via __warningregistry__ |
|
Specific failure details and informal ph |
2026-09-20 |
| pr |
0.05 |
Fix T5 SentencePiece load when precompiled_charsmap is empty |
|
Specific reproduction details and implem |
2026-09-20 |
| pr |
0.05 |
Fix SuperPoint torch.export by replacing data-dependent shap |
|
Specific implementation details and tech |
2026-09-19 |
| pr |
0.05 |
Honor `config.output_router_logits` in the MoE VLM wrappers |
|
Architecture-specific reasoning and nuan |
2026-09-17 |
| pr |
0.05 |
Fix: SentencePieceBackend.convert_tokens_to_string dropping |
|
Short restoration note with no AI-style |
2026-09-18 |
| commit |
0.04 |
Fix sliding window cache when assistant model calls generate |
|
Detailed reproducible test plan and cach |
2026-09-17 |
| commit |
0.04 |
Fix slow integration tests on XPU (#48611) |
|
Detailed XPU diagnostics and measurement |
2026-09-11 |
| pr |
0.04 |
Fix RecurrentGemma compiled generation with StaticCache |
|
Concise explanation contains specific im |
2026-09-20 |
| pr |
0.04 |
Better attn default for gguf |
|
Brief description includes natural wordi |
2026-09-18 |
| pr |
0.04 |
Support per-layer cache configuration and attention-mask sel |
|
Specific implementation scope and review |
2026-08-21 |
| pr |
0.04 |
Modernize ConvNextV2, ResNet, RegNet, RTDetrResnet, SwinV2, |
|
Informal phrasing, typos, and concrete p |
2026-08-27 |
| pr |
0.04 |
Fix tests due to dropping attn mask |
|
Specific failures, abbreviations, typos, |
2026-09-17 |
| pr |
0.04 |
Fix image processor class-level size mutation and min_pixels |
|
Highly domain-specific bug analysis and |
2026-09-18 |
| commit |
0.03 |
Fix docstring argument names that don't match signatures (#4 |
|
Clear, domain-specific documentation fix |
2026-09-17 |
| commit |
0.03 |
Don't fail model loading when the accelerator can't report f |
|
Detailed backend-specific diagnosis and |
2026-09-17 |
| commit |
0.03 |
Fix `StaticCache` for Mllama and enable `torch.compile` (#48 |
|
Detailed benchmark and implementation co |
2026-09-17 |
| commit |
0.03 |
Fix silently random-initializing `RTDetrModel`/`SEWDForCTC` |
|
Detailed technical explanation uses spec |
2026-09-14 |
| commit |
0.03 |
add HyperClovaX Vision (#44314) |
|
Terse, varied commit notes and explicit |
2026-09-11 |
| commit |
0.03 |
Skip the expert-parallel sentinel masking when expert parall |
|
Detailed domain-specific changelog with |
2026-09-10 |
| commit |
0.03 |
Fix deepspeed ci (#48640) |
|
Highly specific CI diagnosis and fixes i |
2026-09-10 |
| pr |
0.03 |
Add opt-in strict checkpoint loading to from_pretrained |
|
Terse, informal, issue-linked wording is |
2026-09-20 |
| pr |
0.03 |
fix-mask-skip-tracing |
|
Brief technical description and informal |
2026-09-18 |
| pr |
0.03 |
Fix Llama left-padding position IDs during training |
|
Specific behavioral details and test ref |
2026-09-19 |
| pr |
0.03 |
Let continuous batching page the call, not the model |
|
Informal, highly specific engineering di |
2026-09-12 |
| pr |
0.03 |
Fix AXK2 integration test: update CUDA expected text and ren |
|
Specific test names, numerical tolerance |
2026-09-18 |
| pr |
0.03 |
fix(opt): preserve attention_mask=None in create_causal_mask |
|
Specific technical diagnosis and testing |
2026-09-18 |
| pr |
0.03 |
Translate audio classification docs to Chinese |
|
Short, straightforward summary with no A |
2026-09-17 |
| pr |
0.03 |
🚨🚧 FeatureExtractor → AudioProcessor |
|
Minimal template content and informal wo |
2026-03-02 |
| pr |
0.03 |
Fix OPT torch.compile SDPA fallback when attention_mask is N |
|
Concise, issue-specific technical descri |
2026-09-18 |
| pr |
0.03 |
Fix assistant masks for processors |
|
Informal phrasing and concrete implement |
2026-09-14 |
| pr |
0.03 |
Modular conversion small fix |
|
Minimal free text and casual wording sug |
2026-09-18 |
| pr |
0.03 |
Restrict remote media URLs in serve |
|
Specific security details, reproduction |
2026-09-18 |
| commit |
0.02 |
Align special tokens on the text config (#48847) |
|
Detailed domain-specific diagnosis and i |
2026-09-18 |
| commit |
0.02 |
DOC: Improve documentation for remap_legacy_layer_types func |
|
Commit history is terse and references R |
2026-09-18 |
| commit |
0.02 |
docs: fix docstring parameters that do not match signatures |
|
Uses precise parameter and class names w |
2026-09-18 |
| commit |
0.02 |
Size the device_map buffer from the largest leaf module (#47 |
|
Specific implementation details and sign |
2026-09-18 |
| commit |
0.02 |
🚨 [vLLM] Fix video token counting for Transformers backend v |
|
Terse commit history with specific autho |
2026-09-17 |
| commit |
0.02 |
Always materialize the causal mask in Doge so sdpa stays cau |
|
Brief iterative commits and specific imp |
2026-09-17 |
| commit |
0.02 |
Fix Idefics2 padding when the first example has no image (#4 |
|
Specific processor and CI fixes with ter |
2026-09-17 |
| commit |
0.02 |
[`feat`] Allow untying `hidden_states[-1]` from `last_hidden |
|
Technical feature description and focuse |
2026-09-17 |
| commit |
0.02 |
Fix DFlash sampled candidates losing the batch dimension (#4 |
|
Detailed domain-specific fix and normal |
2026-09-17 |
| commit |
0.02 |
Add test for causal only variant of some encoder-decoder mod |
|
Technical, repository-specific changelog |
2026-09-16 |
| commit |
0.02 |
Fix `reset` on the dynamic cache layers (#48809) |
|
Specific implementation and testing note |
2026-09-16 |
| commit |
0.02 |
[GPTNeoXJapanese] Fix RoPE ignoring partial_rotary_factor (# |
|
Domain-specific explanation includes inf |
2026-09-16 |
| commit |
0.02 |
Qwen3.8 GGUF (#48660) |
|
Terse, fragmentary changelog language is |
2026-09-15 |
| commit |
0.02 |
[CI] Add link checker (#48160) |
|
Brief informal bullets and concrete CI r |
2026-09-15 |
| commit |
0.02 |
[docs] gguf (#46357) |
|
Minimal, domain-specific commit notes ar |
2026-09-15 |
| commit |
0.02 |
[Fix] Clean-up ternaries in the DeepSeek family (#48447) |
|
Fragmentary review notes and repository- |
2026-09-15 |
| commit |
0.02 |
Fix `AutoModel.from_pretrained` not restoring `modules_to_sa |
|
Specific PEFT implementation details ind |
2026-09-14 |
| commit |
0.02 |
Bump `peft` version requirement (#48716) |
|
Specific versioning rationale and reposi |
2026-09-14 |
| commit |
0.02 |
Cast pixel values to the patch embedding dtype in DeepSeek-O |
|
Detailed reproduction and implementation |
2026-09-14 |
| commit |
0.02 |
Fix TextToAudioPipeline crash for tokenizer-only models (#48 |
|
Concrete error message and API behavior |
2026-09-14 |
| commit |
0.02 |
[InternVL] Normalize num_patches before np.cumsum (#48469) |
|
Terse, domain-specific commit message wi |
2026-09-14 |
| commit |
0.02 |
Enable FSDP2 + expert parallelism via a 2-D (fsdp, tp) devic |
|
Detailed domain-specific changelog with |
2026-09-12 |
| commit |
0.02 |
Add flash_attention_4 in attn_implementation AutoModel docst |
|
Short, task-specific commit messages sho |
2026-09-11 |
| commit |
0.02 |
Fix Qwen2.5-VL temporal RoPE for fractional video intervals |
|
Concise technical commit messages contai |
2026-09-11 |
| commit |
0.02 |
[Fix] yolos offload issue (#48688) |
|
Terse, informal commit wording with a sp |
2026-09-10 |
| pr |
0.02 |
This PR generalizes continuous batching graph support from C |
|
Specific implementation details and revi |
2026-08-27 |
| pr |
0.02 |
Allow image_transforms.resize on normalized (out-of-range) i |
|
Concrete bug context, test cases, and co |
2026-09-20 |
| pr |
0.02 |
Skip mask creation when padding_mask is None under torch.com |
|
Brief and issue-specific wording strongl |
2026-09-18 |
| pr |
0.02 |
Add native LFM2-Audio support |
|
Detailed model-specific parity results a |
2026-08-24 |
| pr |
0.02 |
Fix additional_special_tokens data loss with extra_special_t |
|
Concise, project-specific wording and ca |
2026-08-09 |
| pr |
0.02 |
Return router logits from every MoE model that computes them |
|
Extensive model-specific technical refer |
2026-09-18 |
| pr |
0.02 |
[mLUKE] Un-skip MLukeTokenizerTest |
|
Highly specific measurements, commands, |
2026-08-22 |
| pr |
0.02 |
logits_to_keep bool mask for all ForXXX classes, expose last |
|
Title-only contribution provides no mean |
2026-09-15 |
| pr |
0.02 |
Warm the page cache before sharded loading (opt-in via `HF_S |
|
Detailed filesystem benchmarks and infor |
2026-08-23 |
| commit |
0.01 |
Fix AXK2 integration test: update CUDA expected text and ren |
|
Terse, specific test changes and a norma |
2026-09-18 |
| commit |
0.01 |
docs: fix dead doc links in i18n READMEs and the xlnet docst |
|
Contains concrete repository paths, symb |
2026-09-18 |
| commit |
0.01 |
Improve lazy import error messages (#48602) |
|
Incremental commit titles and specific t |
2026-09-18 |
| commit |
0.01 |
fix(vibevoice-asr): use integer ceiling division for audio t |
|
Precise numerical, dtype, and implementa |
2026-09-18 |
| commit |
0.01 |
Update arXiv citation in NeoMME model doc (#48926) |
|
Brief, concrete documentation change wit |
2026-09-18 |
| commit |
0.01 |
[`DSA`] Only save latents on dsa with indexer as well (#4887 |
|
Informal terse updates and a concrete te |
2026-09-17 |
| commit |
0.01 |
Add auto_docstring task model overrides (#47815) |
|
Short iterative commit messages are char |
2026-09-17 |
| commit |
0.01 |
[Improvement] Rework the docstring and comments of mHC (#488 |
|
Extremely terse review-oriented commits |
2026-09-17 |
| commit |
0.01 |
[CB] Add pause mechanism (#48462) |
|
Terse iterative commit notes and typo in |
2026-09-17 |
| commit |
0.01 |
Fix Pixtral image processor do_pad option (#48872) |
|
Brief, specific fix description with a s |
2026-09-17 |
| commit |
0.01 |
Fix Minimax M2 partial_rotary_factor by remapping legacy rot |
|
Highly domain-specific RoPE details and |
2026-09-17 |
| commit |
0.01 |
Fix kosmos (#48853) |
|
Extremely terse fix message, with no AI- |
2026-09-16 |
| commit |
0.01 |
[generate] Make all methods and logits processors agnostic t |
|
Terse technical notes contain a typo and |
2026-09-16 |
| commit |
0.01 |
QA: deactivate rule 41 (#48852) |
|
Brief conventional QA commit message. |
2026-09-16 |
| commit |
0.01 |
🚨 🚨Bring some dinos to modern standards (#46266) |
|
Typos, shorthand, and iterative developm |
2026-09-15 |
| commit |
0.01 |
[apply_chat_template] pass sampling_rate to call (#48794) |
|
Terse commits and human co-author traile |
2026-09-15 |
| commit |
0.01 |
[generate] Drop attention mask early without padding (#48814 |
|
Informal typos and terse debugging notes |
2026-09-15 |
| commit |
0.01 |
[utils] Add MUSA support for Flash Attention 2 (#48612) |
|
Specific implementation changes and co-a |
2026-09-15 |
| commit |
0.01 |
:rotating_light: [`Kernels`] Bump version (#48714) |
|
Very terse fixes with an informal aside |
2026-09-15 |
| commit |
0.01 |
Fix decoder only path for older bert variants (#48785) |
|
Terse bullets and typos strongly indicat |
2026-09-15 |
| commit |
0.01 |
[docs] mrope and axial rope (#48717) |
|
Terse commit-style notes with no AI phra |
2026-09-14 |
| commit |
0.01 |
Fix kosmos flaky test (#48800) |
|
Brief informal wording and issue-specifi |
2026-09-14 |
| commit |
0.01 |
Restore legacy tensor-parallel initialization for compatibil |
|
Concise compatibility changelog with dom |
2026-09-14 |
| commit |
0.01 |
add kernel mapping entry for RMSNormGated, KDA, Conv1D on XP |
|
Commit history, technical jargon, and si |
2026-09-14 |
| commit |
0.01 |
Fix inkling embedding norm (#48786) |
|
Terse informal notes with model-specific |
2026-09-14 |
| commit |
0.01 |
Kernel api doc, part 2 (#46889) |
|
Commit-style documentation updates and h |
2026-09-14 |
| commit |
0.01 |
fix videomae load error (#48675) |
|
Terse iterative commit notes and standar |
2026-09-14 |
| commit |
0.01 |
unifying device_mesh init to enable PP + TP inference (#4815 |
|
Terse merge and cleanup commit messages |
2026-09-12 |
| commit |
0.01 |
Use huggingface_hub httpx export (#48685) |
|
Brief, repetitive commit message with no |
2026-09-12 |
| commit |
0.01 |
[CI] Replace hardcoded username allowlists with author_assoc |
|
A concise CI change description with a n |
2026-09-11 |
| commit |
0.01 |
[`MoE`] Fix eager EP (#48653) |
|
Extremely terse commit message with no A |
2026-09-10 |
| pr |
0.01 |
[`distributed`] Add expert-parallel token dispatch, default |
|
Informal explanation, typos, and concret |
2026-09-16 |
| pr |
0.01 |
[`distributed`]: Fix DeepSeek-V4 tied embeddings with combin |
|
Specific DeepSeek and tensor-parallel be |
2026-09-15 |
| pr |
0.01 |
Fix typos in DeepseekV4 comments |
|
Terse, specific typo fixes are character |
2026-09-19 |
| pr |
0.01 |
[Umbrella] FSDP2 + expert-parallel token dispatch, with shar |
|
Terse, informal release-branch notes wit |
2026-08-22 |
| pr |
0.01 |
Add MOSS-Transcribe-Diarize model |
|
Contains only a title and automated revi |
2026-08-28 |
| pr |
0.01 |
fix: support NPU/XPU compute stream in ContinuousBatchingSta |
|
Concise, domain-specific explanation wit |
2026-09-18 |
| pr |
0.01 |
[DON'T MERGE] debug mem issues |
|
Informal debug summary with concrete CI |
2026-09-17 |
| pr |
0.01 |
[`distributed`] decoupled tp_plan and ep_plan |
|
Technical rationale includes informal wo |
2026-09-16 |
| pr |
0.01 |
[`distributed`] Add dense and expert device mesh |
|
Contains domain-specific mesh details an |
2026-09-16 |
| pr |
0.01 |
Add Plain-DETR |
|
Extremely terse issue reference with no |
2025-03-28 |
| pr |
0.01 |
Enable new trf rules |
|
Informal phrasing and grammatical errors |
2026-08-10 |
| commit |
0.00 |
Fix tests due to dropping attn mask (#48903) |
|
Highly terse, informal commit fragments |
2026-09-18 |
| commit |
0.00 |
Modular conversion small fix (#48934) |
|
Extremely terse commit messages with inf |
2026-09-18 |
| commit |
0.00 |
Fix compile cache tests (#48923) |
|
Terse, informal fix description with no |
2026-09-18 |
| commit |
0.00 |
Fix T5 tied weights order for lm_head (#48238) |
|
Brief, issue-focused title with no AI in |
2026-09-18 |
| commit |
0.00 |
[serge] Fix 3 integration tests for model `deepseek_vl` fail |
|
— |
2026-09-17 |
| commit |
0.00 |
fix(ci): harden GitHub Actions workflows (#48160) (#48834) |
|
— |
2026-09-16 |
| commit |
0.00 |
[serge] Fix 12 integration tests for model `edgetam` failing |
|
— |
2026-09-14 |
| commit |
0.00 |
[serge] Fix 1 integration test for model `flex_olmo` failing |
|
— |
2026-09-11 |
| commit |
0.00 |
[serge] Fix 2 integration tests for model `mistral` failing |
|
— |
2026-09-11 |
| commit |
0.00 |
[serge] Fix 2 integration tests for model `minimax` failing |
|
— |
2026-09-10 |
| pr |
0.00 |
[serge] Fix 2 integration tests regressed by commit bcb9849a |
|
— |
2026-09-19 |
| pr |
0.00 |
[serge] Fix 2 integration tests for model `moshi` failing wi |
|
— |
2026-09-19 |
| pr |
0.00 |
[`distributed`]: adapt Trainer to work Expert Parallel |
|
Very terse informal status update. |
2026-09-16 |
| pr |
0.00 |
[serge] Fix 2 integration tests for model `llava_next_video` |
|
— |
2026-09-19 |
| pr |
0.00 |
Skip symlink-dependent tests when symlink creation is unavai |
|
Only a concise, issue-linked title is pr |
2026-09-18 |
| pr |
0.00 |
[serge] Fix 2 integration tests for model `nemotron` failing |
|
— |
2026-09-19 |
| pr |
0.00 |
support byte fallback for loading SPM Backend tok |
|
Very terse, informal tokenizer note with |
2026-09-18 |
| pr |
0.00 |
[Enhancement] Get rid of "paged|" implementations |
|
Extremely terse draft title and status a |
2026-09-18 |
| pr |
0.00 |
luke fixe |
|
Typos, terse wording, and a specific fix |
2026-08-29 |
