| commit |
1.00 |
Fix several bugs in `cache_implementation=static` (#46446) |
|
Commit message contains explicit AI assi |
2026-06-19 |
| commit |
1.00 |
Update tokenizer mappings to use TokenizersBackend for addit |
|
Commit message contains explicit AI assi |
2026-06-19 |
| pr |
1.00 |
Fix `TrackioCallback` fails to log evaluation metrics after |
|
PR body explicitly mentions AI collabora |
2026-06-27 |
| pr |
1.00 |
Add attention mask skip tests, tokenizer padding edge case t |
|
PR body explicitly mentions AI collabora |
2026-06-27 |
| pr |
1.00 |
Fix RagTokenizer attribute delegation |
|
PR body explicitly mentions AI collabora |
2026-06-26 |
| pr |
1.00 |
Fix torch.export compatibility for Mixtral MoE experts |
|
PR body explicitly mentions AI collabora |
2026-06-27 |
| pr |
1.00 |
Add Unlimited OCR |
|
PR body explicitly mentions AI collabora |
2026-06-23 |
| pr |
1.00 |
Adapt voxtral_realtime to mistral-common 1.11.5 audio API |
|
PR body explicitly mentions AI collabora |
2026-06-26 |
| pr |
1.00 |
docs: add prepare_for_model() v4→v5 migration example |
|
PR body explicitly mentions AI collabora |
2026-06-26 |
| pr |
1.00 |
[Mistral] Move MistralConverter into integrations/mistral/ p |
|
PR body explicitly mentions AI collabora |
2026-06-12 |
| pr |
1.00 |
Validate shard filenames in checkpoint index to prevent path |
|
PR body explicitly mentions AI collabora |
2026-06-26 |
| pr |
1.00 |
🚨 EP: fix EP router contract for many models + honor FP8 sca |
|
PR body explicitly mentions AI collabora |
2026-06-22 |
| pr |
1.00 |
feat(model): add bailing v2.6 model |
|
PR body explicitly mentions AI collabora |
2026-06-17 |
| pr |
0.70 |
[CB] Changes to increase max_batch_tokens |
|
Uses 'This PR aims at...' and formal per |
2026-06-17 |
| commit |
0.20 |
Fix use_cache with seq_len > 1 ( #46032) (#46084) |
|
Technical jargon, informal notes, not ov |
2026-06-23 |
| commit |
0.20 |
fix: raise `ValueError` for empty conversation in `apply_cha |
|
Direct explanation, technical, not overl |
2026-06-23 |
| commit |
0.20 |
Fix VideoPrismForVideoClassification returning last_hidden_s |
|
Detailed technical content with minor im |
2026-06-23 |
| commit |
0.20 |
Add MiniCPM3 (#41116) |
|
Domain-specific details, fragmented, not |
2026-06-23 |
| pr |
0.20 |
Add TIPSv2 |
|
Technical detail with concise and straig |
2026-06-02 |
| pr |
0.20 |
Add GB10/SM121 Hub-kernel path for Qwen3.6 Gated DeltaNet |
|
Technical, context-embedded, not overly |
2026-06-04 |
| pr |
0.20 |
model: Add NVIDIA Canary-1B-v2 to Transformers |
|
Slightly formal but still domain-specifi |
2026-06-22 |
| commit |
0.10 |
docs(trainer): add JIT checkpointing to trainer recipes (#46 |
|
Concise doc and changelog, domain detail |
2026-06-23 |
| commit |
0.10 |
[skills] help your agent get started (#45732) |
|
Minimal commit messages, no AI patterns, |
2026-06-23 |
| commit |
0.10 |
Fix mask when generating beyond sliding window (#46839) |
|
Conversational style, distinct typo, cle |
2026-06-23 |
| commit |
0.10 |
[RecurrentGemma] Support attn_implementation dispatch (#4632 |
|
Concise, technical commit list; no AI-st |
2026-06-17 |
| commit |
0.10 |
Add some AITER kernel routing for ROCm (#46268) |
|
Terse, domain-specific; no AI hallmarks |
2026-06-17 |
| commit |
0.10 |
feat[vLLM x v5]: Expose max_source_positions on VibeVoiceAsr |
|
Technical changelog, no AI-style formali |
2026-06-17 |
| pr |
0.10 |
Add qwen3 tts |
|
Domain-specific content, no AI-style phr |
2026-03-07 |
| pr |
0.10 |
Add HunYuan VL model |
|
Technical content, terse and to the poin |
2026-06-04 |
| pr |
0.10 |
Fix broken internal documentation links |
|
Contains a changelog table, technical co |
2026-06-28 |
| pr |
0.10 |
Insert a Grafana badge in the PR |
|
Simple, domain-related, informal; no AI |
2026-06-19 |
| pr |
0.10 |
fix(hub): prevent path traversal in sharded checkpoint loadi |
|
Explains a security fix with technical d |
2026-06-03 |
| pr |
0.10 |
Fix Mamba2 chunked-prefill / speculative decoding for Zamba2 |
|
Concise, domain-specific language and re |
2026-06-18 |
| pr |
0.10 |
Don't pin the gated delta net norm to `cuda:0` with a hardco |
|
Technical tone, references coding detail |
2026-06-22 |
| pr |
0.10 |
Fix padding mask skipped for batch size 1 in linear-attentio |
|
Specific to a bug with technical context |
2026-06-19 |
| pr |
0.10 |
fix(cli/serve): import serve handlers lazily so the CLI work |
|
Direct description with domain terminolo |
2026-06-07 |
| pr |
0.10 |
[Gemma4] Fix dtype casting for quantized vision/audio embedd |
|
Bugs and PR references, technical detail |
2026-06-27 |
| pr |
0.10 |
Fix CohereASR training-loss double-shift (same as Moonshine |
|
Describes a specific technical bug with |
2026-06-25 |
| pr |
0.10 |
Bump minimum torch version to 2.5 and clean up torch<2.5 wor |
|
Direct, minimal language and user checks |
2026-06-27 |
| pr |
0.10 |
[NemotronAsrStreaming] processor without modular |
|
Informal, brief update, refers to ongoin |
2026-06-24 |
| pr |
0.10 |
Fix Florence2 training-loss double-shift (same pattern as Mo |
|
Brief, domain-specific, with abbreviatio |
2026-06-25 |
| pr |
0.10 |
Fix YOLOS device mismatch with device_map="auto" |
|
Technical language, concise, uses domain |
2026-06-25 |
| pr |
0.10 |
[docs] continuous batching (offloading behavior, max batch t |
|
Human-like brevity, domain details, lack |
2026-06-26 |
| pr |
0.10 |
[gemma 3/4] Fix bidirectional attention masking crossing sli |
|
Contains specific technical issue; conci |
2026-06-23 |
| pr |
0.10 |
Add support for RADIO models |
|
Terse, uses model name, informal and che |
2026-06-04 |
| pr |
0.10 |
Replace VideoMAE sinusoid encoding helper with PyTorch imple |
|
Domain-specific detail, concise summary, |
2026-06-26 |
| pr |
0.10 |
[Bugfix] [Offloading] Save disk-offloaded buffers, Save conv |
|
Domain-specific jargon, bullet points, i |
2026-06-26 |
| pr |
0.10 |
Fix broken code examples in DIA documentation |
|
Mentions files and bugs, specific detail |
2026-06-26 |
| pr |
0.10 |
Qwen3 ASR and Forced Aligner |
|
Domain abbreviations, informal; checklis |
2026-02-08 |
| pr |
0.10 |
[Docs] Fix full disk offloading docs |
|
Documentation fix with specific versions |
2026-06-26 |
| pr |
0.10 |
🚨 🚨Bring some dinos to modern standards |
|
Informal, touch of humor, specific PR li |
2026-05-28 |
| pr |
0.10 |
Fix flash attention with StaticCache |
|
Terse, informal, domain-specific with a |
2026-06-26 |
| pr |
0.10 |
[new model] Add Zyphra/ZAYA1-8B |
|
Casual tone, uses abbreviations and doma |
2026-05-09 |
| pr |
0.10 |
fix(timesfm2_5): correct method name in window_size moving a |
|
Includes issue reference, concise summar |
2026-06-23 |
| pr |
0.10 |
[wav2vec2] Support attn_implementation=sdpa dispatch |
|
Technical explanation with code referenc |
2026-05-25 |
| pr |
0.10 |
Fix auto-mappings registration for remote code & fixes a few |
|
Multiple fixes explained with clear tech |
2026-06-25 |
| pr |
0.10 |
Enable CB paged decode fast path for MPS hub kernels + sensi |
|
Technical detail, minor template-like se |
2026-06-21 |
| pr |
0.10 |
Add MTP training support for Qwen3.5 |
|
Technical, concise; informal section ble |
2026-04-24 |
| pr |
0.10 |
Fix path traversal via vocab-file arguments in tokenizer_con |
|
Technical, concise description; uses rel |
2026-05-29 |
| pr |
0.10 |
[docs] EP plan validation |
|
Concise, domain-specific, some typo, not |
2026-06-25 |
| pr |
0.10 |
Add SAM 3.1 to Transformers |
|
Terse, uses dev lingo, non-formal, no AI |
2026-05-20 |
| pr |
0.10 |
Update EvalPrediction annotations for batch_eval_metrics |
|
Direct bug fix description, no formal or |
2026-06-25 |
| pr |
0.10 |
TST Run fast PEFT tests in normal CI |
|
PR describes test gating with dev commen |
2026-04-28 |
| pr |
0.10 |
Add heterogeneous model support (per-layer config and modeli |
|
Domain-specific, informal, technical, no |
2026-04-09 |
| pr |
0.10 |
Add heterogeneous config support (per-layer configuration) |
|
Uses domain-specific terms and informal |
2026-04-09 |
| commit |
0.05 |
docs: update apple silicon doc with safetensors `0.8.0` bene |
|
Commit messages have terse, domain-speci |
2026-06-23 |
| commit |
0.05 |
[DiffusionGemma] return_dict_in_generate=False for image-tex |
|
Technical commit messages, brief and inf |
2026-06-23 |
| commit |
0.05 |
fix(cli/serve): import serve handlers lazily so the CLI work |
|
Commit is concise, jargon-rich, and clea |
2026-06-23 |
| commit |
0.05 |
update kernels-community/aiter-rope version (#46810) |
|
Short, terse commit messages indicate hu |
2026-06-22 |
| commit |
0.05 |
Fix prompt lookup decoding crash when no EOS token is config |
|
Technical language and specific explanat |
2026-06-22 |
| commit |
0.05 |
Fix left-padding token selection in `BioGptForSequenceClassi |
|
Detailed technical content with domain j |
2026-06-22 |
| commit |
0.05 |
Fix broken internal links in model documentation (#46807) |
|
Technical doc fix with human phrasing; n |
2026-06-22 |
| commit |
0.05 |
Use cached added-token dicts in per-token decode loops (#465 |
|
Technical, includes co-authored-by trail |
2026-06-22 |
| commit |
0.05 |
Fix offloaded cache device mismatch on hybrid models (#46748 |
|
Technical explanation, informal incomple |
2026-06-22 |
| commit |
0.05 |
Add Videoprism (#39895) |
|
Terse, technical commit messages; human |
2026-06-19 |
| commit |
0.05 |
Add GB10/SM121 Hub-kernel path for Qwen3.6 Gated DeltaNet (# |
|
Technical jargon, informal tone; human-w |
2026-06-19 |
| commit |
0.05 |
Add Nemotron ASR Streaming (#46332) |
|
Very concise messages, domain-specific t |
2026-06-19 |
| commit |
0.05 |
fixes per head minimaxm3 (#46719) |
|
Informal, technical, few details; human- |
2026-06-19 |
| commit |
0.05 |
[fix] enable base64 str audio in load_audio (#46694) |
|
Short, technical, domain references; hum |
2026-06-19 |
| commit |
0.05 |
Lazily build the default kernel mapping to decouple `kernels |
|
Technical changes, informal commit messa |
2026-06-19 |
| commit |
0.05 |
Fix regression in ProcessorMixin._load_tokenizer_from_pretra |
|
Commit messages are concise and contain |
2026-06-19 |
| commit |
0.05 |
fix(aria): use math.ceil in get_number_of_image_patches to m |
|
Technical explanation with context, no s |
2026-06-19 |
| commit |
0.05 |
Fall back to the for-loop grouped_mm on CPU (#46743) |
|
Technical, uses human domain jargon, cle |
2026-06-19 |
| pr |
0.05 |
Add MOSS-TTS-v1.5 (delay) and MOSS-Audio-Tokenizer |
|
Direct, terse engineering style without |
2026-06-05 |
| pr |
0.05 |
Fix dtype cast in TimesFM 2.5 ResidualBlock for quantized we |
|
Technical explanation and concise; no si |
2026-06-28 |
| pr |
0.05 |
[`Dia`] Fix docs |
|
Minimal, informal, refs issues, no AI to |
2026-06-26 |
| pr |
0.05 |
[CI] Use pre-computed `_OLD_MODELS` in `test_new_models_requ |
|
Short, direct, includes a code block; in |
2026-06-25 |
| pr |
0.05 |
CI Test PEFT integration CI |
|
Short, informal; clear test statement; n |
2026-06-25 |
| pr |
0.05 |
call transformers-ci in a nightly run |
|
Brief, domain jargon, typo ('nighly'); h |
2026-06-22 |
| pr |
0.05 |
[CB] [Major] Add multimodal support to continous batching |
|
Technical, brief, direct; review comment |
2026-06-03 |
| pr |
0.05 |
[docs] AITER kernels |
|
Extremely terse, domain-specific, not AI |
2026-06-24 |
| commit |
0.02 |
[`CB`] Add FA2 to the fast path (#46729) |
|
Extremely terse and domain-specific, not |
2026-06-23 |
| commit |
0.02 |
Return logits from semantic segmentation post-process (#4616 |
|
Commit messages are brief, technical, an |
2026-06-19 |
| commit |
0.00 |
Fix broken internal documentation links (#46945) |
|
Concise commit message with domain conte |
2026-06-28 |
| commit |
0.00 |
Insert a Grafana badge in the PR (#46774) |
|
Informal commit message with humanlike t |
2026-06-28 |
| commit |
0.00 |
Add Nemotron 3.5 ASR Streaming (#46565) |
|
Terse commit messages with domain jargon |
2026-06-27 |
| commit |
0.00 |
[NemotronAsrStreaming] fix pipeline (#46870) |
|
Very brief, informal style; typical huma |
2026-06-27 |
| commit |
0.00 |
[NemotronAsrStreaming] processor without modular (#46865) |
|
Short, informal commit log; contains min |
2026-06-27 |
| commit |
0.00 |
[`Dia`] Fix docs (#46923) |
|
Commit uses terse messages and domain te |
2026-06-26 |
| commit |
0.00 |
Qwen3 ASR and Forced Aligner (#43838) |
|
Changelog uses natural, human-like phras |
2026-06-26 |
| commit |
0.00 |
[Docs] Fix full disk offloading docs (#46905) |
|
Brief human message with author signatur |
2026-06-26 |
| commit |
0.00 |
Fix dtype casting for quantized multimodal embedders (#46904 |
|
Multiple brief, iterative commit message |
2026-06-26 |
| commit |
0.00 |
Fix flash attention with StaticCache (#46914) |
|
Minimal, non-formal commit messages; hum |
2026-06-26 |
| commit |
0.00 |
[CB] Changes to increase max_batch_tokens (#46712) |
|
Colloquial commit sequence and direct st |
2026-06-26 |
| commit |
0.00 |
Redirect to diffusers pipe in docs for experimental features |
|
Message is short and direct; contains no |
2026-06-26 |
| commit |
0.00 |
Install in docker (#46910) |
|
Minimal, imperative commit message; typi |
2026-06-26 |
| commit |
0.00 |
[CI] Use pre-computed `_OLD_MODELS` in `test_new_models_requ |
|
Terse, informal style typical for humans |
2026-06-26 |
| commit |
0.00 |
call transformers-ci in a nightly run (#46811) |
|
Casual language and brief entries typica |
2026-06-26 |
| commit |
0.00 |
[docs] full disk offloading (#46893) |
|
Minimal commit message; no AI indicators |
2026-06-25 |
| commit |
0.00 |
TST Run fast PEFT tests in normal CI (#45679) |
|
Contains domain specifics, informal, and |
2026-06-25 |
| commit |
0.00 |
🚨 EP: fix EP router contract for many models + honor FP8 sca |
|
Informal, domain-specific, and includes |
2026-06-25 |
| commit |
0.00 |
[docs] AITER kernels (#46871) |
|
Standard template phrases, no AI signals |
2026-06-25 |
| commit |
0.00 |
nemotron_asr_streaming: set _supports_flex_attn to False (#4 |
|
Concise, standard commit with sign-off; |
2026-06-25 |
| commit |
0.00 |
Add native masked MSE loss for Sapiens2ForPoseEstimation (#4 |
|
Domain-specific, includes code details a |
2026-06-25 |
| commit |
0.00 |
blip 2 fix (#46816) |
|
Brief, domain-specific; no signs of AI g |
2026-06-25 |
| commit |
0.00 |
Use meshgrid for brevity (#46861) |
|
Informal, includes domain details and ab |
2026-06-25 |
| commit |
0.00 |
Add xcodec2 model (#44178) |
|
Detailed commit history with typical dev |
2026-06-25 |
| commit |
0.00 |
Rework some old cache getters/properties (#46862) |
|
Short, domain-focused commit message; no |
2026-06-25 |
| commit |
0.00 |
Prevent auto-class from being modified for all models (#4684 |
|
Terse commit messages with little free t |
2026-06-24 |
| commit |
0.00 |
Add Spanish translation of the torch.compile page (#46852) |
|
Brief, direct, no generic phrasing, stan |
2026-06-24 |
| commit |
0.00 |
docs: Update NeMo AutoModel doc examples (#46857) |
|
Technical phrasing and minor grammar iss |
2026-06-24 |
| commit |
0.00 |
[docs] distributed training (#44420) |
|
Short, domain-jargon, irregular style, h |
2026-06-24 |
| commit |
0.00 |
[docs] require trust_remote_code for custom_generate (#46677 |
|
Very brief, with minimal free text, huma |
2026-06-24 |
| commit |
0.00 |
Documentation for the kernel API (#46754) |
|
Doc-focused, with human feedback and ref |
2026-06-24 |
| commit |
0.00 |
add distributed config (#46705) |
|
Short messages, direct actions, no AI si |
2026-06-24 |
| commit |
0.00 |
[Offloading] [Bugfix] Fix disk offloading of models with exp |
|
Domain details, includes signed-off-by, |
2026-06-24 |
| commit |
0.00 |
Streamable chat parsing (#45847) |
|
Informal, iterative commit messages, hum |
2026-06-24 |
| commit |
0.00 |
Round the ue8m0 FP8 scale before quantizing so dequant match |
|
Technical changelog, clear domain knowle |
2026-06-24 |
| commit |
0.00 |
Fix BitNet packed-weight unpacking dtype (`F.linear` dtype m |
|
Concise commit message and common Signed |
2026-06-24 |
| commit |
0.00 |
Fix typos in code (#46579) |
|
Terse message, domain-specific; includes |
2026-06-24 |
| commit |
0.00 |
Fix Evolla eager attention for the GQA text decoder (#46860) |
|
Short, informal phrasing; no signs of AI |
2026-06-24 |
| commit |
0.00 |
Fix Moonshine training-loss double-shift (train against labe |
|
Detailed but domain-specific, technical, |
2026-06-24 |
| commit |
0.00 |
[CB] Fix issues with FA read / writes (#46765) |
|
Informal language, 'nits', and domain sl |
2026-06-24 |
| commit |
0.00 |
Fix expanded dim in the cache's cumulative length (#46856) |
|
Very terse, typical for fix commits; lac |
2026-06-24 |
| commit |
0.00 |
Switch decorator order (#46853) |
|
Extremely brief commit message; clearly |
2026-06-24 |
| commit |
0.00 |
Fix beam search for mamba models (#46819) |
|
Informal, iterative commit messages; no |
2026-06-24 |
| commit |
0.00 |
Import diffusion_gemma in models init (#46841) |
|
Terse phrasing with informal rationale, |
2026-06-23 |
| commit |
0.00 |
[Offloading] Support full disk offloading (#46749) |
|
Short, iterative commits with signoff tr |
2026-06-23 |
| commit |
0.00 |
Avoid NumPy 2.0 `__array__` copy-keyword deprecation in `cre |
|
Extremely terse summary, no free text, h |
2026-06-23 |
| commit |
0.00 |
Fix flex_attention block mask creation when `get_seq_length` |
|
Very terse, minimal commit; clearly huma |
2026-06-22 |
| commit |
0.00 |
DiffusionGemma: mask layout and CI (#46654) |
|
Very informal, terse, and casual tone; h |
2026-06-22 |
| commit |
0.00 |
fix another flaky test (#46767) |
|
Very informal, terse, casual commit mess |
2026-06-22 |
| commit |
0.00 |
Fix secondary rate limit when downloading artifacts in slack |
|
Technical, concise, includes human-writt |
2026-06-21 |
| commit |
0.00 |
docs: move SmolLM3 to Text models category in _toctree.yml ( |
|
Minimal structured doc update; no AI sig |
2026-06-19 |
| commit |
0.00 |
[CI] Fix artifact download path in self-comment-ci workflow |
|
Brief description, natural tone, human. |
2026-06-19 |
| commit |
0.00 |
[`Kernels`] Trigger proper kernelization on `use_kernels=Tru |
|
Minimal, technical, regular human PR con |
2026-06-19 |
| commit |
0.00 |
[`CI`] Fix some failures introduced by myself :grimacing: ( |
|
Casual, human tone with emoji; not AI. |
2026-06-19 |
| commit |
0.00 |
Kernelize refactor (#46520) |
|
Commit list uses terse, technical phrasi |
2026-06-18 |
| commit |
0.00 |
ci: add comment explaining why secrets are not inherited in |
|
Brief, domain-specific commit messages; |
2026-06-18 |
| commit |
0.00 |
ci: trigger PR CI on ci-* branches (#46746) |
|
Informal, concise commit; contains domai |
2026-06-18 |
| commit |
0.00 |
finegrained v3 (#46742) |
|
Very terse, single word PR title; clear |
2026-06-18 |
| commit |
0.00 |
[docs] metal flash attention (#46349) |
|
Extremely terse docs-related PR; no AI s |
2026-06-18 |
| commit |
0.00 |
Improve AutoImageProcessor error for unavailable backends (# |
|
Technical, structured change log; clear |
2026-06-18 |
| commit |
0.00 |
skip decorators must appear after @parameterized.expand in p |
|
Terse, direct language; human explanatio |
2026-06-18 |
| commit |
0.00 |
:rotating_light: [`Kernels`] Sync to latest version (#46039) |
|
Highly informal, personal tone with typo |
2026-06-17 |
| commit |
0.00 |
[docs] clarify initialization module usage (#46698) |
|
Extremely terse; just a summary, no AI s |
2026-06-17 |
| commit |
0.00 |
feat: bump safetensors to `0.8.0` (#46523) |
|
Conforms to changelog norms, no AI featu |
2026-06-17 |
| commit |
0.00 |
ci: disable CircleCI by replacing config with no-op (#46721) |
|
Very terse with casual language; human s |
2026-06-17 |
| commit |
0.00 |
fix: position ids does not exist in upstream rotary kernel ( |
|
Contains typos, informal phrasing, domai |
2026-06-17 |
| commit |
0.00 |
[CB] Fix offloading (#46587) |
|
Minimal, technical headings, lacks AI-li |
2026-06-17 |
| commit |
0.00 |
[`Templates`] Update members (#46720) |
|
Single-word summary, clearly human style |
2026-06-17 |
| pr |
0.00 |
Fix CohereASR training loss double shift |
|
Very brief/informal, mostly template and |
2026-06-28 |
| pr |
0.00 |
T5: add SDPA and Flash Attention 2 support |
|
Technical, template-structured, content |
2026-06-28 |
| pr |
0.00 |
Add Molmo2 |
|
Non-AI, technical wording, template chun |
2026-01-23 |
| pr |
0.00 |
[serge] Fix 12 integration tests for model `dac` failing wit |
|
— |
2026-06-28 |
| pr |
0.00 |
[serge] Fix 12 integration tests for model `edgetam` failing |
|
— |
2026-06-26 |
| pr |
0.00 |
docs(conditional_detr): fix num_queries default in docstring |
|
Minimal free-text; edit is specific with |
2026-06-27 |
| pr |
0.00 |
Add unit tests for `_ignore_causal_mask_sdpa` in `test_maski |
|
Technical, concise description; some min |
2026-06-27 |
| pr |
0.00 |
Add Gefen optimizer support to Trainer |
|
Technical, specific with truncated summa |
2026-06-16 |
| pr |
0.00 |
Add Nemotron 3.5 ASR Streaming |
|
No free-text or AI-style phrasing; templ |
2026-06-11 |
| pr |
0.00 |
Add V-JEPA 2.1 inference support |
|
Technical, domain-specific, concise; no |
2026-04-17 |
| pr |
0.00 |
Reject assisted generation for LFM2 and LFM2-MoE (set _is_st |
|
Technical, specific free-text; jargon, n |
2026-06-27 |
| pr |
0.00 |
Fix crash in greedy assisted generation with different token |
|
Technical, direct bug description; domai |
2026-06-27 |
| pr |
0.00 |
[NemotronAsrStreaming] fix pipeline |
|
Extremely terse, informal 'Nit fix'; cle |
2026-06-24 |
| pr |
0.00 |
[serge] Fix 8 integration tests for model `blip_2` failing w |
|
— |
2026-06-27 |
| pr |
0.00 |
[serge] Fix 12 integration tests for model `minimax_m3_vl` f |
|
— |
2026-06-27 |
| pr |
0.00 |
[serge] Fix 12 integration tests for model `glm46v` failing |
|
— |
2026-06-27 |
| pr |
0.00 |
Fix yolos |
|
Extremely terse, domain-specific, lacks |
2026-06-26 |
| pr |
0.00 |
Better and more extensive tests for RoPE |
|
Concise and uses domain jargon; no AI-st |
2026-06-26 |
| pr |
0.00 |
fix(qwen3_5): add linear_attn entries to base_model_tp_plan |
|
Technical, concise, and contains abbrevi |
2026-06-23 |
| pr |
0.00 |
Add native FSDP2 module + migration |
|
Brief, domain-specific, and contains a h |
2026-06-17 |
| pr |
0.00 |
Fix PPFormulaNet training-loss double-shift (train against l |
|
Contains technical detail and code refer |
2026-06-26 |
| pr |
0.00 |
Fix dtype casting for quantized multimodal embedders |
|
Technical content, no AI-style patterns |
2026-06-26 |
| pr |
0.00 |
fix(deepspeed): chunk ZeRO-3 missing-key param gather to avo |
|
Domain-specific, technical and concise, |
2026-06-26 |
| pr |
0.00 |
Remove deprecated training args |
|
Direct and technical, standard human eng |
2026-06-26 |
| pr |
0.00 |
[Olmo3] different RoPE per layer type |
|
Minimal, technical, references PR, conta |
2026-06-26 |
| pr |
0.00 |
Fix training-loss double-shift in Florence2 and CohereASR |
|
Brief, focuses on technical issue, no AI |
2026-06-26 |
| pr |
0.00 |
Add new model: Kimi2-6 |
|
Human style, informal comment, domain-wi |
2026-04-24 |
| pr |
0.00 |
[glm-mode-dsa] Indexer uses interleaved rope |
|
Domain-specific, casual, clear human exp |
2026-06-23 |
| pr |
0.00 |
fix(integrations): prevent device mismatch in msa_attention |
|
Off-the-cuff, minimal completion, not AI |
2026-06-23 |
| pr |
0.00 |
wip step 3 7 flash |
|
Terse, lacking structure, no AI signal. |
2026-06-15 |
| pr |
0.00 |
Redirect to diffusers pipe in docs for experimental features |
|
Direct and informal, with domain slang; |
2026-06-25 |
| pr |
0.00 |
Install in docker |
|
Very brief and informal, not indicative |
2026-06-26 |
| pr |
0.00 |
[serge] Fix 8 integration tests for model `deepseek_v32` fai |
|
— |
2026-06-26 |
| pr |
0.00 |
Port ESMC and ESMFold2 to Transformers |
|
Informal tone with abbreviations; no AI- |
2026-06-04 |
| pr |
0.00 |
Fix non-idempotent revert_weight_conversion corrupting trust |
|
Includes specific bug reference, brief t |
2026-06-15 |
| pr |
0.00 |
Fix path traversal in sharded-checkpoint loader via the inde |
|
Direct description with jargon; clear hu |
2026-05-29 |
| pr |
0.00 |
Fix GatedDeltaNet torch fallback prefill vs decode divergenc |
|
Template structure; actual content is te |
2026-06-24 |
| pr |
0.00 |
[docs] full disk offloading |
|
Extremely terse, informal, no AI signals |
2026-06-25 |
| pr |
0.00 |
[docs] update cache apis |
|
Brief, domain-specific, informal; no AI |
2026-06-25 |
