| commit |
1.00 |
tests: reduce processor test memory usage by using tiny Hub |
|
Commit message contains explicit AI assi |
2026-07-10 |
| commit |
1.00 |
Fix flash-attn Docker build broken by setuptools 83 removing |
|
Commit message contains explicit AI assi |
2026-07-10 |
| commit |
1.00 |
Make executorch exporter tests always use xnnpack backend (# |
|
Commit message contains explicit AI assi |
2026-07-09 |
| commit |
1.00 |
[Fix] Make DeepGEMM triton fallback more robust (#47126) |
|
Commit message contains explicit AI assi |
2026-07-09 |
| commit |
1.00 |
Skip caching_allocator_warmup on Neuron (no reuse pool to wa |
|
Commit message contains explicit AI assi |
2026-07-08 |
| commit |
1.00 |
Fix MoE decode optimization for experts living in a submodel |
|
Commit message contains explicit AI assi |
2026-07-07 |
| commit |
1.00 |
processor tests: use tiny Hub repos to reduce CI memory (#47 |
|
Commit message contains explicit AI assi |
2026-07-07 |
| commit |
1.00 |
docs-fix: IMDb dataset link in sequence classification guide |
|
Commit message contains explicit AI assi |
2026-07-06 |
| commit |
1.00 |
Add new model: Kimi2-6 (#45630) |
|
Commit message contains explicit AI assi |
2026-07-03 |
| commit |
1.00 |
Add tiny_model_id support to ProcessorTesterMixin for memory |
|
Commit message contains explicit AI assi |
2026-07-03 |
| pr |
1.00 |
Executorch exporter fixes |
|
PR body explicitly mentions AI collabora |
2026-07-10 |
| pr |
1.00 |
Fix deepgemm on multiple devices |
|
PR body explicitly mentions AI collabora |
2026-07-14 |
| pr |
1.00 |
adding amd quark config class changes |
|
PR body explicitly mentions AI collabora |
2026-07-14 |
| pr |
1.00 |
fix(llama): allow explicit head_dim when hidden_size not div |
|
PR body explicitly mentions AI collabora |
2026-05-26 |
| pr |
1.00 |
pytorch 2.13 is released , updating transformer to handle th |
|
PR body explicitly mentions AI collabora |
2026-07-09 |
| pr |
1.00 |
DeepGEMM shouldn't pad on SM90 |
|
PR body explicitly mentions AI collabora |
2026-07-14 |
| pr |
1.00 |
Optimize flash attention max seqlen computation in vision at |
|
PR body explicitly mentions AI collabora |
2026-07-08 |
| pr |
1.00 |
Refactor image loading in tests to use load_test_image helpe |
|
PR body explicitly mentions AI collabora |
2026-07-09 |
| pr |
1.00 |
Clarify extra_special_tokens documentation to match the exam |
|
PR body explicitly mentions AI collabora |
2026-07-10 |
| pr |
1.00 |
[generation] Fix misleading synced_gpus warning in continuou |
|
PR body explicitly mentions AI collabora |
2026-07-08 |
| pr |
1.00 |
OpenVINO HF Exporter |
|
PR body explicitly mentions AI collabora |
2026-07-01 |
| pr |
1.00 |
Add after-load fusion for static quantized MLPs |
|
PR body explicitly mentions AI collabora |
2026-07-01 |
| pr |
1.00 |
Point to Gemma 4 model in Gemma4ForCausalLM docstring exampl |
|
PR body explicitly mentions AI collabora |
2026-07-10 |
| pr |
1.00 |
Fix AqlmConfig error messages to say "int" instead of "float |
|
PR body explicitly mentions AI collabora |
2026-07-05 |
| pr |
1.00 |
Add repeated n-gram stopping criterion |
|
PR body explicitly mentions AI collabora |
2026-07-12 |
| pr |
1.00 |
Add support for Nemotron Omni |
|
PR body explicitly mentions AI collabora |
2026-06-08 |
| pr |
1.00 |
Add regression test for register() with a string config_clas |
|
PR body explicitly mentions AI collabora |
2026-07-11 |
| pr |
1.00 |
Fix thread-safety of fast (Rust-backed) tokenizers (Issue #4 |
|
PR body explicitly mentions AI collabora |
2026-07-09 |
| pr |
1.00 |
fix: resolve dtype mismatches in DAB-DETR and Grounding DINO |
|
PR body explicitly mentions AI collabora |
2026-07-11 |
| pr |
1.00 |
Fix half-precision torch.compile crash in DETR-family sine p |
|
PR body explicitly mentions AI collabora |
2026-07-10 |
| pr |
1.00 |
Route packed sdpa rows through torch.nn.attention.varlen.var |
|
PR body explicitly mentions AI collabora |
2026-07-09 |
| pr |
0.80 |
Feat: Apply multi-layered Harmonic Seal for data provenance |
|
Verbose, formal tone and atypical 'To th |
2026-07-10 |
| pr |
0.30 |
Add serialization to hf_argparser.py |
|
Slightly formal but focuses on concrete |
2024-03-30 |
| pr |
0.20 |
piepline strming info added in md file |
|
Some grammar issues and informal phrasin |
2026-07-13 |
| pr |
0.20 |
[WIP] Add Fusion-in-Decoder |
|
Brief, factual summary with domain links |
2022-03-02 |
| commit |
0.15 |
[Generation] Add static ensemble verification for lossy spec |
|
Technical detail, terse style, and speci |
2026-07-09 |
| pr |
0.15 |
Fix crash and tokenization mismatch in speculative decoding |
|
Contains concise technical fix summary; |
2026-05-20 |
| pr |
0.10 |
Route chat management calls to the service root (#47138) |
|
Concise, contains domain jargon and dire |
2026-07-13 |
| pr |
0.10 |
Fix hardcoded paths in siglip checkpoint/vocab loading |
|
No free-text section; only title and bad |
2026-07-08 |
| pr |
0.10 |
Add heterogeneous model support (per-layer config and modeli |
|
Title only with no verbose or ChatGPT-st |
2026-04-09 |
| pr |
0.10 |
Update AMD CI runner groups to amd-mi300 |
|
Brief title and direct, domain-specific |
2026-07-13 |
| pr |
0.10 |
Fix Qwen Omni batched text postprocessing |
|
Title only, no sign of AI-generated verb |
2026-07-09 |
| pr |
0.10 |
Kernel native causal_conv1d functions |
|
Title only, no verbose boilerplate or ge |
2026-07-13 |
| pr |
0.10 |
[docs] MPS graph cache |
|
Short, technical doc note with reference |
2026-07-13 |
| pr |
0.10 |
[docs] TokenizersBackend fallback |
|
Short doc note, technical, mentions conc |
2026-07-13 |
| pr |
0.10 |
Fix minor text formatting in transformers (#47254) |
|
Brief free-text with domain linkage and |
2026-07-12 |
| pr |
0.10 |
Polish repository docs in transformers |
|
Short explanation, informal, and pragmat |
2026-07-12 |
| commit |
0.05 |
Add native FSDP2 module + migration (#46707) |
|
Commit history is brief, human-typical m |
2026-07-09 |
| commit |
0.05 |
Fix experts implementation in two spots (#47097) |
|
Very terse, informal language and multip |
2026-07-09 |
| commit |
0.05 |
[Fix] Remove old automatic cross attn pattern from output re |
|
Brief, domain-specific, with code review |
2026-07-09 |
| pr |
0.05 |
Fix StaticLayer.get_seq_length return type annotation (#4598 |
|
Concise technical explanation with human |
2026-05-23 |
| pr |
0.05 |
[WIP] [performance doc] faster/leaner optimizers |
|
Minimal handwritten doc, includes TODO, |
2021-12-09 |
| pr |
0.05 |
Neuron sft exp |
|
No free-text, only title and badge, very |
2026-07-02 |
| pr |
0.05 |
Fix check for interactive stdout in _style function |
|
Only title and CI badge, no AI style det |
2026-07-13 |
| pr |
0.05 |
Resolve continuous batching XPU availability checks at runti |
|
No free-text, only title and CI badge, v |
2026-07-09 |
| pr |
0.05 |
Fix get_json_schema crash on non-string docstring choices |
|
Only title and CI badge, not enough for |
2026-07-04 |
| pr |
0.05 |
Fix inter-chunk recurrence in the Zamba2 / Nemotron-H Mamba2 |
|
No free-text, concise title and badge on |
2026-07-10 |
| commit |
0.00 |
[docs] DeepGEMM requirements (#47324) |
|
Commit message is terse, references doma |
2026-07-14 |
| commit |
0.00 |
:rotating_light: Fix GPTBigCode and GPTNeoX for the Transfor |
|
Human-like informal tone, includes domai |
2026-07-14 |
| commit |
0.00 |
DeepGEMM shouldn't pad on SM90 (#47313) |
|
The brief 'claude's fix' is highly infor |
2026-07-14 |
| commit |
0.00 |
Fix half-precision torch.compile crash in DETR-family sine p |
|
Detailed technical explanation and chang |
2026-07-14 |
| commit |
0.00 |
Fix hardcoded paths in siglip checkpoint/vocab loading (#471 |
|
Terse, issue-focused commit message, no |
2026-07-14 |
| commit |
0.00 |
[generation] Fix misleading synced_gpus warning in continuou |
|
Describes a fix and test update with hum |
2026-07-14 |
| commit |
0.00 |
Update AMD CI runner groups to amd-mi300 (#47307) |
|
Short and specific update message; no AI |
2026-07-14 |
| commit |
0.00 |
Point to Gemma 4 model in Gemma4ForCausalLM docstring exampl |
|
Docstring update noted concisely, inform |
2026-07-13 |
| commit |
0.00 |
Fix Qwen Omni batched text postprocessing (#47197) |
|
Technical domain language and brief sign |
2026-07-13 |
| commit |
0.00 |
Fix AqlmConfig error messages to say "int" instead of "float |
|
Problem-focused commit, uses informal ex |
2026-07-13 |
| commit |
0.00 |
Fix check for interactive stdout in _style function (#47283) |
|
Concise, domain-specific fix summary in |
2026-07-13 |
| commit |
0.00 |
Fix get_json_schema crash on non-string docstring choices (# |
|
Technical explanation, domain abbreviati |
2026-07-13 |
| commit |
0.00 |
Make `MODEL_IDS_TO_TOKENIZERS_BACKEND` capture all DeepSeek |
|
Commit trailer, standard style, terse la |
2026-07-13 |
| commit |
0.00 |
Update doc preprocessing regex to prevent ReDoS (#47187) |
|
Short commit description, no AI signals, |
2026-07-13 |
| commit |
0.00 |
[generate] Add proper MTP support (#46229) |
|
Terse, informal progress logs, no AI pat |
2026-07-13 |
| commit |
0.00 |
Shard on read Dtensor aware (#46717) |
|
Contains typos, domain jargon, informal |
2026-07-13 |
| commit |
0.00 |
Switch AMD daily CI to mi300 runners (#47259) |
|
Terse and technical commit message; huma |
2026-07-11 |
| commit |
0.00 |
Torch compile backend defaults to "neuron" (#47035) |
|
Commit messages are brief, use domain te |
2026-07-10 |
| commit |
0.00 |
Add heterogeneous config support (per-layer configuration) ( |
|
Commit uses domain jargon, succinct huma |
2026-07-10 |
| commit |
0.00 |
[fix] update integration test values (#47146) |
|
Minimal, terse commit language; nothing |
2026-07-10 |
| commit |
0.00 |
Fix DeepSpeed SP loss aggregation and LocalLayerRepository k |
|
Detailed but technical, with typos and i |
2026-07-10 |
| commit |
0.00 |
Fix FA performance regression (#47134) |
|
Brief notes, domain language, co-author |
2026-07-10 |
| commit |
0.00 |
tests only for the top 10 download models (#47244) |
|
Very terse, human style; no AI hallmarks |
2026-07-10 |
| commit |
0.00 |
Fix InputTokensDetails missing cache_write_tokens for openai |
|
Only a title and co-author line, typical |
2026-07-10 |
| commit |
0.00 |
Revert "Trigger a scheduled run" (#47249) |
|
Revert with references, human phrasing, |
2026-07-10 |
| commit |
0.00 |
Fix CI read-only cache failures by patching cached_files in |
|
Informal iterative commit notes, human t |
2026-07-10 |
| commit |
0.00 |
Remove Executorch from CI until latest version is supported |
|
Short, imperative style; matches typical |
2026-07-10 |
| commit |
0.00 |
Be more defensive with `remap_legacy_layer_types` for custom |
|
Brief, domain-specific message with a sm |
2026-07-10 |
| commit |
0.00 |
Fix DistributedConfig docstring for unimplemented sp_plan ( |
|
Free text includes detailed technical re |
2026-07-10 |
| commit |
0.00 |
Switch mlinter to 0.1.2 (#47172) |
|
Succinct, telegraphic commit messages; n |
2026-07-09 |
| commit |
0.00 |
trainer: clear MPS graph cache via torch_empty_cache_steps ( |
|
Highly detailed, technical, domain-speci |
2026-07-09 |
| commit |
0.00 |
Trigger a scheduled run (#47209) |
|
Minimal, repetitive commit word; no AI p |
2026-07-09 |
| commit |
0.00 |
No agent PR descriptions (#45790) |
|
Direct changelog-style commits, with dom |
2026-07-09 |
| commit |
0.00 |
Clarify that max_steps is required for datasets without __le |
|
Brief, technical, and informal commit me |
2026-07-09 |
| commit |
0.00 |
Cleanup pipelines, stop materializing generators (#47142) |
|
Concise, typical commit phrasing; no sig |
2026-07-09 |
| commit |
0.00 |
[Fix] Remove deprecated argument from `kernels` call (#47100 |
|
Brief, direct summary and technical lang |
2026-07-09 |
| commit |
0.00 |
Fix device_map computation when the no_split_modules have di |
|
Single-word, informal 'fix' message, cle |
2026-07-09 |
| commit |
0.00 |
Remove executorch from all-latest-gpu image + add torch smok |
|
Detailed explanation with domain errors |
2026-07-09 |
| commit |
0.00 |
Fix crash in greedy assisted generation with different token |
|
Thorough technical bug description; no A |
2026-07-09 |
| commit |
0.00 |
🌐 [i18n-KO] Translate accelerator_selection.md to Korean (#4 |
|
Standard i18n commit, contains co-author |
2026-07-08 |
| commit |
0.00 |
[i18n-KO] Translate optimum.md to Korean and fix Furiosa typ |
|
Brief translation commit, no AI signals |
2026-07-08 |
| commit |
0.00 |
[docs] fix curly quotes rendering to straight quotes (#47135 |
|
Typical terse docs commit message, fully |
2026-07-08 |
| commit |
0.00 |
Fix custom code which doesn't know about the new linear laye |
|
Standard fix message with Git trailer, n |
2026-07-08 |
| commit |
0.00 |
Reject path traversal in the `transformers_weights` config f |
|
Technical description with domain detail |
2026-07-08 |
| commit |
0.00 |
[docs] Custom code conversion mapping (#47114) |
|
Minimal free-text, standard docs update |
2026-07-08 |
| commit |
0.00 |
Add exporters min version requirements and test skip (#47161 |
|
Human-like terse project maintenance com |
2026-07-08 |
| commit |
0.00 |
tests: reduce processor test memory usage and use tiny test |
|
Terse and repetitive test commit with hu |
2026-07-08 |
| commit |
0.00 |
Clarify input device placement in the Quicktour inference ex |
|
Short human-style commit, no AI phrasing |
2026-07-08 |
| commit |
0.00 |
Extend continuous batching memory prediction test to XPU (#4 |
|
Technical, domain-specific commit, signe |
2026-07-08 |
| commit |
0.00 |
Fix case where `_LazyAutoMapping.register` is passed a `str` |
|
Technical commit message, concise and do |
2026-07-08 |
| commit |
0.00 |
[docs] MoE decode switching (#47149) |
|
Terse commit with domain reference, huma |
2026-07-08 |
| commit |
0.00 |
add XPU output expectations for minicpm3 tests (#47092) |
|
Short, specific, uses abbreviations, typ |
2026-07-08 |
| commit |
0.00 |
Diffusion gemma: fix failed test cases (#47025) |
|
Short, domain-relevant, no AI phrasing p |
2026-07-08 |
| commit |
0.00 |
add XPU Expectation for cosmos3_omni tests (#46880) |
|
Straightforward, terse, domain-specific |
2026-07-08 |
| commit |
0.00 |
Fix IndexError Bug in XLMRoberta/Camembert ForMultipleChoice |
|
Fix summary is direct, uses technical te |
2026-07-08 |
| commit |
0.00 |
Restrict QuantizedCache to only full attention (#47160) |
|
Extremely brief message; typical of huma |
2026-07-08 |
| commit |
0.00 |
[docs] continuous batching (offloading behavior, max batch t |
|
Concise conventional commit messages, hu |
2026-07-07 |
| commit |
0.00 |
[docs] fix autolinks (#46968) |
|
Brief, minimal text; typical human style |
2026-07-07 |
| commit |
0.00 |
Adapt voxtral_realtime to mistral-common 1.11.5 audio API (# |
|
Contains domain-specific changelog, no A |
2026-07-07 |
| commit |
0.00 |
revert #47121 (#47144) |
|
Standard revert message; no signs of AI |
2026-07-07 |
| commit |
0.00 |
Fix output labels for AudioFlamingo3 (and related) models (# |
|
Terse, informal commit chunks; human eng |
2026-07-07 |
| commit |
0.00 |
Fix false __len__ claims in Trainer docstrings (#47131) |
|
Brief, targeted doc corrections; distinc |
2026-07-07 |
| commit |
0.00 |
Fix: accept scalar audio formats in Voxtral transcription re |
|
Direct and concise, no signals of AI-gen |
2026-07-07 |
| commit |
0.00 |
[serge] Fix 12 integration tests for model `dac` failing wit |
|
— |
2026-07-07 |
| commit |
0.00 |
Fix CLI compatibility with huggingface_hub 1.22 (#47059) (#4 |
|
Detailed technical explanation, human-st |
2026-07-07 |
| commit |
0.00 |
Add TIPSv2 (#46347) |
|
Natural, informal commit fragments; clea |
2026-07-07 |
| commit |
0.00 |
we want to run the CI in the release branches (#47125) |
|
Informal tone; typical team communicatio |
2026-07-07 |
| commit |
0.00 |
Small improvement (#47128) |
|
Terse commit message with minimal conten |
2026-07-07 |
| commit |
0.00 |
[cache] Simplify cache dispatch based on layer_types (#47118 |
|
Succinct, iterative commit messages with |
2026-07-07 |
| commit |
0.00 |
[sdpa] Allow prefill to use FA kernel with StaticCache (#470 |
|
Terse change list and domain-specific co |
2026-07-07 |
| commit |
0.00 |
[Model] Support `use_cache=False` for DeepSeek V4 (#46965) |
|
Technical, to-the-point changelog with h |
2026-07-07 |
| commit |
0.00 |
Fix CANINE fp16 dtype mismatch in attention mask causing ONN |
|
Technical, concise, specific to code; hu |
2026-07-06 |
| commit |
0.00 |
Make doc builds faster (#47099) |
|
Standard changelog entries, informal ton |
2026-07-06 |
| commit |
0.00 |
Fix AltCLIP text embedding resize test (#47079) |
|
Minimal and terse commit messages; clear |
2026-07-06 |
| commit |
0.00 |
fix mask return-type contract regression and add correctness |
|
Domain-specific, informal, emails; stron |
2026-07-06 |
| commit |
0.00 |
Fix save_pretrained with offloading and weight conversions ( |
|
Terse, colloquial commit messages; defin |
2026-07-06 |
| commit |
0.00 |
Update dev (#47044) |
|
Extremely terse; characteristic of human |
2026-07-03 |
| commit |
0.00 |
[`Gemma4`] Update 1 integration test (#47042) |
|
Very short, direct commit message; human |
2026-07-03 |
| commit |
0.00 |
Update workflow callers to use `transformers-ci` (#47040) |
|
Brief, to-the-point; has a 'co-authored- |
2026-07-03 |
| commit |
0.00 |
Fix/qwen3 variants tp plan (#47041) |
|
Technical details and problem explanatio |
2026-07-03 |
| commit |
0.00 |
Add HunYuan VL model (#46417) |
|
Domain jargon, detailed code changes, an |
2026-07-03 |
| commit |
0.00 |
[Gemma4] Fix dtype casting for quantized vision/audio embedd |
|
Short, practical commit; typical human e |
2026-07-03 |
| commit |
0.00 |
chore(linter): add TRF018 modeling rule (#46259) |
|
Informal, stream-of-consciousness tone; |
2026-07-03 |
| commit |
0.00 |
Fix silent SDPA math-kernel fallback for GQA when key/value |
|
Technical content with human-style expla |
2026-07-02 |
| commit |
0.00 |
[PoC] HF exporters (#41992) |
|
Minimal commit messages and highly techn |
2026-07-02 |
| commit |
0.00 |
TST Skip PEFT tests if PEFT version is too low (#47027) |
|
Technical, concise phrasing, with minor |
2026-07-02 |
| pr |
0.00 |
Add Granite-swa and Granitemoe-swa model support |
|
Only a PR title is present; no AI signal |
2026-07-08 |
| pr |
0.00 |
[docs] Improve CLI copy |
|
PR content is just a brief, domain-speci |
2026-07-14 |
| pr |
0.00 |
Add Cosmos3 Edge model support |
|
Single, domain-specific model support ti |
2026-07-08 |
| pr |
0.00 |
Fix MoE expert decompression for non-32-divisible bit widths |
|
Brief, technical fix in PR title; contai |
2026-07-14 |
| pr |
0.00 |
:rotating_light: Fix GPTBigCode and GPTNeoX for the Transfor |
|
Short, technical PR title referencing mo |
2026-07-09 |
| pr |
0.00 |
Fix assistant tokens mask when a generation span ends at tok |
|
Technical issue stated succinctly, with |
2026-07-14 |
| pr |
0.00 |
Add RoMa keypoint matcher |
|
Only a technical model addition in the t |
2026-07-05 |
| pr |
0.00 |
[docs] DeepGEMM requirements |
|
Concise, human-written dependency note i |
2026-07-14 |
| pr |
0.00 |
[AudioFlamingo3] Fix bfloat16 dtype mismatch in audio encode |
|
Brief PR title with domain language; no |
2026-07-11 |
| pr |
0.00 |
Port ESMC and ESMFold2 to Transformers |
|
Title is a crisp, domain-specific portin |
2026-06-04 |
| pr |
0.00 |
[Nit] Add expectations for gemma4 tests on H100 |
|
Only a terse PR title without any AI hal |
2026-07-14 |
| pr |
0.00 |
Deprecate the old response_schema |
|
Short, domain-specific title lacking AI- |
2026-07-14 |
| pr |
0.00 |
[`peft`] Support key_mapping with PEFT models |
|
Human-style direct, technical PR title o |
2026-06-19 |
| pr |
0.00 |
LocateAnything 3B |
|
No free-text; brief domain term used as |
2026-05-30 |
| pr |
0.00 |
Fix half-precision torch.compile crash in DETR-family sine p |
|
Template structure; custom title only, n |
2026-07-10 |
| pr |
0.00 |
[docs] MTP support |
|
Concise, domain-specific PR title only, |
2026-07-13 |
| pr |
0.00 |
Fix PPFormulaNet training loss: prevent incorrect label shif |
|
PR title is technical and concise, not A |
2026-07-14 |
| pr |
0.00 |
[Improvement] Re-structure sonic MoE kernel interface to use |
|
Technical PR title with no signs of AI p |
2026-07-14 |
| pr |
0.00 |
fix(integrations): prevent device mismatch in msa_attention |
|
Technical, terse title only, no AI or te |
2026-06-23 |
| pr |
0.00 |
Avoid ~2x transient host memory when loading to a unified-me |
|
Informal, succinct domain-specific PR ti |
2026-07-11 |
| pr |
0.00 |
[Nit] Add kernels_fallback_ok kwarg to is_flash_attn_N_avail |
|
Domain-specific title, no indicators of |
2026-07-14 |
| pr |
0.00 |
Fix misleading synced_gpus warning in continuous batching |
|
Template usage; free-text is minimal and |
2026-07-13 |
| pr |
0.00 |
Keep user-typed multimodal placeholders out of the placehold |
|
Human-like, domain-focused and concise P |
2026-07-14 |
| pr |
0.00 |
feat(rope): fused Triton kernel for apply_rotary_pos_emb via |
|
Technical feat-style title with no AI ch |
2026-07-12 |
| pr |
0.00 |
Add Fun-ASR-Nano model |
|
Template use; only model name filled, no |
2026-05-24 |
| pr |
0.00 |
Pipeline parallel naive inference |
|
Only a PR title present; content is too |
2026-07-13 |
| pr |
0.00 |
[serge] Fix 20 integration tests for model `whisper` failing |
|
— |
2026-07-08 |
| pr |
0.00 |
🚨 Pass tp_plan from lm_heads |
|
Single technical phrase with emoji, clea |
2026-07-10 |
| pr |
0.00 |
[CB] [Major] Add multimodal support to continous batching |
|
Only a PR title present; no AI indicator |
2026-06-03 |
| pr |
0.00 |
[Tests] Fix slow video tensor creation from list of numpy ar |
|
Only a PR title present; no AI hallmarks |
2026-03-15 |
| pr |
0.00 |
[security] Add SSRF validation to load_image URL fetch |
|
Template structure; custom title only, n |
2026-07-10 |
| pr |
0.00 |
FSDP orchestration: apply + loading/saving |
|
PR title is brief and technical with no |
2026-07-01 |
| pr |
0.00 |
🚨[wav2vec2] Support attn_implementation=sdpa dispatch |
|
Title only, uses domain tags; nothing to |
2026-05-25 |
| pr |
0.00 |
[serge] Fix 10 integration tests for model `gemma3` failing |
|
— |
2026-07-14 |
| pr |
0.00 |
[serge] Fix 12 integration tests for model `glm46v` failing |
|
— |
2026-07-13 |
| pr |
0.00 |
Experiments with Neuron |
|
Casual and minimal, no sign of AI or for |
2026-07-13 |
| pr |
0.00 |
Fix Double Application of Softmax for Router Logits in MoE m |
|
Brief technical fix title, no AI cues pr |
2026-04-09 |
| pr |
0.00 |
Fix ZeroDivisionError in polynomial decay when warmup equals |
|
Straightforward bugfix title, no formal |
2026-07-13 |
| pr |
0.00 |
Fix pad_token_id=0 silently replaced by eos_token_id in beam |
|
Terse, domain-specific bugfix, not AI-ge |
2026-07-13 |
| pr |
0.00 |
Fix NaN loss in metrics when eval dataset produces no losses |
|
Human-written, short and technical probl |
2026-07-13 |
| pr |
0.00 |
Fix token not propagated to get_config_dict in PreTrainedCon |
|
Concise technical fix, no unnatural lang |
2026-07-13 |
| pr |
0.00 |
Fix OLMo 3 scaled RoPE handling for sliding attention |
|
Begins technical summary section, normal |
2026-05-13 |
| pr |
0.00 |
Standardize COCO test image fixture loading |
|
No free-text; only title and CI badge, h |
2026-07-11 |
| pr |
0.00 |
[xLSTM] Fix train mode crashes with return_last_states=False |
|
Only a technical, terse PR title with no |
2026-07-03 |
| pr |
0.00 |
fix: restore dict return type in create_masks_for_generate ( |
|
Brief, technical commit title typical of |
2026-07-12 |
| pr |
0.00 |
Also size the device_map buffer from the largest leaf module |
|
Only title and CI badge, lacks signs of |
2026-07-09 |
| pr |
0.00 |
Fix important model failures on MI450 |
|
Title and CI badge only; no AI hallmarks |
2026-07-13 |
| pr |
0.00 |
Fix Hunyuan-VL PIL image resize parity with reference prepro |
|
Title and CI badge only; style typical f |
2026-07-10 |
| pr |
0.00 |
Make `MODEL_IDS_TO_TOKENIZERS_BACKEND` capture all DeepSeek |
|
Only title and CI; direct and highly dom |
2026-07-13 |
| pr |
0.00 |
Fix orthogonal_ init for low-precision dtypes (bf16/fp16) |
|
Title and CI badge only; technical and c |
2026-07-10 |
| pr |
0.00 |
Fix TimesFM window_size AttributeError (#46821) |
|
Title and CI info; no free-form language |
2026-07-09 |
| pr |
0.00 |
Fix Swin2SR over-padding when dimensions are already aligned |
|
PR content is just a terse, domain-speci |
2026-07-09 |
| pr |
0.00 |
fix(musicflamingo): Fix dtype mismatch with _keep_in_fp32_mo |
|
Patch title is brief, has specific techn |
2026-07-13 |
| pr |
0.00 |
Fix double-shift training loss in CohereAsrForConditionalGen |
|
Concise bug fix title using technical te |
2026-07-10 |
| pr |
0.00 |
Fix off-by-one in MoonshineStreaming sliding window right co |
|
Describes classic off-by-one bug, terse |
2026-07-10 |
| pr |
0.00 |
Fix misleading T5Stack error messages when stack is used as |
|
Short, targeted fix message, typical for |
2026-07-10 |
| pr |
0.00 |
Fix wrong method name call in TimesFM 2.5 preprocessing |
|
Brief, clear summary with specific refer |
2026-07-10 |
