| COMMIT |
1.00 |
Block image_start/end_token_id in generation test sampling ( |
|
Commit message contains explicit AI assi |
2026-05-15 |
| COMMIT |
1.00 |
Fix M-RoPE device mismatch in Qwen3VL family under FSDP2 CPU |
|
Commit message contains explicit AI assi |
2026-05-15 |
| COMMIT |
1.00 |
Extract dynamic vision/audio tensors into standalone pure fu |
|
Commit message contains explicit AI assi |
2026-05-13 |
| COMMIT |
1.00 |
fix(testing_utils): guard get_device_capability with torch.c |
|
Explicit mention of AI collaboration: 'd |
2026-05-11 |
| COMMIT |
1.00 |
Add Qwen3.5 support for token classification (#45833) |
|
Commit message contains explicit AI assi |
2026-05-08 |
| COMMIT |
1.00 |
Fix WeightConverter regex incorrectly matching shared_expert |
|
Commit message contains explicit AI assi |
2026-05-06 |
| COMMIT |
1.00 |
Add Granite 4.1 Vision (granite4_vision) (#45597) |
|
Commit message contains explicit AI assi |
2026-05-05 |
| COMMIT |
1.00 |
Unwrap `text_config` in `AutoModelFor*.from_config` (#45770) |
|
Commit message contains explicit AI assi |
2026-05-05 |
| COMMIT |
1.00 |
Add EXAONE 4.5 implementations (#45471) |
|
Commit message contains explicit AI assi |
2026-05-04 |
| PR |
1.00 |
Add Sapiens2 Model |
|
PR body explicitly mentions AI collabora |
2026-05-12 |
| PR |
1.00 |
fix: ModuleNotFoundError caused by distributed race conditio |
|
PR body explicitly mentions AI collabora |
2026-05-08 |
| PR |
1.00 |
docs: Fix formatting issues in weightconverter.md |
|
PR body explicitly mentions AI collabora |
2026-05-15 |
| PR |
1.00 |
[Llama4] Adopt `@use_experts_implementation` standard interf |
|
PR body explicitly mentions AI collabora |
2026-05-15 |
| PR |
1.00 |
Fix `MoeTensorParalellExperts` crash for Llama4 pre-weighted |
|
Explicit statement of AI assistance (Cla |
2026-05-14 |
| PR |
1.00 |
fix: add HF_USE_MLX opt-out for MLX detection |
|
PR body explicitly mentions AI collabora |
2026-05-09 |
| PR |
1.00 |
Fix colqwen2 test |
|
PR body explicitly mentions AI collabora |
2026-05-14 |
| PR |
1.00 |
fix feature extractor error messages to mention processor_co |
|
PR body explicitly mentions AI collabora |
2026-05-13 |
| PR |
1.00 |
fix: restore `_attn_implementation `and fix request offset i |
|
PR body explicitly mentions AI collabora |
2026-05-13 |
| PR |
1.00 |
[Cache] Add `Cache.snapshot()` / `Cache.restore(snapshot)` f |
|
PR body explicitly mentions AI collabora |
2026-05-08 |
| PR |
1.00 |
Add unified Cache-layer management for GLM-5 DSA Indexer key |
|
PR body explicitly mentions AI collabora |
2026-04-23 |
| PR |
1.00 |
torch.backends.fp32_precision cascade conv/rnn so removing t |
|
PR body explicitly mentions AI collabora |
2026-05-04 |
| COMMIT |
0.20 |
Warn about forgetting attention mask functions (#45811) |
|
Somewhat more formal and explanatory, bu |
2026-05-11 |
| PR |
0.20 |
Fix decoder_attention_mask None handling in generation utils |
|
Section headings are template; summary s |
2026-05-15 |
| PR |
0.20 |
[Generation] Add static ensemble verification for lossy spec |
|
Direct technical explanation, but some f |
2026-05-14 |
| PR |
0.20 |
Add TurboQuant KV cache backend |
|
Structured but not overly formal, contai |
2026-05-13 |
| PR |
0.20 |
🚨 [ALM] Add base model without head |
|
Casual tone, includes emoji, domain deta |
2026-04-20 |
| PR |
0.20 |
Enhance the handling of Union types in HfArgumentParser |
|
Slightly formal but has domain details a |
2025-10-08 |
| PR |
0.20 |
Drop `content=None` from messages in `apply_chat_template` |
|
Succinct technical language with domain |
2026-04-14 |
| PR |
0.20 |
Qwen3 ASR and Forced Aligner |
|
Technical content, fragmentation due to |
2026-02-08 |
| PR |
0.15 |
Streamable chat parsing |
|
Natural, informal, some domain specifics |
2026-05-08 |
| PR |
0.15 |
GgufLinear: inference-time GGUF matmul on Apple Silicon — ll |
|
Uses domain-specific quantization names, |
2026-05-14 |
| PR |
0.15 |
Parakeet tdt |
|
Succinct, domain-specific; lacks ChatGPT |
2026-02-20 |
| COMMIT |
0.10 |
Fix slow Trainer path with 4D attention mask (#45852) |
|
Domain-specific detail, natural technica |
2026-05-12 |
| COMMIT |
0.10 |
fix(rope): read original_max_position_embeddings from yarn v |
|
Technical explanation, domain-specific p |
2026-05-12 |
| PR |
0.10 |
[Weight converter] Account for base model prefix in scoped w |
|
Terse, technical description with domain |
2026-05-15 |
| PR |
0.10 |
RFDetr - use correct Roboflow org for release |
|
Brief and specific, lacks AI-style forma |
2026-05-13 |
| PR |
0.10 |
Trainer.compute_loss: fix loss over-counting under TP and EP |
|
Uses technical language and context-spec |
2026-05-15 |
| PR |
0.10 |
Refacto GGUF weight conversion |
|
Colloquial 'vendored from', technical re |
2026-03-17 |
| PR |
0.10 |
[new model] Add Zyphra/ZAYA1-8B |
|
Minimal, human-like introduction with di |
2026-05-09 |
| PR |
0.10 |
Block image_start/end_token_id in generation test sampling |
|
Informal summary and explicit reference |
2026-05-12 |
| PR |
0.10 |
fix model parallel issues for deimv2 |
|
Direct, terse, with real test output sni |
2026-05-08 |
| PR |
0.10 |
Fix TAPAS tokenizer crash on pandas 3.x string-dtype tables |
|
Technical, mentions pandas and dtype, hu |
2026-05-12 |
| PR |
0.10 |
fix: enable single-scale input support in SAM3 mask decoder |
|
Concise language and technical phrasing; |
2026-05-15 |
| PR |
0.10 |
fix: enable _supports_flash_attn_2 for EOMT |
|
Brief, technical, and uses project-speci |
2026-05-15 |
| PR |
0.10 |
feat: enable Flash Attention 2 for T5Gemma2 |
|
Efficient, domain-specific language with |
2026-05-15 |
| PR |
0.10 |
[PoC] HF exporters |
|
Informal edits and casual language; no A |
2025-11-03 |
| PR |
0.10 |
Fix MLX backend detection to respect HF_USE_MLX environment |
|
Direct, technical fix description; style |
2026-05-15 |
| PR |
0.10 |
[pp_formulanet] Fix polynomial ReDoS in remove_chinese_text_ |
|
Succinct, technical, references bug boun |
2026-05-13 |
| PR |
0.10 |
perf: replace degenerate Conv3d with Linear in Qwen VL patch |
|
Technical detail and abbreviations; natu |
2026-05-15 |
| PR |
0.10 |
Correct wrong function signatures on the docs website |
|
Succinct, human tone; checkboxes are fro |
2021-08-20 |
| PR |
0.10 |
Fix M-RoPE device mismatch in Qwen3VL family under FSDP2 CPU |
|
Project-specific, informal tone; structu |
2026-05-09 |
| PR |
0.10 |
hf_argparser: fix parse_yaml_file missing utf-8 encoding and |
|
Details specific bugs, precise bug repor |
2026-05-14 |
| PR |
0.10 |
test: add regression test for chat_template kwarg override ( |
|
Jargon, technical summary, informal tone |
2025-09-17 |
| PR |
0.10 |
chore(ci): replace hardcoded maintainer allowlists with team |
|
Technical wording, abbreviations, and tr |
2026-05-14 |
| PR |
0.10 |
Enable kernels-community/metal-flash-sdpa on MPS |
|
Technical, highly specific context, abbr |
2026-05-14 |
| PR |
0.10 |
Support Audio Flamingo Next checkpoints |
|
Domain terms, code refs, concise summary |
2026-03-18 |
| PR |
0.10 |
FSDP2 native support in transformers |
|
Mix of French, domain-specific shorthand |
2026-02-17 |
| PR |
0.10 |
Add Maximal Update Parametrization (μP) |
|
Technical and domain-specific, no AI-lik |
2026-05-08 |
| PR |
0.10 |
[docs] contributing |
|
Uses informal points and typical commit/ |
2026-04-15 |
| PR |
0.10 |
Fix M-RoPE `inv_freq` device and `meta` → `to_empty` re-init |
|
Short, issue-linked description, uses te |
2026-05-12 |
| PR |
0.10 |
Deepseek v4 csa mask collapse |
|
Brief, informal reference to external di |
2026-05-13 |
| PR |
0.10 |
Extended n-to-1 kernel fusion via `KernelConfig` |
|
Technical explanation with clear formatt |
2026-04-27 |
| PR |
0.10 |
Improve handling of QuantizedLayer.reset |
|
Very brief technical explanation; not AI |
2026-02-10 |
| COMMIT |
0.05 |
🚨 Refactor ViT to updated standards (#41693) |
|
Commit messages are terse, use domain te |
2026-05-08 |
| PR |
0.05 |
[CB] [Major] Add tensor paralellism |
|
Jargon-heavy, bullet points, abbreviatio |
2026-05-07 |
| PR |
0.05 |
tokenizer: decoding an empty batch returns an empty list |
|
Explains edge case with jargon; informal |
2026-05-14 |
| PR |
0.05 |
[docs] decode fast path |
|
Concise, domain-specific, notes human re |
2026-05-11 |
| PR |
0.05 |
chore(ci): remove dead env vars from circleci-failure-summar |
|
Technical summary, YAML example; not AI |
2026-05-14 |
| PR |
0.05 |
tests: fix duplicated "for" in test_video_utils FIXME commen |
|
Minimal, casual, terse commit-style lang |
2026-05-14 |
| PR |
0.05 |
Fix deepseek v4 |
|
Dense technical jargon and incomplete se |
2026-05-11 |
| PR |
0.05 |
GGUF: optional Metal dequant fast path via kernels-community |
|
Concise, precise, and domain-focused des |
2026-05-14 |
| PR |
0.05 |
fix(ci): set persist-credentials: false on actions/checkout |
|
Direct technical phrasing, clear domain |
2026-05-14 |
| PR |
0.05 |
chore(ci): set default workflow permissions to contents: rea |
|
Terse, focused, uses standard engineerin |
2026-05-14 |
| PR |
0.05 |
fix(ci): remove template injection on pull_request_target wo |
|
Technical and direct, no AI-style formal |
2026-05-14 |
| PR |
0.05 |
chore(ci): pin all GitHub Actions and reusable workflows by |
|
Clear technical intent, with accurate, d |
2026-05-14 |
| PR |
0.05 |
Add FP8 kernel acceleration for compressed-tensors quantized |
|
Technical summary and detailing, with ty |
2026-04-29 |
| COMMIT |
0.00 |
🚨Fix memory leaks caused by lru decorators in vision models |
|
Commit log uses terse, technical, and in |
2026-05-15 |
| COMMIT |
0.00 |
RFDetr - use correct Roboflow org for release (#45946) |
|
Contains only summary and a human co-aut |
2026-05-15 |
| COMMIT |
0.00 |
docs: Fix formatting issues in weightconverter.md (#45988) |
|
Brief, human-formulated commit summary. |
2026-05-15 |
| COMMIT |
0.00 |
Fix colqwen2 test (#45981) |
|
Extremely terse, typical of human commit |
2026-05-15 |
| COMMIT |
0.00 |
[docs] chat template prefill (#45947) |
|
Single word 'docs' is terse and human-li |
2026-05-14 |
| COMMIT |
0.00 |
[docs] decode fast path (#45899) |
|
Terse technical commit, no AI signal. |
2026-05-14 |
| COMMIT |
0.00 |
fix: restore `_attn_implementation `and fix request offset i |
|
Informal, includes technical shorthand a |
2026-05-14 |
| COMMIT |
0.00 |
Support Audio Flamingo Next checkpoints (#44830) |
|
Changelog with domain jargon and informa |
2026-05-14 |
| COMMIT |
0.00 |
Expose `per_layer_inputs` for every Gemma4 variants (#45927) |
|
Very brief commit messages; no AI indica |
2026-05-14 |
| COMMIT |
0.00 |
chore: update benchmark_v2.yml (#45966) |
|
— |
2026-05-14 |
| COMMIT |
0.00 |
chore: update build-ci-docker-images.yml (#45959) |
|
— |
2026-05-14 |
| COMMIT |
0.00 |
fix(ci): set persist-credentials: false on actions/checkout |
|
Very technical, detailed, with domain-sp |
2026-05-14 |
| COMMIT |
0.00 |
chore(ci): set default workflow permissions to contents: rea |
|
Technical explanation with human tone; n |
2026-05-14 |
| COMMIT |
0.00 |
fix(ci): remove template injection on pull_request_target wo |
|
Technical with filenames and error refer |
2026-05-14 |
| COMMIT |
0.00 |
chore(ci): pin all GitHub Actions and reusable workflows by |
|
Detailed and technical; lists tool and f |
2026-05-14 |
| COMMIT |
0.00 |
[docs] ALMModelTest (#45900) |
|
Minimal commit messages; no AI signals. |
2026-05-13 |
| COMMIT |
0.00 |
[docs] new attention mask helpers (#45949) |
|
Standard template usage and terse messag |
2026-05-13 |
| COMMIT |
0.00 |
Enhance apply_chat_template to support custom field prefilli |
|
Contains domain-specific detail and huma |
2026-05-13 |
| COMMIT |
0.00 |
Pass packed boundary metadata to Qwen3.5 linear-attention fa |
|
Dense technical commit log, human-like s |
2026-05-13 |
| COMMIT |
0.00 |
BUGFIX: Support hubert models that don't have conv_pos_batch |
|
Short and focused human edits, plus coau |
2026-05-13 |
| COMMIT |
0.00 |
Revert 45777 (#45942) |
|
Single-word revert message; human typica |
2026-05-13 |
| COMMIT |
0.00 |
pass the otel secrets (#45933) |
|
Informal lowercase message; no AI traits |
2026-05-13 |
| COMMIT |
0.00 |
Add initial torch_tpu backend support (#45918) |
|
Detailed technical changes and realistic |
2026-05-13 |
| COMMIT |
0.00 |
[CB] Hide activation footprint by using the CUDA graph pool |
|
Short, informal commit language, domain |
2026-05-13 |
| COMMIT |
0.00 |
Require input_ids for repetition penalty (#45389) |
|
Contains typos, informal style, and coau |
2026-05-13 |
| COMMIT |
0.00 |
Fix undefined 'input' variable (#45895) |
|
Terse commit message, no AI signals, no |
2026-05-13 |
| COMMIT |
0.00 |
Deepseek v4 csa mask collapse (#45928) |
|
Technical detail and abbreviations; no A |
2026-05-13 |
| COMMIT |
0.00 |
[fix] Add `fatal_error` to `ContinuousBatchingManager` so th |
|
Terse, domain-specific, terse comments; |
2026-05-13 |
| COMMIT |
0.00 |
Fix fsdp2 for gemma models with shared kv states (#45912) |
|
Casual wording and co-author trailer ind |
2026-05-13 |
| COMMIT |
0.00 |
[docs] update model cards (#45612) |
|
Terse commit style, no AI signals. |
2026-05-12 |
| COMMIT |
0.00 |
torch.backends.fp32_precision cascade conv/rnn so removing t |
|
Informal tone, no signs of AI generation |
2026-05-12 |
| COMMIT |
0.00 |
:rotating_light: [`Attn`] Remove all old mask APIs from mode |
|
Changelog only, terse, notably informal. |
2026-05-12 |
| COMMIT |
0.00 |
fix(text-generation): use token-level slicing for return_ful |
|
Terse commit summaries, human style. |
2026-05-12 |
| COMMIT |
0.00 |
[CI] AMD docker: bump to ROCm 7.2.2 / PyTorch 2.10 + prebuil |
|
Short, task-focused fragments, human-wri |
2026-05-12 |
| COMMIT |
0.00 |
Automatically inherit properties from children in composite |
|
One-word commit summaries, human pattern |
2026-05-12 |
| COMMIT |
0.00 |
Fix deepseek v4 (#45892) |
|
Informal fragments, explicitly criticize |
2026-05-12 |
| COMMIT |
0.00 |
Fix/pe audio video bugs (#45886) |
|
Concise task-based commit messages, huma |
2026-05-12 |
| COMMIT |
0.00 |
Forward `revision` to `list_repo_files` in tokenizer loading |
|
Terse, technical commit; no AI signals. |
2026-05-12 |
| COMMIT |
0.00 |
Remove `_checkpoint_conversion_mapping` from model attribute |
|
Brief, to-the-point message; no AI signa |
2026-05-12 |
| COMMIT |
0.00 |
Use `_keys_to_ignore_on_load_unexpected/missing` recursively |
|
Informal, minimal changelog-style messag |
2026-05-12 |
| COMMIT |
0.00 |
Do not keep refs to submodules internally (#45889) |
|
List of short, informal entries; no AI-g |
2026-05-12 |
| COMMIT |
0.00 |
[docs] paper cuts (#45798) |
|
Minimal, informal phrasing with no AI ma |
2026-05-11 |
| COMMIT |
0.00 |
[Weight converter] Revert unnecessary changes to `rename_sou |
|
Informal, pragmatic, and terse—lacks AI |
2026-05-11 |
| COMMIT |
0.00 |
fix(minicpmv4_6): skip invalid failing tests (#45836) |
|
Terse, technical, specific—no AI style d |
2026-05-11 |
| COMMIT |
0.00 |
audio tester class (#45391) |
|
Domain-specific shorthand and brief note |
2026-05-11 |
| COMMIT |
0.00 |
Remove deprecation cycle for inputs embeds (#45885) |
|
Extremely brief, laconic style; no AI si |
2026-05-11 |
| COMMIT |
0.00 |
Remove deprecation cycle for `cache_position` in masking pri |
|
Minimal, informal commit message. |
2026-05-11 |
| COMMIT |
0.00 |
Revert "[deepseek_v4] fix CSA per-query masking in eager pat |
|
Standard revert message; no AI hallmarks |
2026-05-11 |
| COMMIT |
0.00 |
Revert "Merge branch 'main' of github.com:huggingface/transf |
|
Auto-generated revert, template-based fo |
2026-05-11 |
| COMMIT |
0.00 |
[deepseek_v4] fix CSA per-query masking in eager path |
|
In-depth technical detail, human tone, a |
2026-05-11 |
| COMMIT |
0.00 |
fix(rf_detr): fix failing tests (#45845) |
|
Direct, minimal language, issue context, |
2026-05-08 |
| COMMIT |
0.00 |
fix(granite4_vision): auto-fix failing tests (#45844) |
|
Direct technical description, signatures |
2026-05-08 |
| COMMIT |
0.00 |
fix(laguna): fix failing tests (#45842) |
|
Minimal edit, direct cause, no AI hallma |
2026-05-08 |
| COMMIT |
0.00 |
:rotating_light: Generic Sequence Classifier works for multi |
|
Informal, nonstandard grammar, domain-sp |
2026-05-08 |
| COMMIT |
0.00 |
Fix import error in moe.py by providing explicit schema to c |
|
Technical context, precise, realistic er |
2026-05-08 |
| COMMIT |
0.00 |
fix: correct typo 'seperate' -> 'separate' in comments acros |
|
Direct typo fix description, minimal det |
2026-05-08 |
| COMMIT |
0.00 |
🚨 [Fuyu] Remove FuyuBatchFeature subclass, use BatchFeature |
|
Terse style, clear purpose, co-author hu |
2026-05-08 |
| COMMIT |
0.00 |
Keep deleting (#45802) |
|
Extremely terse, informal commit message |
2026-05-08 |
| COMMIT |
0.00 |
Fix gemma4 with multi-gpu setup (#45826) |
|
Brief and informal; no AI hallmarks pres |
2026-05-07 |
| COMMIT |
0.00 |
Add HyperCLOVAX SEED Think 14B (#44956) |
|
Commit trailers but no AI signals; conci |
2026-05-07 |
| COMMIT |
0.00 |
Fix `kernelize()` crash for gpt_oss: missing `@use_kernel_fu |
|
Technical fixes, domain jargon, no AI ph |
2026-05-07 |
| COMMIT |
0.00 |
Cache `merged_typed_dict` to not break `validate_typed_dict` |
|
Direct changelog with domain terms; lack |
2026-05-07 |
| COMMIT |
0.00 |
Get rid of deprecated use_return_dict call. (#45815) |
|
Minimal, terse, human-written style, no |
2026-05-07 |
| COMMIT |
0.00 |
fix(qianfan_ocr): add XPU expectations (#45615) |
|
Technical listing; informal, lacks AI te |
2026-05-07 |
| COMMIT |
0.00 |
Fix shared config mutation issue in flash_attn_from_config ( |
|
Domain-specific, concise, signed-off; no |
2026-05-07 |
| COMMIT |
0.00 |
Add RF-DETR (#36895) |
|
Changelog is technical and terse, lacks |
2026-05-07 |
| COMMIT |
0.00 |
[Weight Converter] More fine-grained mappings on classes, sc |
|
Highly technical, multiple fixes, domain |
2026-05-07 |
| COMMIT |
0.00 |
refector: renamed file glob to cache to make it clearer (#45 |
|
Brief, informal commit message indicates |
2026-05-06 |
| COMMIT |
0.00 |
Fix decorator order (#45806) |
|
Very terse, sometimes typoed commit mess |
2026-05-06 |
| COMMIT |
0.00 |
[`Granite 4.1 Vision`] Fixup integration tests (#45805) |
|
Terse commit messages, human abbreviatio |
2026-05-06 |
| COMMIT |
0.00 |
fix: validate special token ids against attribute values (#4 |
|
Technical, concise; shows informal engin |
2026-05-06 |
| COMMIT |
0.00 |
Blockwise mask fn as opt arg in all masking functions (#4547 |
|
Highly informal, fragmented notes; no si |
2026-05-06 |
| COMMIT |
0.00 |
[CB] Refactor any model-related code in a separate class (#4 |
|
Uses domain jargon, informal corrections |
2026-05-06 |
| COMMIT |
0.00 |
fix: forward use_cache kwarg to attention mixer in nemotron_ |
|
Brief technical commit messages with dom |
2026-05-05 |
| COMMIT |
0.00 |
fix: correct spelling in continuous_api docstring (#45749) |
|
Minimal, terse technical correction; typ |
2026-05-05 |
| COMMIT |
0.00 |
Fix link to modular transformers documentation (#45746) |
|
Direct update explanation; clear, concis |
2026-05-05 |
| COMMIT |
0.00 |
Gemma4: fix failed test cases (#45568) |
|
List of technical fixes and updates, sig |
2026-05-05 |
| COMMIT |
0.00 |
First model (#45788) |
|
Casual tone and specific jargon; reflect |
2026-05-05 |
| COMMIT |
0.00 |
Fix CI: Allow more artifacts to be download in CI (#45785) |
|
Debug note and domain-specific context, |
2026-05-05 |
| COMMIT |
0.00 |
Add `concurrency` to `PR CI` workflow file (`pr-ci-caller.ym |
|
Single word commit, terse; standard huma |
2026-05-05 |
| COMMIT |
0.00 |
Reorder decorators for autodoc and dataclass (#45702) |
|
Short, technical log phrases typical of |
2026-05-05 |
| COMMIT |
0.00 |
deepseek r1 distilled tokenizer fix for qwen2 mapping (#4574 |
|
Typo present, informal tone; clear human |
2026-05-05 |
| COMMIT |
0.00 |
fix: Added Mps support in float fallback backends list (#45 |
|
Structured code fixes, informal language |
2026-05-05 |
| COMMIT |
0.00 |
Github Actions PR CI (caller) (#45476) |
|
Commit message is terse and includes dom |
2026-05-04 |
| COMMIT |
0.00 |
make sure we call check_auto in CI (#45775) |
|
Informal tone; domain-specific and conci |
2026-05-04 |
| COMMIT |
0.00 |
Better Grouped GEMM + EP (#45621) |
|
Commit log is informal, terse, and techn |
2026-05-04 |
| COMMIT |
0.00 |
DeepSeek OCR specifies an incorrect tokenizer class on the H |
|
Human-written; issue title is concise an |
2026-05-04 |
| PR |
0.00 |
trainer: clear MPS graph cache after each optimizer step (py |
|
Domain-specific summary, technical with |
2026-05-07 |
| PR |
0.00 |
Add V-JEPA 2.1 inference support |
|
Technical and domain-specific, no AI hal |
2026-04-17 |
| PR |
0.00 |
cache: store StaticLayer.cumulative_length as a 0-dim scalar |
|
Domain-specific, technical, terse with c |
2026-05-15 |
| PR |
0.00 |
🚨Fix memory leaks caused by lru decorators in vision models |
|
Contains technical explanation, concise |
2026-05-12 |
| PR |
0.00 |
feat: add crop() to StaticCache layers for assisted generati |
|
Terse, domain-specific, no generic AI ph |
2026-05-02 |
| PR |
0.00 |
FSDP + TP & native save/load distributed |
|
Technical, domain jargon, and cut-off te |
2026-03-26 |
| PR |
0.00 |
bugfix(ci): avoid E2BIG in pr_slow_ci_suggestion |
|
Brief, technical, terse; no AI phrasing. |
2026-05-14 |
| PR |
0.00 |
Fix models for which we don't have a dedicated tokenizer cla |
|
Casual, slightly unpolished; uses projec |
2026-05-13 |
| PR |
0.00 |
gpt_oss multi-GPU AMD support |
|
Uses technical jargon and code, domain-s |
2026-04-30 |
| PR |
0.00 |
cache_utils: fix StaticSlidingWindowLayer.get_mask_sizes ret |
|
Brief, domain-specific bugfix with terse |
2026-05-13 |
| PR |
0.00 |
Strip language_model. prefix when loading multimodal Gemma 4 |
|
Concise, domain-specific phrasing; no AI |
2026-05-14 |
| PR |
0.00 |
[CB] Remove OpenTelemetry |
|
Structured, context-specific and uses sh |
2026-05-15 |
| PR |
0.00 |
add DeepSeek-V4-Flash-Base support, also add the testcase(de |
|
Highly terse, with typos and informal la |
2026-05-11 |
| PR |
0.00 |
Revert 45777 |
|
Casual justification, mentions a specifi |
2026-05-13 |
| PR |
0.00 |
no empty label when Grounding Dino detects nothing |
|
Terse, informal, code context; no AI sig |
2026-05-14 |
| PR |
0.00 |
[docs] chat template prefill |
|
Informal, domain-specific, no AI indicat |
2026-05-13 |
| PR |
0.00 |
🚨🚧 FeatureExtractor → AudioProcessor |
|
Checklist, project-specific, human-like |
2026-03-02 |
| PR |
0.00 |
Expose `per_layer_inputs` for every Gemma4 variants |
|
Technical, direct, filled template, no A |
2026-05-13 |
| PR |
0.00 |
chore: update pr_build_doc_with_comment.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update build-ci-docker-images.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update assign-reviewers.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update circleci-failure-summary-comment.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update circleci-failure-summary-comment.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update anti-slop.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update pr_slow_ci_suggestion.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update benchmark_v2.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update build-ci-docker-images.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update assign-reviewers.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update build-ci-docker-images.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update trl-ci-bot.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
chore: update assign-reviewers.yml |
|
— |
2026-05-14 |
| PR |
0.00 |
avoid divid zero errors. |
|
Typo in title and informal language; no |
2025-08-27 |
| PR |
0.00 |
fix(rope): read original_max_position_embeddings from yarn v |
|
Technical phrasing and incomplete senten |
2026-05-11 |
| PR |
0.00 |
[docs] ALMModelTest |
|
Minimal, context-specific entry in templ |
2026-05-11 |
| PR |
0.00 |
[docs] new attention mask helpers |
|
Minimal template content, informal style |
2026-05-13 |
| PR |
0.00 |
Fix OLMo 3 scaled RoPE handling for sliding attention |
|
Human-style summary, technical context, |
2026-05-13 |
| PR |
0.00 |
Enhance apply_chat_template to support custom field prefilli |
|
Direct and technical enhancement summary |
2026-05-11 |
