| commit |
1.00 |
Add Molmo2 (#43451) |
|
Commit message contains explicit AI assi |
2026-10-09 |
| commit |
1.00 |
Aria, Ernie4.5-VL-MoE, Gemma4, Inkling, MiniMax-M3-VL, Qwen3 |
|
Commit message contains explicit AI assi |
2026-10-09 |
| commit |
1.00 |
Fix CI slack reporting: surface crashed/missing model jobs i |
|
Commit message contains explicit AI assi |
2026-10-09 |
| commit |
1.00 |
Optimize Whisper generation for TPU and device sync (#49426) |
|
Commit message contains explicit AI assi |
2026-10-08 |
| commit |
1.00 |
Complete PR #49228: remove unused device=None from rope __in |
|
Commit message contains explicit AI assi |
2026-10-08 |
| commit |
1.00 |
Fix data-dependent branch in _ignore_causal_mask_sdpa under |
|
Commit message contains explicit AI assi |
2026-10-08 |
| commit |
1.00 |
[Model] Add PP-DocLayoutV4 Model Support (#48387) |
|
Commit message contains explicit AI assi |
2026-10-08 |
| commit |
1.00 |
PR CI: grant actions: read to the reusable workflow (#49417) |
|
Commit message contains explicit AI assi |
2026-10-08 |
| commit |
1.00 |
[tests] Fix flaky CMVN tests in Speech2TextFeatureExtraction |
|
Commit message contains explicit AI assi |
2026-10-07 |
| commit |
1.00 |
[Docker] Upgrade to torch 2.14/CUDA 13.0 and optimize FA2 bu |
|
Commit message contains explicit AI assi |
2026-10-07 |
| commit |
1.00 |
Fix DataCollatorForLanguageModeling ignoring seed=0 (#49349) |
|
Commit message contains explicit AI assi |
2026-10-06 |
| commit |
1.00 |
GPT-OSS: read the clamped SwiGLU's alpha and limit from the |
|
Commit message contains explicit AI assi |
2026-10-06 |
| commit |
1.00 |
Bump doc-builder main docs workflow pin (#49350) |
|
Commit message contains explicit AI assi |
2026-10-06 |
| commit |
1.00 |
Fix run_glue.py label ids becoming strings with pandas>=3 (# |
|
Commit message contains explicit AI assi |
2026-10-05 |
| commit |
1.00 |
Fix chunked prefill with multi-axis position ids (Qwen3-VL) |
|
Commit message contains explicit AI assi |
2026-10-05 |
| commit |
1.00 |
Do RoPE with mul instead of matmul (#49265) |
|
Commit message contains explicit AI assi |
2026-10-05 |
| commit |
1.00 |
Enable model test on upcoming `torch_tpu` backend (#49207) |
|
Commit message contains explicit AI assi |
2026-10-05 |
| commit |
1.00 |
[CI] Avoid fetching all artifacts when only specific ones ar |
|
Commit message contains explicit AI assi |
2026-10-03 |
| commit |
1.00 |
[CI] check_bad_commit: use EFS cache to avoid Xet FUSE OOM ( |
|
Commit message contains explicit AI assi |
2026-10-02 |
| commit |
1.00 |
[ESM/Persimmon/NllbMoe/ESMFold] use safetensors repos to fix |
|
Commit message contains explicit AI assi |
2026-10-01 |
| pr |
1.00 |
[RFC] Add configurable decoder layer execution (reference im |
|
PR body explicitly mentions AI collabora |
2026-10-10 |
| pr |
1.00 |
Add TDT loss kernel |
|
Explicitly states that it was generated |
2026-05-19 |
| pr |
1.00 |
Fix uniform video frame index sampling |
|
PR body explicitly mentions AI collabora |
2026-10-10 |
| pr |
1.00 |
Fix MetalLinear batched inputs by flattening before kernel d |
|
PR body explicitly mentions AI collabora |
2026-10-07 |
| pr |
1.00 |
Add LFM2 to the GGUF loading path |
|
PR body explicitly mentions AI collabora |
2026-10-07 |
| pr |
1.00 |
Deprecate unused get_uniform_frame_indices |
|
Explicitly states that an AI assistant h |
2026-10-09 |
| pr |
1.00 |
[generate] Opt-in normalized repetition penalty |
|
PR body explicitly mentions AI collabora |
2026-09-09 |
| pr |
1.00 |
Fix uniform frame indices 49410 upstream |
|
PR body explicitly mentions AI collabora |
2026-10-09 |
| pr |
1.00 |
Trainer: retain unused-parameter detection with non-reentran |
|
PR body explicitly mentions AI collabora |
2026-10-09 |
| pr |
1.00 |
[Gemma4 Assistant] Force is_causal=False for bidirectional a |
|
PR body explicitly mentions AI collabora |
2026-10-09 |
| pr |
1.00 |
docs: update quantized cache hqq/quanto examples and documen |
|
PR body explicitly mentions AI collabora |
2026-07-09 |
| pr |
1.00 |
Add Kolibri 1 |
|
PR body explicitly mentions AI collabora |
2026-10-09 |
| pr |
1.00 |
Add GLiNER2 |
|
The description explicitly states it was |
2026-10-08 |
| pr |
1.00 |
Add `force_gqa` flag in `sdpa_attention_forward` |
|
PR body explicitly mentions AI collabora |
2026-10-09 |
| pr |
1.00 |
[MiMo-V2-Flash] Count the query token in the sliding window |
|
PR body explicitly mentions AI collabora |
2026-09-28 |
| pr |
1.00 |
feat: Add/update AGENTS.md for AI agent guidance |
|
Explicitly states the content was genera |
2026-10-09 |
| pr |
1.00 |
[Generation] Fix WatermarkDetector batch ordering bug with m |
|
PR body explicitly mentions AI collabora |
2026-10-05 |
| pr |
1.00 |
[vLLM] Add Qwen-Omni Thinker base models and Gemma 3n per-la |
|
The description explicitly states that A |
2026-10-08 |
| pr |
1.00 |
Add Intel XPU CI |
|
PR body explicitly mentions AI collabora |
2026-10-02 |
| pr |
1.00 |
[Model] add Step 3.7 Flash |
|
PR body explicitly mentions AI collabora |
2026-06-11 |
| pr |
1.00 |
Update sliding window indexing in continuous batching |
|
PR body explicitly mentions AI collabora |
2026-08-12 |
| pr |
1.00 |
Fall back to model type for unresolvable image processor nam |
|
PR body explicitly mentions AI collabora |
2026-09-05 |
| pr |
1.00 |
Fix TextGenerationPipeline missing tokenizer for stop_string |
|
PR body explicitly mentions AI collabora |
2026-09-06 |
| pr |
1.00 |
Expose last_hidden_state in Gemma4ForConditionalGeneration |
|
PR body explicitly mentions AI collabora |
2026-09-15 |
| pr |
1.00 |
Composite lm head tp plan |
|
PR body explicitly mentions AI collabora |
2026-09-24 |
| pr |
1.00 |
Ernie4.5-VL-MoE: shard the text and vision experts under TP |
|
PR body explicitly mentions AI collabora |
2026-10-09 |
| pr |
1.00 |
🚨 MoE: shard the experts in every TP/EP plan, and test for i |
|
PR body explicitly mentions AI collabora |
2026-09-30 |
| pr |
1.00 |
Support quantization in Dynamo, ONNX, OpenVINO and ExecuTorc |
|
PR body explicitly mentions AI collabora |
2026-08-03 |
| pr |
1.00 |
Aria, Ernie4.5-VL-MoE, Gemma4, Inkling, MiniMax-M3-VL, Qwen3 |
|
PR body explicitly mentions AI collabora |
2026-10-07 |
| pr |
1.00 |
Add dots3-note Preview model support |
|
PR body explicitly mentions AI collabora |
2026-08-08 |
| pr |
1.00 |
fix(whisper): auto-cast input_features to encoder dtype to f |
|
The description explicitly discloses Cod |
2026-08-06 |
| pr |
1.00 |
[vLLM] Add Omni token counting and text replacement offsets |
|
Explicitly acknowledges AI assistance in |
2026-10-08 |
| pr |
1.00 |
[Model] Add native Ling 3.0 Flash support |
|
PR body explicitly mentions AI collabora |
2026-10-09 |
| pr |
0.86 |
Fix Trainer checkpoint rotation matching backup directories |
|
Highly polished HTML summary uses generi |
2026-10-09 |
| pr |
0.82 |
Add VideoToTextPipeline with smart frame sampling and system |
|
Polished, generic feature-summary langua |
2025-11-26 |
| pr |
0.72 |
fix(processing): Filter kwargs in ProcessorMixin call to pre |
|
Highly polished explanatory prose uses g |
2025-10-15 |
| pr |
0.68 |
Fix: add None check for extractors in video_processor_class_ |
|
Highly polished explanatory sections use |
2025-12-10 |
| pr |
0.68 |
Deprecate `max_size` in ConditionalDetrImageProcessor with w |
|
Polished headings and generic migration |
2025-09-20 |
| pr |
0.30 |
Preserve positional exist_ok in AutoVideoProcessor.register |
|
Polished and exhaustive, though specific |
2026-10-09 |
| pr |
0.25 |
Fix Trainer prediction handling with context parallelism |
|
Detailed technical prose but no clear st |
2026-10-10 |
| pr |
0.22 |
Serialize MPS transfers in async model loading (fix #48029) |
|
Highly polished debugging report, but gr |
2026-09-28 |
| pr |
0.20 |
Make sinusoidal position embeddings compileable |
|
Careful technical explanation with model |
2026-10-10 |
| pr |
0.20 |
Keep AutoVideoProcessor class lookup lazy |
|
Polished and detailed, but lacks distinc |
2026-10-09 |
| pr |
0.08 |
docs: clarify GPU memory requirements in run_clm example |
|
Concise, domain-specific documentation c |
2026-10-10 |
| pr |
0.08 |
fix: check GPU architecture before enabling GQA in SDPA on C |
|
Concise, specific performance issue and |
2026-09-09 |
| pr |
0.08 |
Gate enable_gqa=True on actual flash-attention eligibility |
|
Detailed domain-specific discussion and |
2026-05-04 |
| pr |
0.08 |
Support custom kernels for processing ops |
|
Detailed technical explanation includes |
2026-06-19 |
| pr |
0.08 |
Move processing defaults to class |
|
Detailed domain-specific discussion and |
2026-09-24 |
| pr |
0.08 |
Fix Qwen3-VL Processor flattening multi-image batches (fix # |
|
Concise, specific change summary is cons |
2025-10-18 |
| pr |
0.08 |
Fix SinglePositionMultiTokenCandidateGenerator signature |
|
Specific API details, reproduction steps |
2026-10-08 |
| pr |
0.07 |
Add WeatherNext 2 weather forecasting model |
|
Highly technical, specific details and a |
2026-08-10 |
| commit |
0.05 |
:rotating_light: SubConfigSpec for centralized init of subco |
|
Terse, typo-filled commit notes strongly |
2026-10-08 |
| pr |
0.05 |
Use one static Transformers CI badge in PR descriptions |
|
Polished but highly specific implementat |
2026-10-09 |
| pr |
0.05 |
Update metal quant kernels |
|
Informal wording and domain-specific det |
2026-10-09 |
| pr |
0.05 |
Fix loading legacy remote-code models without `all_tied_weig |
|
Detailed model-loading context and concr |
2026-10-09 |
| commit |
0.04 |
Generation cleanup in pipelines (#49269) |
|
Short series of concrete cleanup notes, |
2026-10-08 |
| pr |
0.04 |
Keep packed sequences apart in the torch references of the c |
|
Specific implementation details, benchma |
2026-10-09 |
| pr |
0.04 |
Raise when a torch fallback would drop packed-sequence bound |
|
Concrete bug behavior, package reference |
2026-10-09 |
| pr |
0.04 |
Fix CI slack reporting: surface crashed/missing model jobs i |
|
Detailed CI-specific failure modes and i |
2026-10-08 |
| pr |
0.04 |
Fix MusicGen forward(labels=...) when decoder.decoder_start_ |
|
Specific regression details, tests, and |
2026-09-25 |
| pr |
0.04 |
Don't warn about class defaults when diffing a config |
|
Detailed domain-specific debugging narra |
2026-07-28 |
| pr |
0.04 |
Support static cache in assisted decoding |
|
Detailed implementation caveats and benc |
2026-10-08 |
| pr |
0.04 |
Fix a port allocation race in FSDP tests |
|
Specific race details, commands, and tes |
2026-10-09 |
| commit |
0.03 |
[Qwen3VLMoe] Update expected text in test_small_model_integr |
|
Specific implementation details and a no |
2026-10-08 |
| commit |
0.03 |
This PR fixes Gemma3n fp16 loading when missing weights need |
|
Detailed domain-specific failure analysi |
2026-10-08 |
| commit |
0.03 |
[`distributed`] Add expert-parallel token dispatch, default |
|
Domain-specific distributed-systems deta |
2026-10-06 |
| commit |
0.03 |
[`distributed`] decoupled tp_plan and ep_plan (#48859) |
|
Specific implementation details, typos, |
2026-10-06 |
| commit |
0.03 |
[`distributed`] Add dense and expert device mesh (#48857) |
|
Technical mesh terminology and informal |
2026-10-06 |
| commit |
0.03 |
🚨 Return router logits from every MoE model that computes th |
|
Highly domain-specific changelog with it |
2026-10-05 |
| commit |
0.03 |
Add Trainer.loss_is_scaled_for_ga to declare whether compute |
|
Technical, repository-specific wording a |
2026-10-02 |
| pr |
0.03 |
Stop treating HunYuan's DynamicNTKAlpha RoPE as dynamic |
|
Highly specific technical diagnosis and |
2026-10-09 |
| pr |
0.03 |
Fix BLOOM crash with static cache / compiled generate |
|
Informal phrasing and typo strongly sugg |
2026-10-10 |
| pr |
0.03 |
Use the Hub rotary kernel for inference only on CUDA |
|
Highly specific kernel behavior, benchma |
2026-10-08 |
| pr |
0.03 |
[DON'T MERGE YET] [tests] Some fixes for test suite failures |
|
Informal grammar and TPU-specific implem |
2026-10-08 |
| pr |
0.03 |
Guard the checks for current_accelerator() and bump Torch ve |
|
Concise release-specific explanation and |
2026-10-07 |
| pr |
0.03 |
fix: Pass the MTP layer type to per-layer-type rotary embedd |
|
Concise, specific bug description with n |
2026-10-09 |
| pr |
0.03 |
Uniformize model processors (models w/o special arg names) |
|
Technical model list, terse notes, and i |
2024-08-16 |
| pr |
0.03 |
Uniformize model processors (models *with* special arg names |
|
Specific compatibility details and infor |
2024-08-16 |
| pr |
0.03 |
handle get_input_embeddings() on models like gemma3 graceful |
|
Brief, informal technical description wi |
2025-12-02 |
| pr |
0.03 |
Move the MI300 docker image to ROCm 10 and update test expec |
|
Concise release-change description uses |
2026-10-09 |
| pr |
0.03 |
:rotating_light: Kernels=True ?! |
|
Informal shorthand, emojis, and concrete |
2026-08-04 |
| commit |
0.02 |
Second phase of torch 2.6 cleanups (#49396) |
|
Terse, repository-specific cleanup title |
2026-10-09 |
| commit |
0.02 |
fix: treat every declared field of the config class as a def |
|
Repetitive fix wording is typical of a c |
2026-10-09 |
| commit |
0.02 |
[Atlas Inference] Enable torch.compile for the Strix Halo Ga |
|
Terse, duplicated technical title and de |
2026-10-09 |
| commit |
0.02 |
[Atlas Inference] Fix GB10 Gated DeltaNet hub kernel for tra |
|
Brief, specific kernel fix description w |
2026-10-09 |
| commit |
0.02 |
More RoPE simplification, TF32 fixes (#49328) |
|
Brief technical commit title with reposi |
2026-10-08 |
| commit |
0.02 |
Fix Inkling moe_intermediate_size when dense_intermediate_si |
|
Specific configuration fix and brief exp |
2026-10-08 |
| commit |
0.02 |
[Bugfix] [Modeling] Fix DSV4 norm (#47486) |
|
Brief technical commits and Signed-off-b |
2026-10-08 |
| commit |
0.02 |
Smthn smthn (#49364) |
|
Highly specific implementation details, |
2026-10-06 |
| commit |
0.02 |
Move CI to Python 3.11, with the version set in `.python-ver |
|
Technical CI details and terse commit hi |
2026-10-06 |
| commit |
0.02 |
Allow registering new processing backends (#48984) |
|
Informal terse notes and concrete debugg |
2026-10-06 |
| commit |
0.02 |
remove colwise_gather_output for lm_head for DeepseekV4 (#4 |
|
Terse, repository-specific revert notes |
2026-10-06 |
| commit |
0.02 |
Fix quantized cache (#48700) |
|
Detailed implementation and test notes a |
2026-10-05 |
| commit |
0.02 |
🚨 Select the OWLv2 image query by objectness (#49200) |
|
Specific technical change with a normal |
2026-10-02 |
| commit |
0.02 |
Support per-layer cache configuration (#48178) |
|
Long iterative history contains specific |
2026-10-02 |
| commit |
0.02 |
Remove rotaries deprecation cycle (#49228) |
|
Terse, informal commit-style wording sug |
2026-10-02 |
| commit |
0.02 |
[CB] Make CB more device agnostic and add support for XPU (# |
|
Detailed domain-specific summary and ite |
2026-10-01 |
| pr |
0.02 |
fix bridgetower cross-attn mask shape bug |
|
Specific debugging details, error output |
2026-10-09 |
| pr |
0.02 |
[`GradientCkpting`] Fix non grad arg inputs |
|
Informal phrasing, typos, and concrete i |
2026-10-07 |
| pr |
0.02 |
Fix Gemma 4 tool call parsing for names containing - or . |
|
Brief, specific fix description with an |
2026-10-08 |
| pr |
0.02 |
Update utils.py: fix nan |
|
Terse, informal wording and reviewer dis |
2025-08-14 |
| pr |
0.02 |
Revert text_positions in Qwen25VL |
|
Brief technical description with issue r |
2025-08-14 |
| pr |
0.02 |
Uniformize kwargs for Layoutlm (2, 3, X) processors |
|
Concise technical changelog with project |
2024-07-24 |
| pr |
0.02 |
Adds uniform processing kwargs to paligemma. |
|
Terse, domain-specific wording and minim |
2024-08-01 |
| pr |
0.02 |
fix: qwen2.5 omni apply_chat_template system content check |
|
Repetition, placeholders, and terse word |
2025-04-15 |
| pr |
0.02 |
Fixes StaticCache Crashes |
|
Terse issue-focused text with benchmark |
2025-11-28 |
| pr |
0.02 |
Count ep and fsdp in caching_allocator_warmup byte calculati |
|
Terse informal wording and specialized t |
2026-10-07 |
| commit |
0.01 |
Use the Hub rotary kernel for inference only on CUDA (#49439 |
|
Detailed kernel and gradient behavior is |
2026-10-09 |
| commit |
0.01 |
Guard the checks for current_accelerator() and bump Torch ve |
|
Brief technical description with specifi |
2026-10-09 |
| commit |
0.01 |
Use one static Transformers CI badge in PR descriptions (#49 |
|
Concise, concrete CI change phrased like |
2026-10-09 |
| commit |
0.01 |
Keep AutoVideoProcessor class lookup lazy (#49458) |
|
Short, specific change with a plausible |
2026-10-09 |
| commit |
0.01 |
[MiMo-V2-Flash] Count the query token in the sliding window |
|
Detailed model-specific reasoning and it |
2026-10-09 |
| commit |
0.01 |
Ci fetches multimodal models recursively (#49389) |
|
Terse implementation notes and repositor |
2026-10-09 |
| commit |
0.01 |
Recurse for all subconfigs (#49461) |
|
Minimal technical commit message with no |
2026-10-09 |
| commit |
0.01 |
Fix backend attribute in documentation and default backend i |
|
Terse, repetitive commit wording is char |
2026-10-08 |
| commit |
0.01 |
Revert "[CI] Temporarily disable AMD scheduled CI trigger" ( |
|
Standard revert message generated by Git |
2026-10-08 |
| commit |
0.01 |
Fix QuantizedCache residual flushing for multi-token updates |
|
Concise, specific bug-fix wording with n |
2026-10-08 |
| commit |
0.01 |
fix bug for DebertaV2ForMaskedLM non-legacy load issue (#491 |
|
Typos, terse wording, and signed-off tra |
2026-10-08 |
| commit |
0.01 |
Fix Whisper batch timestamp decoding for tensors (#49419) |
|
Concise technical commit title with no A |
2026-10-08 |
| commit |
0.01 |
Do not wrap the model in FSDP for expert-parallel token disp |
|
Specific distributed-systems jargon and |
2026-10-08 |
| commit |
0.01 |
[docs] Update Spaces (#49242) |
|
Terse commit-style bullets are typical h |
2026-10-07 |
| commit |
0.01 |
handle empty precompiled charsmap in spm converter (#49211) |
|
Typos and terse wording strongly suggest |
2026-10-07 |
| commit |
0.01 |
🚨🚨🚨 luke fixe (#48408) |
|
Informal iterative commit notes and a hu |
2026-10-07 |
| commit |
0.01 |
Let token dispatch run with `ep_size != tp_size` (#49368) |
|
Terse technical title and standard human |
2026-10-07 |
| commit |
0.01 |
[tests] Skip onnx.exactness for BitForImageClassification (O |
|
Specific ORT behavior, tolerance, and mo |
2026-10-06 |
| commit |
0.01 |
Link the padding side section from SSM and hybrid model docs |
|
Concrete documentation references and it |
2026-10-06 |
| commit |
0.01 |
Revert "[CI] Temporarily disable AMD scheduled CI trigger" ( |
|
Short conventional revert description wi |
2026-10-06 |
| commit |
0.01 |
Fix explicit kernelization modes for evaluation models (#493 |
|
Brief technical fix with review iteratio |
2026-10-06 |
| commit |
0.01 |
fix(cohere_compass): add video to input_modalities (#49320) |
|
Conventional commit syntax and precise m |
2026-10-06 |
| commit |
0.01 |
[`distributed`] Default MoE `ep_plans` to token dispatch (#4 |
|
Detailed domain-specific history and ter |
2026-10-06 |
| commit |
0.01 |
[`distributed`]: adapt Trainer to work Expert Parallel (#488 |
|
Specific distributed-training details an |
2026-10-06 |
| commit |
0.01 |
Bump doc-builder workflow pin (#49342) |
|
Short, specific dependency-pin updates a |
2026-10-06 |
| commit |
0.01 |
Fix WatermarkDetector repeated-ngram counting (#49315) |
|
Specific regression and review reference |
2026-10-06 |
| commit |
0.01 |
Fix minicpm 4.6V video batching and CI (#49259) |
|
Very terse informal debugging notes stro |
2026-10-06 |
| commit |
0.01 |
Fix the `labels` docstring of CLIPSegForImageSegmentation (# |
|
Concise, technically specific documentat |
2026-10-05 |
| commit |
0.01 |
[CI] Temporarily disable AMD scheduled CI trigger (#49271) |
|
Brief CI change with a normal co-author |
2026-10-05 |
| commit |
0.01 |
Remove the unused pandas entry from setup.py deps (#49329) |
|
Specific dependency and CI-version detai |
2026-10-05 |
| commit |
0.01 |
[Fix] Remove old test file with two ancient tests (#49280) |
|
Short, informal commit sequence typical |
2026-10-05 |
| commit |
0.01 |
Fix multi-image feature ordering during generation batch exp |
|
Concise fix description with a normal co |
2026-10-05 |
| commit |
0.01 |
Why override when you can fix in `Mixin` (#49105) |
|
Fragmentary informal commit notes strong |
2026-10-05 |
| commit |
0.01 |
Deprecated pipelines should redirect; removed pipelines shou |
|
Brief review-driven changelog with a nor |
2026-10-05 |
| commit |
0.01 |
Fix encoder repetition penalty source rows after batch expan |
|
Terse, domain-specific commit message wi |
2026-10-05 |
| commit |
0.01 |
Preserve masked EOS scores in exponential decay length penal |
|
Concise technical commit message; no sig |
2026-10-05 |
| commit |
0.01 |
Revert "[CI] ssh-runner: add optional cache_type input to sw |
|
Concise revert message with a specific c |
2026-10-02 |
| commit |
0.01 |
Don't forward text-level flash-attention kwargs to multimoda |
|
Terse commit notes and an informal CI re |
2026-10-02 |
| commit |
0.01 |
[`GTE`] Add memory mixin (#49234) |
|
Terse, informal wording is characteristi |
2026-10-01 |
| commit |
0.01 |
[docs] TP for PEFT adapters (#49060) |
|
Brief, informal commit notes lack AI-sty |
2026-10-01 |
| commit |
0.01 |
correct final metrics due to resuming from checkpoint (#4920 |
|
Concise wording is typical of a human co |
2026-10-01 |
| commit |
0.01 |
🚨 Attention 🚨 Deprecate "paged|" prefix for SDPA and flash |
|
Terse commit sequence and domain-specifi |
2026-10-01 |
| commit |
0.01 |
🚨 Attention 🚨 Make regular attention support CB (#49101) |
|
Informal review notes, typos, and co-aut |
2026-10-01 |
| commit |
0.01 |
🚨 [CB] 🚨 Fuse update for index and block table path (#49088) |
|
Minimal phrase is a typical human squash |
2026-10-01 |
| commit |
0.01 |
[Refactor] Make some flash-attention utils more readable (#4 |
|
Short technical change list and informal |
2026-10-01 |
| pr |
0.01 |
Second phase of torch 2.6 cleanups |
|
Very brief and informal stacked-PR summa |
2026-10-07 |
| pr |
0.01 |
Remove imports from other models |
|
Minimal, casual description with reposit |
2026-10-09 |
| pr |
0.01 |
fix: treat every declared field of the config class as a def |
|
Very terse and informal description is c |
2026-10-09 |
| pr |
0.01 |
Add Molmo2 |
|
Mostly template content with no substant |
2026-01-23 |
| pr |
0.01 |
fix colpali mapping |
|
Contains a casual typo and specific proj |
2025-07-11 |
| pr |
0.01 |
Updating Chameleon Image handling documentation |
|
Concise, specific documentation request |
2024-09-24 |
| pr |
0.01 |
Fix audio-related config naming for Gemma3n |
|
Specific technical context and informal |
2025-06-28 |
| pr |
0.01 |
fix immediate quantization of the first token in QuantizedCa |
|
Brief, issue-specific commit-style descr |
2025-01-17 |
| pr |
0.01 |
fix qwen2.5-omini cant be loaded from AutoModel |
|
Informal phrasing and typos strongly sug |
2025-04-25 |
| pr |
0.01 |
Allow qwen/emu3 to process low res images |
|
Specific benchmark context and typo indi |
2025-05-15 |
| pr |
0.01 |
remove vision2seq vs image-text-to-text |
|
Informal reference to internal chat and |
2025-09-04 |
| pr |
0.01 |
Ci fetches multimodal models recursively |
|
Terse implementation note uses repositor |
2026-10-07 |
| pr |
0.01 |
Recurse for all subconfigs |
|
Very brief, informal description with a |
2026-10-09 |
| pr |
0.01 |
Image decoding moves to torchcodec |
|
Informal deferment note and concrete ver |
2026-08-14 |
| pr |
0.01 |
Fix `create_chunked_causal_mask()` crashing with `block_sequ |
|
Brief, repository-specific bug descripti |
2026-10-09 |
| commit |
0.00 |
Image decoding moves to torchcodec (#47963) |
|
Commit history is terse and includes a n |
2026-10-09 |
| commit |
0.00 |
[CI] Temporarily disable AMD scheduled CI trigger (#49394) |
|
Concise revert message with a specific c |
2026-10-07 |
| commit |
0.00 |
Fix Phi3 LongRoPE cache rebuild at the context boundary (#49 |
|
A terse, specific bug-fix title with no |
2026-10-07 |
| commit |
0.00 |
Fix Pixtral Flash Attention image boundaries after axial RoP |
|
Specific regression details and signed-o |
2026-10-07 |
| commit |
0.00 |
ALM testx fIxing... (#49355) |
|
Fragmentary typo-filled message is stron |
2026-10-06 |
| commit |
0.00 |
Fix pp chart2table inference (#49225) |
|
Informal typo-filled notes are character |
2026-10-02 |
| commit |
0.00 |
fix precomputed flash kwargs clashing (#47928) |
|
Extremely terse and informal commit mess |
2026-10-02 |
| commit |
0.00 |
Reapply modular examples (#49230) |
|
Brief, informal commit notes with no AI- |
2026-10-02 |
| commit |
0.00 |
Remove some stuff that does not exist (#49232) |
|
Terse wording and informal phrasing indi |
2026-10-02 |
| commit |
0.00 |
Remove deprecation cycle on mask and cache (#49231) |
|
Minimal, informal revert description. |
2026-10-02 |
| pr |
0.00 |
Smthn smthn |
|
Extremely terse and informal, with no in |
2026-10-06 |
| pr |
0.00 |
[serge] Fix 2 integration tests for model `qwen2_moe` failin |
|
— |
2026-10-10 |
| pr |
0.00 |
Olmo3 CI |
|
Extremely terse and informal wording str |
2026-07-08 |