| commit |
1.00 |
[Glm 5.3 Flash] GLM 5.3 Flash Support (#48342) |
|
The commit history explicitly mentions e |
2026-08-26 |
| commit |
1.00 |
Fix kernel commit and repo paths for ESMFold2 (#48186) |
|
Commit message contains explicit AI assi |
2026-08-26 |
| commit |
1.00 |
[debug] Trace previous CI run selection in get_previous_dail |
|
Commit message contains explicit AI assi |
2026-08-26 |
| commit |
1.00 |
Normalize HunYuanVL's legacy field aliases via attribute_map |
|
Commit message contains explicit AI assi |
2026-08-26 |
| commit |
1.00 |
[Gemma4] Investigate flaky test_generation_beyond_sliding_wi |
|
Commit message contains explicit AI assi |
2026-08-25 |
| commit |
1.00 |
Fix nemotron_h save_pretrained emitting singular backbone.em |
|
Commit message contains explicit AI assi |
2026-08-24 |
| commit |
1.00 |
[Gemma4] Fix stale expected values in integration tests (#48 |
|
Commit message contains explicit AI assi |
2026-08-23 |
| commit |
1.00 |
[TableTransformer, PI0] Fix stale expected values and OOM in |
|
Commit message contains explicit AI assi |
2026-08-21 |
| commit |
1.00 |
Fix stale expected values in integration tests (cuda sm_86 / |
|
Commit message contains explicit AI assi |
2026-08-21 |
| commit |
1.00 |
Fix `gpt_oss` runs on GPU (#48118) |
|
Commit message contains explicit AI assi |
2026-08-21 |
| commit |
1.00 |
[EsmFold2] Fix stale expected distogram logit values (#48182 |
|
Commit message contains explicit AI assi |
2026-08-21 |
| commit |
1.00 |
Fix Apr 05 integration test regressions (cuda sm_86) (#48170 |
|
Commit message contains explicit AI assi |
2026-08-21 |
| commit |
1.00 |
Fix integration test expected values for cuda sm_86 (Mar 15 |
|
Commit message contains explicit AI assi |
2026-08-21 |
| commit |
1.00 |
[LLaVA] Fix pixtral integration tests for cuda sm_86 (#48166 |
|
Commit message contains explicit AI assi |
2026-08-20 |
| commit |
1.00 |
Port ESMC and ESMFold2 to Transformers (#46419) |
|
Commit message contains explicit AI assi |
2026-08-20 |
| commit |
1.00 |
[InternVL] Fix stale expected values for Llama integration t |
|
Commit message contains explicit AI assi |
2026-08-20 |
| commit |
1.00 |
Retry transient network errors (RemoteDisconnected) in githu |
|
Commit message contains explicit AI assi |
2026-08-20 |
| commit |
1.00 |
[Whisper] Fix speculative decoding: preserve cleared suppres |
|
Commit message contains explicit AI assi |
2026-08-20 |
| commit |
1.00 |
Enable mlinter findings artifact for inline PR reviews (#481 |
|
Commit message contains explicit AI assi |
2026-08-19 |
| commit |
1.00 |
Delete old mlinter review comments before posting new ones ( |
|
Commit message contains explicit AI assi |
2026-08-19 |
| commit |
1.00 |
Fix mlinter artifact path (#48088) |
|
Commit message contains explicit AI assi |
2026-08-19 |
| commit |
1.00 |
[Whisper] Fix decoder position IDs for left-padded batches i |
|
Commit message contains explicit AI assi |
2026-08-19 |
| commit |
1.00 |
[PE] Skip test_sdpa_can_dispatch_on_flash for TimmWrapper-ba |
|
Commit message contains explicit AI assi |
2026-08-18 |
| commit |
1.00 |
[Gemma3] Update integration test expected values for A10G (# |
|
Commit message contains explicit AI assi |
2026-08-18 |
| pr |
1.00 |
Vision (2d/3d) rotary embeddings |
|
The author explicitly states that AI was |
2026-08-19 |
| pr |
1.00 |
Add support for Nemotron Omni |
|
PR body explicitly mentions AI collabora |
2026-06-08 |
| pr |
1.00 |
Fix SwitchTransformers router outputs |
|
PR body explicitly mentions AI collabora |
2026-08-25 |
| pr |
1.00 |
[debug] BigBirdPegasus gradient checkpointing flakiness |
|
Explicitly states the text was generated |
2026-08-26 |
| pr |
1.00 |
Ignore a stale best checkpoint recorded in a resumed trainer |
|
PR body explicitly mentions AI collabora |
2026-08-25 |
| pr |
1.00 |
Batch Rebalance Data Sampler |
|
PR body explicitly mentions AI collabora |
2026-07-15 |
| pr |
1.00 |
[qwen4_exp] disable torch/onnx export tests due to data-depe |
|
Explicitly states it was generated with |
2026-08-26 |
| pr |
1.00 |
Raise the intended diagnostic when the stop string probe can |
|
Explicitly discloses development with AI |
2026-08-26 |
| pr |
1.00 |
fix precomputed flash kwargs clashing |
|
PR body explicitly mentions AI collabora |
2026-08-12 |
| pr |
1.00 |
Add Unlimited OCR |
|
PR body explicitly mentions AI collabora |
2026-06-23 |
| pr |
1.00 |
Unified multimodal mixin (mrope) |
|
PR body explicitly mentions AI collabora |
2026-08-14 |
| pr |
1.00 |
Fix infeasible cost matrix errors in hugarian matcher losses |
|
PR body explicitly mentions AI collabora |
2026-08-03 |
| pr |
1.00 |
Assert cached decode matches recomputing without a cache |
|
PR body explicitly mentions AI collabora |
2026-08-25 |
| pr |
1.00 |
Add an opt-in per-frame pixel cap (cap_pixels_per_frame) to |
|
PR body explicitly mentions AI collabora |
2026-08-18 |
| pr |
1.00 |
Add shared ImageProcessingTester |
|
PR body explicitly mentions AI collabora |
2026-08-03 |
| pr |
1.00 |
[Improvement] Make gated delta rule more explicit and suppor |
|
PR body explicitly mentions AI collabora |
2026-07-29 |
| pr |
1.00 |
Fix Moshi output fields |
|
PR body explicitly mentions AI collabora |
2026-08-25 |
| pr |
1.00 |
Narrow Sentencepiece deserialization exception handling |
|
PR body explicitly mentions AI collabora |
2026-08-26 |
| pr |
1.00 |
Stop pairing two-word pretokenized documents in the python t |
|
PR body explicitly mentions AI collabora |
2026-08-25 |
| pr |
0.92 |
Strip the metaspace prefix space only when it is unambiguous |
|
Reviewer feedback explicitly identifies |
2026-08-10 |
| pr |
0.78 |
Add native TR-HASH model support |
|
Highly polished, exhaustive structure an |
2026-08-21 |
| pr |
0.72 |
Restore BC for the tensor-parallel API |
|
Highly polished, exhaustive structure an |
2026-08-25 |
| pr |
0.68 |
[conftest] Use get_cpu_ram_total_gib for psutil patch (cgrou |
|
Polished explanatory prose and generic r |
2026-08-25 |
| pr |
0.68 |
Add support for Lumma Model |
|
Polished overview and unusually deferent |
2026-08-12 |
| pr |
0.48 |
[ONNX] Skip affected models on torch 2.13 (two dynamo regres |
|
Highly polished summary with boilerplate |
2026-08-21 |
| pr |
0.35 |
[Glm 5.3 Flash] GLM 5.3 Flash Support |
|
The polished marketing-style model summa |
2026-08-26 |
| pr |
0.28 |
fix(mluke): resolve ent ids via vocab not positional index |
|
Structured wording feels somewhat agent- |
2026-08-25 |
| pr |
0.18 |
Fix `PhimoeIntegrationTest` |
|
Very polished explanation, but it contai |
2026-06-10 |
| commit |
0.08 |
Ignore a stale best checkpoint recorded in a resumed trainer |
|
Specific failure mode and test results i |
2026-08-26 |
| pr |
0.08 |
Add auto_docstring task model overrides |
|
Technical, repository-specific wording s |
2026-08-06 |
| pr |
0.08 |
Fix LlamaTokenizer overwriting pre_tokenizer and decoder fro |
|
Detailed domain-specific explanation and |
2026-08-22 |
| pr |
0.08 |
[debug] Trace previous CI run selection in get_previous_dail |
|
Specific CI details and debugging hypoth |
2026-08-26 |
| pr |
0.08 |
Fix interval merge invariant in _find_disjoint |
|
Polished and unusually thorough, but its |
2026-08-10 |
| pr |
0.08 |
Register activation kernel layers on XPU |
|
Specific XPU details, benchmarks, and de |
2026-08-10 |
| pr |
0.07 |
Restore .item() on the packed-path max_seqlen (third site mi |
|
Specific regression history, hardware de |
2026-08-26 |
| pr |
0.07 |
[generate] Fix CUDAGraphs tensor overwrite for encoder-decod |
|
Detailed domain-specific diagnosis is no |
2026-08-21 |
| pr |
0.06 |
[Qwen3-VL-MoE] Add native TP and FSDP sharding plan for expe |
|
Uses precise project terminology and con |
2026-08-26 |
| pr |
0.06 |
model: Add NVIDIA Canary-1B-v2 to Transformers |
|
Template-driven model summary with concr |
2026-06-22 |
| commit |
0.05 |
Fix `BayesianDetectorModel.from_pretrained()` by calling `po |
|
Specific debugging details and issue ref |
2026-08-25 |
| pr |
0.05 |
[`Kernels`] Add swiglu and geglu MLP across models |
|
Terse, domain-specific wording and infor |
2026-08-26 |
| pr |
0.05 |
fix(cli): migrate chat.py from requests to httpx to fix tran |
|
Polished structure, but it contains conc |
2026-08-25 |
| commit |
0.04 |
Create the continuous batching CPU group with local synchron |
|
Detailed distributed-systems explanation |
2026-08-26 |
| commit |
0.04 |
Resolve continuous batching config against the text config f |
|
Specific composite-model configuration d |
2026-08-26 |
| pr |
0.04 |
Fix `safe_open` mmap memory exhaustion on Windows by using ` |
|
Concise, specific engineering summary wi |
2026-08-26 |
| pr |
0.04 |
Add multimodal sequence classification support for Mistral 3 |
|
Detailed model-specific explanation with |
2026-08-21 |
| pr |
0.04 |
Normalize HunYuanVL's legacy field aliases via attribute_map |
|
Highly specific config and checkpoint de |
2026-08-24 |
| pr |
0.04 |
Cpmant fix use cache |
|
Detailed technical diagnosis is specific |
2026-08-17 |
| pr |
0.04 |
fix: decode() batch path respects self.clean_up_tokenization |
|
Concise, code-specific bug description w |
2026-08-05 |
| commit |
0.03 |
Keep MXFP4 weights quantized on XPU when use_kernels is set |
|
Concise domain-specific commit message w |
2026-08-26 |
| commit |
0.03 |
Restore BC for the tensor-parallel API (#48300) |
|
Technical details and informal notes lik |
2026-08-26 |
| commit |
0.03 |
Add an opt-in per-frame pixel cap (cap_pixels_per_frame) to |
|
Detailed domain-specific engineering rat |
2026-08-26 |
| commit |
0.03 |
Fix dtype mismatch in grouped_mm_fallback for LoRA training |
|
Technical specificity, review references |
2026-08-25 |
| commit |
0.03 |
🚨[wav2vec2] Support attn_implementation=sdpa dispatch (#4619 |
|
Technical incremental commit notes and d |
2026-08-20 |
| commit |
0.03 |
Always tie embeddings for LongT5 and Pop2Piano (#47620) |
|
Detailed, technically specific debugging |
2026-08-20 |
| commit |
0.03 |
Improve group images by shape (#47964) |
|
Very terse wording and a typo are strong |
2026-08-19 |
| commit |
0.03 |
fix(pipeline): preserve model.generation_config precedence o |
|
Concise technical fix with review feedba |
2026-08-18 |
| pr |
0.03 |
Shape cross-attention keys and values by the states they com |
|
Terse, highly technical description with |
2026-08-27 |
| pr |
0.03 |
Fix generate() crash when reusing image_hidden_states with p |
|
Detailed regression context and concrete |
2026-08-02 |
| pr |
0.03 |
[Quantization] Add MXFP4 support |
|
Technical details and informal phrasing |
2026-08-20 |
| pr |
0.03 |
Simplify qwen4 tests |
|
Brief informal wording is consistent wit |
2026-08-26 |
| pr |
0.03 |
fix(zamba): add use_associative_scan config flag to avoid t |
|
Specific implementation details and info |
2026-08-26 |
| pr |
0.03 |
Fix AutoTokenizer returning TokenizersBackend for DeepSeek-R |
|
Commit history, model details, and failu |
2026-08-22 |
| pr |
0.03 |
[Umbrella] Long-context SFT fixes (do not merge) |
|
Concrete benchmarks and repository refer |
2026-08-26 |
| pr |
0.03 |
[WIP] unlimited-ocr |
|
Sparse WIP text with placeholders and te |
2026-08-24 |
| commit |
0.02 |
Fix missing FP8 TP layer overrides (#48343) |
|
Terse, repository-specific change summar |
2026-08-26 |
| commit |
0.02 |
[qwen4_exp] disable torch/onnx export tests due to data-depe |
|
Concise technical change with a standard |
2026-08-26 |
| commit |
0.02 |
Add a regression test for force_accelerate_hooks signature |
|
Specific regression details, test behavi |
2026-08-26 |
| commit |
0.02 |
Add shared ImageProcessingTester (#47745) |
|
Iterative, repository-specific test chan |
2026-08-26 |
| commit |
0.02 |
Stop pairing two-word pretokenized documents in the python t |
|
Specific tokenizer behavior, tests, and |
2026-08-25 |
| commit |
0.02 |
Let gradient checkpointing skip layers with every_n_layers ( |
|
Technical change notes and standard co-a |
2026-08-25 |
| commit |
0.02 |
Fix `tie_word_embeddings` not lifted from `text_config` for |
|
Commit progression and repository-specif |
2026-08-25 |
| commit |
0.02 |
`force_accelerate_hooks` should not hide the signature it wr |
|
Extremely terse fix description, with no |
2026-08-24 |
| commit |
0.02 |
[Optimization] Avoid redundant token length initialization l |
|
Technical optimization title with concis |
2026-08-24 |
| commit |
0.02 |
[xcodec2] Fix flex attention and flash dispatch tests (#4824 |
|
Detailed domain-specific debugging expla |
2026-08-24 |
| commit |
0.02 |
Post two CI badges on a PR: CPU PR CI and GPU run-slow (#481 |
|
Specific CI behavior and issue reference |
2026-08-21 |
| commit |
0.02 |
[Qwen2.5-Omni] Update stale expected values for cuda sm_86 ( |
|
Concise, domain-specific commit text wit |
2026-08-20 |
| commit |
0.02 |
[Mistral3] Fix batched integration tests: padding_side=left |
|
Specific test and hardware details indic |
2026-08-20 |
| commit |
0.02 |
Use generator with seed for LengthGroupedSampler in Trainer. |
|
Specific implementation details and ters |
2026-08-19 |
| commit |
0.02 |
Fix Qwen2.5-Omni / Qwen3-Omni-MoE generation with a compilab |
|
Model-specific implementation details an |
2026-08-19 |
| commit |
0.02 |
feat: add nvfp4 quantization (#47883) |
|
Specific feature and documentation commi |
2026-08-18 |
| pr |
0.02 |
Quiet continuous batching at default verbosity |
|
Technical validation details and informa |
2026-08-25 |
| pr |
0.02 |
add HyperClovaX Vision |
|
Specific project context and somewhat aw |
2026-02-27 |
| pr |
0.02 |
Fix MTP generation test regex gate for escaped layer ignore |
|
Specific regex behavior and model names |
2026-08-24 |
| pr |
0.02 |
Fix fp8 dtype |
|
Concise description uses specific quanti |
2026-08-26 |
| pr |
0.02 |
[docs] Fix links and remove TokenizerFast |
|
Concise, repository-specific change list |
2026-08-03 |
| pr |
0.02 |
[generation] Encode multimodal data only once |
|
Specific implementation details and casu |
2026-05-05 |
| pr |
0.02 |
Implement VibeVoice |
|
Brief, repository-specific wording provi |
2025-08-29 |
| pr |
0.02 |
Fix video-llama modular conversion |
|
Informal phrasing, typo, and concrete CI |
2026-08-26 |
| pr |
0.02 |
Fail fast when the continuous batching background thread has |
|
Detailed implementation and failure-mode |
2026-08-25 |
| pr |
0.02 |
CI: gate the hunyuan-moe slow test |
|
Typos, terse wording, and specific CI co |
2026-08-26 |
| pr |
0.02 |
Docstring check didn't match some file - fix it |
|
Casual wording and uncertainty are stron |
2026-08-19 |
| pr |
0.02 |
Fix additional_special_tokens data loss with extra_special_t |
|
Specific technical diagnosis and validat |
2026-08-09 |
| pr |
0.02 |
[Umbrella] Continuous batching fixes for training-while-gene |
|
Highly domain-specific measurements, bug |
2026-08-25 |
| pr |
0.02 |
Wait for the first request in the async continuous batching |
|
Specific bug behavior, reproduction, and |
2026-08-25 |
| pr |
0.02 |
[Umbrella] Enable FSDP2 + expert parallelism via a 2-D (fsdp |
|
Highly specific benchmark and design det |
2026-08-22 |
| commit |
0.01 |
[docs] Fix links and remove TokenizerFast (#47748) |
|
Brief informal commit sequence is charac |
2026-08-26 |
| commit |
0.01 |
Fix incorrect token classification prefix for ESMC (#48348) |
|
Extremely terse, issue-specific commit m |
2026-08-26 |
| commit |
0.01 |
[`CI`] Unblock fast CI for now (failing tests) (#48344) |
|
Minimal informal message lacks AI-genera |
2026-08-26 |
| commit |
0.01 |
Add Qwen4Exp model (#48337) |
|
Terse, domain-specific commit notes and |
2026-08-26 |
| commit |
0.01 |
Fix video-llama modular conversion (#48336) |
|
Extremely terse, repository-specific fix |
2026-08-26 |
| commit |
0.01 |
CI: gate the hunyuan-moe slow test (#48330) |
|
Brief, informal commit wording is consis |
2026-08-26 |
| commit |
0.01 |
Fix `scores` type in stopping criteria docstrings (#47676) |
|
A concise documentation fix title provid |
2026-08-26 |
| commit |
0.01 |
Docstring check didn't match some file - fix it (#48121) |
|
Short, fragmented implementation notes a |
2026-08-26 |
| commit |
0.01 |
fix bug for blt model parallel bug (#48327) |
|
Minimal bug-fix wording and a developer |
2026-08-26 |
| commit |
0.01 |
Cpmant fix use cache (#48013) |
|
Highly specific debugging details and do |
2026-08-26 |
| commit |
0.01 |
Fix AutoTokenizer returning TokenizersBackend for DeepSeek-R |
|
Specific regression history, identifiers |
2026-08-26 |
| commit |
0.01 |
[docs] Fix failing doctests (#47687) |
|
Terse one-word commit message is charact |
2026-08-25 |
| commit |
0.01 |
Fix build_2d_sinusoidal_position_embedding on MPS (#47897) |
|
Concise, domain-specific fix title with |
2026-08-25 |
| commit |
0.01 |
Refactor audio buffer creation to use NumPy (#48295) |
|
Short technical refactoring title appear |
2026-08-25 |
| commit |
0.01 |
[docs] Kernel supported models (#48258) |
|
Minimal docs-related commit notes are te |
2026-08-25 |
| commit |
0.01 |
Disable daily nightly CI (#48292) |
|
Brief informal CI note is characteristic |
2026-08-25 |
| commit |
0.01 |
Pipeline parallel naive inference (#47289) |
|
Long technical history contains terse fr |
2026-08-25 |
| commit |
0.01 |
gs (#48288) |
|
Informal incremental notes and domain-sp |
2026-08-25 |
| commit |
0.01 |
CI: fix muse OOMs (#48284) |
|
Concise CI fix message with a reviewer m |
2026-08-25 |
| commit |
0.01 |
replace xpu-smi subprocess call in benchmark_v2 (#48083) |
|
Specific implementation choices and sign |
2026-08-25 |
| commit |
0.01 |
Compute MoE load-balancing loss per layer to avoid giant one |
|
Terse, issue-specific wording and a conc |
2026-08-24 |
| commit |
0.01 |
deterministic layer_types buffer registration in multiple mo |
|
Short technical commit-style bullets wit |
2026-08-24 |
| commit |
0.01 |
fix(data_collator): align TokenClassification numpy_call wit |
|
Conventional commit wording and a specif |
2026-08-24 |
| commit |
0.01 |
[docs] Cache crop (#47950) |
|
Minimal, informal entries are typical of |
2026-08-24 |
| commit |
0.01 |
Fix a typo in a use of a local variable field_ in a test (#4 |
|
Contains a natural grammatical typo and |
2026-08-24 |
| commit |
0.01 |
doc: Fix typo in SigLIP2 Flash Attention code example (#481 |
|
Brief typo-fix wording with repository-s |
2026-08-24 |
| commit |
0.01 |
[Fix] Export crashes on kernel-decorated function (#47808) |
|
Short commit bullets and review notes ar |
2026-08-24 |
| commit |
0.01 |
Assign a reviewer even when a codeowner has left, and route |
|
Highly domain-specific details, metrics, |
2026-08-21 |
| commit |
0.01 |
Fix ASR pipeline mono conversion for channels-last audio (fi |
|
Detailed domain-specific debugging and r |
2026-08-19 |
| commit |
0.01 |
Fix CpmAnt loading: size lm_head to vocab_size (#48012) |
|
Highly specific checkpoint and tensor-sh |
2026-08-19 |
| commit |
0.01 |
[Video] Warn and return all frames when num_frames exceeds t |
|
Repetitive review notes and CI adjustmen |
2026-08-19 |
| commit |
0.01 |
Accept artifact dir as argument in post_mlinter_review.py (# |
|
Concise repository-specific follow-up wi |
2026-08-19 |
| commit |
0.01 |
[Video] Fix convert_to_rgb channel slicing and alpha blendin |
|
Specific review feedback and enum/testin |
2026-08-19 |
| pr |
0.01 |
Fix Zamba2 construction for num_mem_blocks > 1 checkpoints |
|
Detailed checkpoint-specific reasoning a |
2026-08-26 |
| pr |
0.01 |
Modernize FSMT: shared attention interface and remove legacy |
|
Terse, domain-specific summary with conc |
2026-08-27 |
| pr |
0.01 |
Fall back to the base attention implementation when a paged |
|
Concrete implementation paths, numerical |
2026-08-25 |
| pr |
0.01 |
Fix missing FP8 TP layer overrides |
|
Focused bug description and exact test c |
2026-08-26 |
| pr |
0.01 |
Use the shared attention interface in encoder-decoder attent |
|
Concise model-specific technical details |
2026-08-26 |
| pr |
0.01 |
Branch aware export |
|
Extremely brief follow-up description wi |
2026-08-26 |
| pr |
0.01 |
Add NeoMME and NeoMME-Retriever |
|
Domain-specific model details, informal |
2026-08-15 |
| pr |
0.01 |
🚨🚧 FeatureExtractor → AudioProcessor |
|
Very brief, fragmentary description is m |
2026-03-02 |
| pr |
0.01 |
Add WeatherNext 2 weather forecasting model |
|
Highly specific implementation details a |
2026-08-10 |
| pr |
0.01 |
[CI] Add link checker |
|
Informal wording, typo, and emoticon are |
2026-08-20 |
| pr |
0.01 |
Keep MXFP4 weights quantized on XPU when use_kernels is set |
|
Concrete hardware, dependency, and perfo |
2026-08-12 |
| pr |
0.01 |
Fix incorrect token classification prefix for ESMC |
|
Brief, informal, model-specific explanat |
2026-08-26 |
| pr |
0.01 |
Add support for NeuCodec |
|
The brief, informal description contains |
2026-07-07 |
| pr |
0.01 |
Create the continuous batching CPU group with local synchron |
|
Concrete distributed-systems diagnosis a |
2026-08-25 |
| pr |
0.01 |
[CB] Fail faster |
|
Typos, terse phrasing, and concrete impl |
2026-08-26 |
| pr |
0.01 |
Resolve continuous batching config against the text config f |
|
Specific implementation details and vali |
2026-08-25 |
| pr |
0.01 |
Fix kernel commit and repo paths for ESMFold2 |
|
Concise, specific release-preparation wo |
2026-08-21 |
| pr |
0.01 |
[`CI`] Unblock fast CI for now (failing tests) |
|
Extremely terse title and description pr |
2026-08-26 |
| pr |
0.01 |
Add Granite Speech 5.0 |
|
The minimal, fragmentary text provides n |
2026-08-25 |
| pr |
0.01 |
Made SAM3 to return loss in outputs |
|
Informal grammar, specific implementatio |
2026-06-13 |
| pr |
0.01 |
Add a regression test for force_accelerate_hooks signature |
|
Very brief, informal description with is |
2026-08-24 |
| pr |
0.01 |
Experiments with Neuron |
|
Terse informal note with a clear experim |
2026-07-13 |
| pr |
0.01 |
Fix `scores` type in stopping criteria docstrings |
|
Short, issue-specific documentation note |
2026-07-31 |
| pr |
0.01 |
skip mtp slow tests for now (#48328) |
|
Brief description contains a natural spe |
2026-08-26 |
| pr |
0.01 |
fix bug for blt model parallel bug |
|
Informal wording, typo, and direct revie |
2026-08-26 |
| pr |
0.01 |
Byte-level tokenizer.json + Hub tokenizer_class LlamaTokeniz |
|
Terse, lowercase phrasing and specific t |
2026-08-25 |
| pr |
0.01 |
[docs] Fix failing doctests |
|
Brief, issue-specific wording with no AI |
2026-07-31 |
| pr |
0.01 |
Warn once when a hub-kernel function falls back to its refer |
|
Highly specific benchmarks, package mapp |
2026-08-21 |
| pr |
0.01 |
Handle empty FSDP shards in sharded loading: crash and hang |
|
Deeply specific sharding reproducers and |
2026-08-23 |
| commit |
0.00 |
[Fix] Avoid duplicating tests in CI (#48287) |
|
Terse commit-style notes show no meaning |
2026-08-25 |
| commit |
0.00 |
[`GDN`] Fix recurrent FLA fallback (#48266) |
|
The one-word follow-up is plainly terse |
2026-08-25 |
| commit |
0.00 |
ignore mlinter ci file (#48267) |
|
Extremely terse commit message with no A |
2026-08-25 |
| commit |
0.00 |
[serge] Fix 2 integration tests regressed by commit 16780c86 |
|
— |
2026-08-24 |
| commit |
0.00 |
[VITS] Un-skip test_model_forward (#46375) |
|
Terse commit message with a human co-aut |
2026-08-21 |
| commit |
0.00 |
Apply context parallelism to the evaluation path (#48167) |
|
Brief, technical commit message with no |
2026-08-21 |
| commit |
0.00 |
🚨 TP dtensor API inference + training (#47579) |
|
Long series of terse commits, typos, and |
2026-08-21 |
| commit |
0.00 |
Add MLU support to is_flash_linear_attention_available (#469 |
|
Concise domain-specific commit title wit |
2026-08-21 |
| commit |
0.00 |
Fix ROCm SDPA-flash skip guard that crashes on RDNA GPUs (#4 |
|
Terse technical commit message with an i |
2026-08-21 |
| commit |
0.00 |
Fix DynamicCache reconstruction during ExecuTorch export (#4 |
|
Iterative technical commits and human co |
2026-08-21 |
| commit |
0.00 |
Revert "Support per-layer cache configuration and attention- |
|
Standard revert message with direct comm |
2026-08-21 |
| commit |
0.00 |
Support per-layer cache configuration and attention-mask sel |
|
Brief technical commit series with a hum |
2026-08-21 |
| commit |
0.00 |
Fix DeepSeek V2 default vocab size (#48159) |
|
Terse conventional fix message with a hu |
2026-08-20 |
| commit |
0.00 |
Revert "[Quantization]: Refactor is_quantization_compressed |
|
Standard terse revert message. |
2026-08-20 |
| commit |
0.00 |
fix bugs for clvp model (#47127) |
|
Terse commit series with repeated human |
2026-08-20 |
| pr |
0.00 |
draft modular v2 + transformers format |
|
Empty template section provides no evide |
2026-08-25 |
| pr |
0.00 |
Add Qwen4Exp model |
|
Empty template section provides no evide |
2026-08-26 |
| pr |
0.00 |
[serge] Fix 12 integration tests for model `edgetam` failing |
|
— |
2026-08-25 |
| pr |
0.00 |
[serge] Fix 2 integration tests regressed by commit 6034e90c |
|
— |
2026-08-22 |
