| commit |
1.00 |
fix(fetcher): case-sensitive submodule check on case-insensi |
|
Commit message contains explicit AI assi |
2026-11-03 |
| commit |
1.00 |
fix(cosmos3_edge): bump backend_equivalence_atol for Pillow |
|
Commit message contains explicit AI assi |
2026-11-03 |
| commit |
1.00 |
feat(ci): runtime dep recorder — track transformers imports |
|
Commit message contains explicit AI assi |
2026-11-03 |
| commit |
1.00 |
perf(pipelines): lazy-load pipeline submodules in pipelines/ |
|
Commit message contains explicit AI assi |
2026-11-03 |
| commit |
1.00 |
Fix EncoderDecoderCache missing .layers in torch export test |
|
Commit message contains explicit AI assi |
2026-11-03 |
| commit |
1.00 |
Add SGLang compatibility tests to Transformers CI (#48891) |
|
Commit message contains explicit AI assi |
2026-11-03 |
| commit |
1.00 |
Docs: remove duplicate hf_token secret in build_documentatio |
|
Commit message contains explicit AI assi |
2026-11-02 |
| commit |
1.00 |
fix(tests_fetcher): handle `X as X` re-export syntax and dup |
|
Commit message contains explicit AI assi |
2026-11-02 |
| commit |
1.00 |
deal with Nan values for aggregating loss (#50172) |
|
Commit message contains explicit AI assi |
2026-11-02 |
| commit |
1.00 |
[VibeVoice] Add `sample` argument to disable VAE noise in au |
|
Commit message contains explicit AI assi |
2026-11-02 |
| commit |
1.00 |
fix(tests_fetcher): use posix paths throughout, eliminate al |
|
Commit message contains explicit AI assi |
2026-11-02 |
| commit |
1.00 |
Remove the GitHub API cache-bust from the AMD docker images |
|
Commit message contains explicit AI assi |
2026-11-02 |
| commit |
1.00 |
Allow untokenized output from `apply_transcription_request` |
|
Commit message contains explicit AI assi |
2026-10-30 |
| commit |
1.00 |
Fix `WatermarkDetector` p-value computation (#50123) |
|
Commit message contains explicit AI assi |
2026-10-30 |
| commit |
1.00 |
test: lazy-load pipeline classes to eliminate collection-pha |
|
Commit message contains explicit AI assi |
2026-10-30 |
| commit |
1.00 |
Docs: build English only, other languages are machine-transl |
|
Commit message contains explicit AI assi |
2026-10-30 |
| commit |
1.00 |
Add CI dashboard badge when a pull request opens (#50106) |
|
Commit message contains explicit AI assi |
2026-10-29 |
| commit |
1.00 |
feat(ci): surface pre-existing failures in PR run-slow comme |
|
Commit message contains explicit AI assi |
2026-10-29 |
| commit |
1.00 |
Fix `_init_weights` gaps in 7 audio models and enable `test_ |
|
Commit message contains explicit AI assi |
2026-10-29 |
| pr |
1.00 |
Fold GQA query heads onto their KV head when decoding instea |
|
PR body explicitly mentions AI collabora |
2026-10-19 |
| pr |
1.00 |
[vLLM] Add Omni token counting and text replacement offsets |
|
The description explicitly states that A |
2026-10-08 |
| pr |
1.00 |
Fix error message and remove duplicate code in convert_slow_ |
|
PR body explicitly mentions AI collabora |
2026-10-28 |
| pr |
1.00 |
Flag special tokens typed in the input text in special_token |
|
PR body explicitly mentions AI collabora |
2026-10-28 |
| pr |
1.00 |
Docs: note in the v5 migration guide that hidden_states[-1] |
|
PR body explicitly mentions AI collabora |
2026-11-03 |
| pr |
1.00 |
Check for an empty batch in convert_to_tensors without flatt |
|
The description explicitly discloses Cla |
2026-10-23 |
| pr |
1.00 |
[Masking] Avoid device-to-host sync in `_ignore_causal_mask_ |
|
PR body explicitly mentions AI collabora |
2026-11-02 |
| pr |
1.00 |
[tests] Pin the chat-template date in Llama's FA static-cach |
|
Explicitly states it was generated with |
2026-11-03 |
| pr |
1.00 |
Add Maximal Update Parametrization (μP) |
|
Explicitly declares assistance by Claude |
2026-05-08 |
| pr |
1.00 |
Exporters: ONNX Runtime fusions at export time (`OnnxConfig( |
|
Explicitly states it was generated with |
2026-11-03 |
| pr |
1.00 |
feat(ci): runtime dep recorder — track transformers imports |
|
Explicitly states it was generated with |
2026-11-03 |
| pr |
1.00 |
Fix Top-H normalized prefix entropy and full-distribution bu |
|
PR body explicitly mentions AI collabora |
2026-11-03 |
| pr |
1.00 |
OpenVINO exporter: let OpenVINO fuse attention, RoPE, the KV |
|
PR body explicitly mentions AI collabora |
2026-11-03 |
| pr |
1.00 |
Add dots3-note Preview model support |
|
PR body explicitly mentions AI collabora |
2026-08-08 |
| pr |
1.00 |
[Qwen4 Exp] Vectorize the QSA indexer for prefix masks |
|
PR body explicitly mentions AI collabora |
2026-10-30 |
| pr |
1.00 |
Fix CLAP loss double-counting the text-to-audio direction |
|
PR body explicitly mentions AI collabora |
2026-11-03 |
| pr |
1.00 |
[Fix] Prevent duplicate weight renaming on save_pretrained f |
|
PR body explicitly mentions AI collabora |
2026-11-02 |
| pr |
1.00 |
Reject assisted generation with stateful logits processors |
|
PR body explicitly mentions AI collabora |
2026-11-03 |
| pr |
1.00 |
Optim: add SST optimizer wrapper and distributed Trainer sup |
|
PR body explicitly mentions AI collabora |
2026-10-23 |
| pr |
1.00 |
Fix invalid JSON in serve streaming error events [ Issue #49 |
|
PR body explicitly mentions AI collabora |
2026-11-02 |
| pr |
1.00 |
🚨 Exporters: export multimodal models per component (`export |
|
Explicitly states it was generated with |
2026-11-02 |
| pr |
1.00 |
Exporters: load optimum-intel OpenVINO exports (legacy forma |
|
Explicitly states it was generated with |
2026-11-02 |
| pr |
1.00 |
🚨 Exporters: decompose generative models into one decode gra |
|
Explicitly states it was generated with |
2026-11-02 |
| pr |
1.00 |
Exporters: run generation exports through `generate` (`Expor |
|
Explicitly states it was generated with |
2026-11-02 |
| pr |
1.00 |
Exporters: run multimodal exports (`ExportedMultimodalModel` |
|
Explicitly states it was generated with |
2026-11-02 |
| pr |
1.00 |
Support backend-defined MoE parameter targeting in torchao |
|
PR body explicitly mentions AI collabora |
2026-10-30 |
| pr |
1.00 |
🚨 Exporters: `ExportArtifacts`, Hub save/load, one backend r |
|
Explicitly states it was generated with |
2026-10-27 |
| pr |
1.00 |
perf(pipelines): lazy-load pipeline submodules in pipelines/ |
|
Explicitly states it was generated with |
2026-10-30 |
| pr |
1.00 |
Fix Doge load balancing loss pairing expert loads with rank- |
|
PR body explicitly mentions AI collabora |
2026-11-02 |
| pr |
1.00 |
Add grt model |
|
PR body explicitly mentions AI collabora |
2026-11-03 |
| pr |
1.00 |
sam3: enhance output boxes for any resolution |
|
PR body explicitly mentions AI collabora |
2026-10-24 |
| pr |
1.00 |
Keep image inputs on the first chunk of a chunked prefill |
|
PR body explicitly mentions AI collabora |
2026-10-20 |
| pr |
0.25 |
[utils] Add MUSA support for flash linear attention |
|
Polished and detailed, but uses specific |
2026-10-28 |
| pr |
0.24 |
Add generic heterogeneous model support (per-layer config an |
|
Polished explanatory prose is somewhat g |
2026-04-09 |
| pr |
0.22 |
fix(pipelines): validate zero-shot detection batch labels |
|
Polished structured summary, but it cont |
2026-11-03 |
| pr |
0.22 |
Fix QuantizedCache.crop() IndexError breaking speculative de |
|
Highly polished technical explanation, b |
2026-11-03 |
| pr |
0.20 |
Fix RoPE inv_freq bf16 regression under DeepSpeed ZeRO-3 |
|
Formal and comprehensive, but contains s |
2026-11-03 |
| pr |
0.14 |
fix(pipelines): preserve batched generation metadata |
|
Well-structured technical summary, but n |
2026-11-01 |
| pr |
0.12 |
Restore BLT's 512-byte sliding windows (entropy patcher + lo |
|
Highly polished and structured, but grou |
2026-09-29 |
| pr |
0.12 |
Fix model_input_names singleton issue causing shared state |
|
Generic explanation is polished, but the |
2025-11-06 |
| pr |
0.12 |
Fix repetition penalty for masked padding tokens |
|
Polished structure is present, but the t |
2026-11-01 |
| pr |
0.12 |
Fix out-of-bounds error in continuous batching top-k process |
|
Specific debugging details and terse tes |
2026-11-03 |
| pr |
0.08 |
[PR3] [Modular Clean-Up] RoPE canonical parents |
|
Dense technical detail is structured but |
2026-11-03 |
| pr |
0.08 |
fix: preserve multiple candidates in multimodal pipelines |
|
Specific regression details and validati |
2026-10-31 |
| pr |
0.08 |
Add LLaVA-OneVision-1.5 model and related configurations |
|
Brief generic description, but no clear |
2025-09-23 |
| pr |
0.08 |
[Qwen2-VL] Precompute data-dependent vision tensors on CPU f |
|
Polished and structured, though its deta |
2026-10-23 |
| commit |
0.06 |
Fix `bboxes` being overwritten before use in `prepare_examp |
|
Clear domain-specific bug explanation, w |
2026-11-02 |
| pr |
0.06 |
Replace torch.FloatTensor with torch.Tensor in the model fil |
|
Detailed mechanical migration notes and |
2026-10-30 |
| pr |
0.06 |
Replace torch.FloatTensor with torch.Tensor outside the mode |
|
Mechanical change description uses speci |
2026-10-30 |
| pr |
0.05 |
[PR2] [Modular Clean-Up] *Norm Canonical Update and Renaming |
|
Technical implementation details and dup |
2026-11-03 |
| pr |
0.05 |
Exporters: run exported models (`ExportedModel`) |
|
Highly domain-specific details and human |
2026-10-27 |
| pr |
0.05 |
Simplify the quantizer API changes |
|
Very brief and generic, but lacks distin |
2026-10-23 |
| commit |
0.04 |
Read add_bos_token and add_eos_token from GGUF files (#49649 |
|
Detailed but technically specific change |
2026-11-02 |
| pr |
0.04 |
Modernize ConvNextV2, ResNet, RegNet, RTDetrResnet, SwinV2, |
|
Informal phrasing and specific project d |
2026-08-27 |
| pr |
0.04 |
fix(pipelines): respect audio classification problem type |
|
Concrete regression cases and checkpoint |
2026-11-03 |
| pr |
0.04 |
Add overridable Trainer.get_resumed_dataloader |
|
Concise, technically specific proposal w |
2026-11-02 |
| commit |
0.03 |
Docs: note in the v5 migration guide that hidden_states[-1] |
|
Specific documentation edits and review |
2026-11-03 |
| commit |
0.03 |
Rename `gradient_checkpointing` to `activation_checkpointing |
|
Detailed but natural API migration notes |
2026-11-03 |
| commit |
0.03 |
[serve] Respect max_completion_tokens in /v1/chat/completion |
|
Concise issue-specific explanation using |
2026-11-02 |
| commit |
0.03 |
[CI] Docs bot (#49999) |
|
Terse, task-focused commit-style notes w |
2026-10-31 |
| pr |
0.03 |
fix(tokenizers): stop treating a missing transformers_versio |
|
Detailed, issue-specific debugging and t |
2026-10-19 |
| pr |
0.03 |
fix(fetcher): case-sensitive submodule check on case-insensi |
|
Highly specific Windows behavior and imp |
2026-11-03 |
| pr |
0.03 |
Object Detection Losses refactor - from dedicated folder to |
|
Domain-specific refactoring details and |
2026-10-15 |
| pr |
0.03 |
fix(cosmos3_edge): bump backend_equivalence_atol for Pillow |
|
Specific benchmark results and Pillow be |
2026-11-03 |
| pr |
0.03 |
Rename GradientCheckpointingLayer and supports_gradient_chec |
|
Concrete compatibility details and revie |
2026-10-02 |
| pr |
0.03 |
[PR1] [Modular Clean-Up] Remove No-op Overrides |
|
Structured planning is technical and inc |
2026-11-03 |
| pr |
0.03 |
Fix Moe aux loss scaling |
|
Contains specific implementation reasoni |
2026-11-03 |
| pr |
0.03 |
Keep overrides of the deprecated gradient checkpointing meth |
|
Detailed project-specific compatibility |
2026-11-03 |
| pr |
0.03 |
Test the dispatched FA3 kernel per device |
|
Highly specific runtime and device detai |
2026-11-03 |
| pr |
0.03 |
Fix Qwen4-Exp eager attention NaNs with left padding |
|
Detailed causal debugging and repository |
2026-10-23 |
| pr |
0.03 |
🚨 Keep static sliding-window decode on one graph |
|
Detailed implementation tradeoffs, bench |
2026-10-12 |
| pr |
0.03 |
Add MiMo V2 multimodal support, including MiMo-V2.5, MiMo-V2 |
|
Detailed model-specific implementation n |
2026-10-26 |
| pr |
0.03 |
[`FA`] Fix head dim manipulation for more special models |
|
Brief, awkward wording and informal styl |
2026-10-22 |
| commit |
0.02 |
[Video] Require one metadata entry per video and refuse retu |
|
Terse, domain-specific commit series wit |
2026-11-04 |
| commit |
0.02 |
Fix model_input_names singleton issue causing shared state ( |
|
Detailed, code-specific explanation and |
2026-11-03 |
| commit |
0.02 |
Keep overrides of the deprecated gradient checkpointing meth |
|
Specific regression context and internal |
2026-11-03 |
| commit |
0.02 |
Fix Qwen4-Exp eager attention NaNs with left padding (#49923 |
|
Specific model, padding, and regression |
2026-11-03 |
| commit |
0.02 |
fix(pipelines): preserve batched generation metadata (#50210 |
|
Terse technical commit history with norm |
2026-11-03 |
| commit |
0.02 |
🚨 Safe-ize pi0 graphs (#50160) |
|
Technical commit bullets are terse and i |
2026-11-03 |
| commit |
0.02 |
Chat template output tensor type (#50103) |
|
Terse, informal commit fragments strongl |
2026-11-03 |
| commit |
0.02 |
[Trainer] Compile with static shapes when the backend is neu |
|
Short, specific commit title with no AI- |
2026-11-02 |
| commit |
0.02 |
Dequantize MXFP4 experts to the requested dtype (#50256) |
|
Brief, technically precise commit messag |
2026-11-02 |
| commit |
0.02 |
[Gemma4] Pass num_frames_per_video to get_video_features in |
|
Brief, repetitive commit-style wording i |
2026-11-02 |
| commit |
0.02 |
Use repo-wide linguist-generated globs so new models don't t |
|
Specific CI and repository rationale rea |
2026-11-02 |
| commit |
0.02 |
[PaliGemma] Apply final logit softcapping (#50156) |
|
Concise, domain-specific description wit |
2026-11-02 |
| commit |
0.02 |
Speed up `import transformers` by building the package distr |
|
Specific importlib and torch details sug |
2026-10-30 |
| commit |
0.02 |
Add `TRANSFORMERS_SDPA_FORCE_GQA` envvar in `sdpa_attention_ |
|
Concise technical change with a specific |
2026-10-30 |
| commit |
0.02 |
Fix AttributeError on LinearAttentionLayer in export generat |
|
Specific code details, model names, and |
2026-10-30 |
| commit |
0.02 |
[LwDetr] Initialize encoder heads like the detection heads ( |
|
Detailed implementation references and i |
2026-10-30 |
| commit |
0.02 |
[OneFormer] Fix loss computation when is_training=False (#50 |
|
Specific model behavior and terse revisi |
2026-10-30 |
| commit |
0.02 |
Shard the parameters kept in float32 in their own FSDP2 grou |
|
Terse commit history with reversions and |
2026-10-30 |
| commit |
0.02 |
🚨 build and expand once the tp/ep_plan/fsdp_plan (#49527) |
|
Iterative commit messages and precise co |
2026-10-30 |
| commit |
0.02 |
handle local sink in TP (#49567) |
|
Brief, informal commits with concrete im |
2026-10-30 |
| commit |
0.02 |
[Qwen4 Exp] Fix _no_split_modules naming a class that does n |
|
Specific technical fix with concise, rep |
2026-10-30 |
| commit |
0.02 |
Don't pass code_revision to non-Auto model classes in pipeli |
|
Specific pipeline behavior described pla |
2026-10-29 |
| commit |
0.02 |
FP8 backward: chunked expert dequant, skip autograd.Function |
|
Terse, domain-specific commit bullets an |
2026-10-29 |
| commit |
0.02 |
Train adapters on top of FP8 weights (#50031) |
|
Technical, incremental commit notes are |
2026-10-29 |
| commit |
0.02 |
[loading] Allow `PermuteForRope` to act on single tensors wh |
|
Terse, typo-filled commit bullets strong |
2026-10-28 |
| pr |
0.02 |
[Video] Require one metadata entry per video and refuse retu |
|
Terse, domain-specific wording and infor |
2026-10-28 |
| pr |
0.02 |
docs(watermarking): fix WatermarkDetector.__call__ docstring |
|
Brief documentation fix with concrete ed |
2026-11-03 |
| pr |
0.02 |
Drop `content=None` from messages in `apply_chat_template` |
|
Detailed domain-specific reasoning and c |
2026-04-14 |
| pr |
0.02 |
[Model] LFM3: Conv kernel prerequisites |
|
Terse technical title and informal revie |
2026-10-27 |
| pr |
0.02 |
[`Rope`] Fix per layer rope in head dim manipulation |
|
Informal shorthand and domain-specific r |
2026-10-22 |
| pr |
0.02 |
:rotating_light: Kernels=True ?! |
|
Informal title and highly specific kerne |
2026-08-04 |
| pr |
0.02 |
Improve Mistral Large 4 cache sharing |
|
Very brief informal explanation tied to |
2026-11-03 |
| pr |
0.02 |
Add LoMa keypoint matching model |
|
Specific model integration scope, checkp |
2026-07-18 |
| pr |
0.02 |
remove ep_router |
|
Informal domain-specific explanation and |
2026-10-14 |
| pr |
0.02 |
fix failed cases for beam model |
|
Terse wording, typos, and informal revie |
2026-11-02 |
| pr |
0.02 |
fix failed test cases for nemotron3_5_transcirbe model |
|
Informal phrasing and typo in the review |
2026-11-02 |
| commit |
0.01 |
:rotating_light: Finally drop rope deltas from ModelOutput! |
|
Very terse informal commit messages stro |
2026-11-04 |
| commit |
0.01 |
[tests] Pin the chat-template date in Llama's FA static-cach |
|
Short, specific test-fix message with no |
2026-11-03 |
| commit |
0.01 |
Fix a cache test (#50320) |
|
Extremely terse iterative fixes are char |
2026-11-03 |
| commit |
0.01 |
Improve Mistral Large 4 cache sharing (#50305) |
|
Informal, domain-specific incremental me |
2026-11-03 |
| commit |
0.01 |
Test the dispatched FA3 kernel per device (#50304) |
|
Detailed ROCm/CUDA terminology and revie |
2026-11-03 |
| commit |
0.01 |
upgrade version for mamba-ssm and fla kernel (#50294) |
|
Brief kernel version updates with a huma |
2026-11-03 |
| commit |
0.01 |
Fix Doge load balancing loss pairing expert loads with rank- |
|
Concise, domain-specific fix description |
2026-11-03 |
| commit |
0.01 |
Prevent deepspeed from downcasting rope Buffers (#50275) |
|
Terse iterative commit notes typical of |
2026-11-03 |
| commit |
0.01 |
[`FA`] Less flaky FA tests (#49861) |
|
Very brief, informal technical message. |
2026-11-03 |
| commit |
0.01 |
[`FA`] Fix old models that do not use a proper interface (#4 |
|
Short informal notes with a specific com |
2026-11-03 |
| commit |
0.01 |
[`Moe`] Fix grouped mm fallback (#49857) |
|
Brief domain-specific fix and style note |
2026-11-03 |
| commit |
0.01 |
[Distributed] Set `device_id` only for NCCL (#50281) |
|
Concise technical message with a normal |
2026-11-03 |
| commit |
0.01 |
[Improvement] Formalize KV sharing (#47290) |
|
Typos, informal fragments, and iterative |
2026-11-02 |
| commit |
0.01 |
Fix OOM regression for FSDP2 + cpu_ram_efficient_loading on |
|
Terse iterative fixes and human co-autho |
2026-11-02 |
| commit |
0.01 |
Fix resize_embeddings tests for deepspeed (#49909) |
|
Short implementation notes and repeated |
2026-11-02 |
| commit |
0.01 |
fix failed test cases for pocket_tts model (#50159) |
|
Terse technical changelog with human sig |
2026-11-02 |
| commit |
0.01 |
fix Load DETR-family panoptic masks with load_image_as_tenso |
|
Concise technical fix description with s |
2026-11-02 |
| commit |
0.01 |
update to torch 2.15 for xpu (#50236) |
|
Terse version-update notes and human sig |
2026-11-02 |
| commit |
0.01 |
Fix/trainer parallel token count 50212 (#50214) |
|
Detailed commit breakdown uses specific |
2026-11-02 |
| commit |
0.01 |
Modular suffix overlap - allow non-model classes (#49754) |
|
Informal iterative notes and typos stron |
2026-11-02 |
| commit |
0.01 |
🚨 [serve][breaking] update attn for mac users (#49730) |
|
Terse, informal wording is characteristi |
2026-10-30 |
| commit |
0.01 |
Restore the `MegaBlocksMoeMLP` hub kernel mapping for MXFP4 |
|
Detailed domain-specific explanation and |
2026-10-30 |
| commit |
0.01 |
Fix ZeRO-3 init for composite models built via _from_config |
|
Brief technical message with a sign-off |
2026-10-30 |
| commit |
0.01 |
[docs] Quantization (#50093) |
|
Short informal changelog bullets show no |
2026-10-30 |
| commit |
0.01 |
expose private parameter in `trainer.push_to_hub` (#49874) |
|
Incremental commit-style bullets and con |
2026-10-30 |
| commit |
0.01 |
Trainer: `save_pretrained` on every rank for PEFT model (#49 |
|
Long sequence of specific implementation |
2026-10-30 |
| commit |
0.01 |
[docs] Vision backend differences (#50131) |
|
Minimal informal wording provides no ind |
2026-10-30 |
| commit |
0.01 |
fix(glm4v): bump backend_equivalence_atol to 0.5 for Pillow |
|
Specific Pillow behavior and cross-refer |
2026-10-30 |
| commit |
0.01 |
need to check model_type and model_name (#50128) |
|
Awkward, terse grammar is more consisten |
2026-10-30 |
| commit |
0.01 |
Remove torch.jit.script (#50082) |
|
Terse, repository-specific commit messag |
2026-10-30 |
| commit |
0.01 |
Mark generated files as linguist-generated (#50061) |
|
Incremental technical commit notes and f |
2026-10-30 |
| commit |
0.01 |
Standardise processing requires (#50042) |
|
Brief informal sequence of cleanup commi |
2026-10-30 |
| commit |
0.01 |
[Kosmos2] Fix SDPA applying a causal mask in the image-to-te |
|
Technical shorthand and iterative fix no |
2026-10-30 |
| commit |
0.01 |
v5.23.0-dev0 (#50155) |
|
Simple version-bump message with no gene |
2026-10-30 |
| commit |
0.01 |
Config annotations (#50067) |
|
Typos, informal phrasing, and repository |
2026-10-30 |
| commit |
0.01 |
Reuse the TP/EP test ranks across tests instead of spawning |
|
Informal commit notes, typo, and detaile |
2026-10-30 |
| commit |
0.01 |
Fix check repo (#49899) |
|
Terse notes, typos, and uncertainty stro |
2026-10-30 |
| commit |
0.01 |
Update MiniCPM OCR docs and tests (#50134) |
|
Specific test and formatting changes wit |
2026-10-29 |
| commit |
0.01 |
[Masking] Skip linear layers for the default full-attention |
|
Concise technical iterations and reposit |
2026-10-29 |
| commit |
0.01 |
[docs] correct the MiniCPM-OCR link (#50052) |
|
Brief, specific documentation correction |
2026-10-29 |
| commit |
0.01 |
Remove expired deprecation cycles (#50037) |
|
Terse fix notes with informal wording an |
2026-10-29 |
| commit |
0.01 |
[Fix] Remove WIP doc for Ml4 (#50124) |
|
Short, repository-specific description o |
2026-10-29 |
| commit |
0.01 |
Add Mistral Large 4 (#50121) |
|
Long, messy technical history with typos |
2026-10-29 |
| commit |
0.01 |
Fix NameError in accelerate.py and clarify error for PyTorch |
|
Specific bug and compatibility issue sta |
2026-10-29 |
| commit |
0.01 |
Processor unnecessary warnings (#49812) |
|
Brief informal notes with typos strongly |
2026-10-29 |
| commit |
0.01 |
Fix boolean padding in ClapFeatureExtractor (#23648) (#49960 |
|
A concise, specific fix title with no AI |
2026-10-29 |
| commit |
0.01 |
Fix image backend equivalence tests for Pillow 13 (#50105) |
|
Short, specific fix title with no genera |
2026-10-29 |
| commit |
0.01 |
[Fix] Count cropping of an uninitialized layer as a no-op (# |
|
Informal wording and typo indicate a hum |
2026-10-29 |
| pr |
0.01 |
:rotating_light: Finally drop rope deltas from ModelOutput! |
|
Casual phrasing, emoji, and informal ton |
2026-10-30 |
| pr |
0.01 |
Import torchcodec lazily in processing_utils |
|
Informal wording and a typo are characte |
2026-11-03 |
| pr |
0.01 |
pin pyarrow<26 to fix datasets import with numpy<2 in ci |
|
Highly specific dependency details, erro |
2026-11-04 |
| pr |
0.01 |
[Distributed] Set `device_id` only for NCCL |
|
Concise, issue-focused explanation with |
2026-11-02 |
| pr |
0.01 |
Fix CJK decoding for LlamaTokenizerFast by loading full pipe |
|
Specific bug details and explicit author |
2026-10-02 |
| pr |
0.01 |
Add `transcription` input arg to `apply_transcription_reques |
|
Terse, specific wording and review conte |
2026-10-22 |
| pr |
0.01 |
test: dummy change to vit/llama/mistral to verify runtime de |
|
Specific CI purpose and informal wording |
2026-11-03 |
| pr |
0.01 |
Quantize on load from the checkpoint tensor, not its cast to |
|
Contains highly specific implementation |
2026-11-03 |
| pr |
0.01 |
[CB] Scale the per-token memory estimate by the TP size |
|
Concise, domain-specific explanation wit |
2026-09-06 |
| pr |
0.01 |
Draft updates for contribution/agent rules |
|
Informal opinions, emoji, and repository |
2026-10-20 |
| pr |
0.01 |
[Model] LFM3 Family: LFM3-Text |
|
Brief, highly specific model details and |
2026-11-03 |
| pr |
0.01 |
Fixes two issues related to TP with tied weights and DTensor |
|
Dense, domain-specific debugging details |
2026-11-02 |
| pr |
0.01 |
Fix a cache test |
|
Brief, informal description with a speci |
2026-11-03 |
| pr |
0.01 |
Add minimax vl 01 |
|
Minimal WIP text and informal human revi |
2026-08-04 |
| pr |
0.01 |
Use 32-row flex attention blocks for head_dim > 256 |
|
Highly specific benchmarks, hardware det |
2026-10-09 |
| pr |
0.01 |
[compressed-tensors] dequantize while loading through the co |
|
Dense domain-specific details and implem |
2026-11-02 |
| pr |
0.01 |
Save dequantized models without the quantizer's weight conve |
|
Concise, specific regression details and |
2026-11-02 |
| pr |
0.01 |
Trainer: make checkpoints work under tensor parallelism |
|
Specific distributed-training behavior, |
2026-09-06 |
| pr |
0.01 |
ZeRO-3: read weights on rank 0 only and convert them one gro |
|
Specific memory measurements, ZeRO-3 beh |
2026-11-03 |
| pr |
0.01 |
Prevent deepspeed from downcasting rope Buffers |
|
Terse, informal wording and a specific t |
2026-11-02 |
| commit |
0.00 |
[serge] Fix 2 integration tests regressed by commit 7d3ae42e |
|
— |
2026-11-02 |
| commit |
0.00 |
Audio from video in jinja (#49382) |
|
Numerous typos and fragmented iterative |
2026-11-02 |
| pr |
0.00 |
[`kernels`] Bump version |
|
Terse, informal wording and a concrete t |
2026-11-03 |
| pr |
0.00 |
upgrade version for mamba-ssm and fla kernel |
|
Informal abbreviations and concise versi |
2026-11-03 |