v4.57.0 Branch#41310

Merged

LysandreJik merged 167 commits intov4from

v4-backup

Oct 3, 2025

Commits on Sep 18, 2025

Update expected values for one more `test_speculative_generation` after #40949 (#40967)
ydshieh
authored
FIX(trainer): ensure final checkpoint is saved when resuming training (#40347)

authored
Add new model LFM2-VL (#40624)

zucchini-nlp
and
ankke
authored
Fix outdated version checks of accelerator (#40969)
cyyever
authored
Use `skip_predictor=True` in vjepa2 `get_vision_features` (#40966)
hamishs
authored
[Trainer] Fix DP loss (#40799)

SunMarc
and
S1ro1
authored
[timm_wrapper] better handling of "Unknown model" exception in timm (#40951)
harshaljanjani
authored
Fix Issue #39030: AutoTokenizer.from_pretrained does not propagate token (#40956)

brandenkmurray
and
Ubuntu
authored
[tests] Really use small models in all fast tests (#40945)
Cyrilvallez
authored
Add captured actual outputs to CI artifacts (#40965)
ydshieh
authored
Revert change in `compile_friendly_resize` (#40645)
qubvel
authored
Track the CI (model) jobs that don't produce test output files (process being killed etc.) (#40981)
ydshieh
authored

Commits on Oct 3, 2025

Remove `set_model_tester_for_less_flaky_tests` (#40982)

Cyrilvallez
authored and
ArthurZucker
committed
Benchmarking v2 GH workflows (#40716)

ahadnagy
authored and
ArthurZucker
committed
ENH: Enable readline support for transformers chat (#40911)

BenjaminBossan
authored and
ArthurZucker
committed
[testing] test `num_hidden_layers` being small in model tester (#40992)

ydshieh
authored and
ArthurZucker
committed
blt wip (#38579)

authored and
ArthurZucker
committed
[`RMSNorm`] Fix rms norm init for models that center around 1 (#40796)

vasqu
authored and
ArthurZucker
committed
Make `EfficientLoFTRModelTest` faster (#41000)

ydshieh
authored and
ArthurZucker
committed
Fix typoes in src and tests (#40845)

cyyever
authored and
ArthurZucker
committed
Fix more dates in model cards and wrong modalities in _toctree.yml (#40955)

yonigozlan
authored and
ArthurZucker
committed
RUFF fix on CI scripts (#40805)

cyyever
authored and
ArthurZucker
committed
fix dict like init for ModelOutput (#41002)

SunMarc
authored and
ArthurZucker
committed
[tests] update `test_left_padding_compatibility` (and minimize overwrites) (#40980)

gante
authored and
ArthurZucker
committed
Patch more `unittest.case.TestCase.assertXXX` methods (#41008)

ydshieh
authored and
ArthurZucker
committed
🚨 [lightglue] fix: matches order changed because of early stopped indices (#40859)

authored and
ArthurZucker
committed
Fix `PhimoeIntegrationTest` (#41007)

ydshieh
authored and
ArthurZucker
committed
Fix Glm4v test (#41011)

Cyrilvallez
authored and
ArthurZucker
committed
Update after #41007 (#41014)

ydshieh
authored and
ArthurZucker
committed
Fix benchmark runner argument name (#41012)

ahadnagy
authored and
ArthurZucker
committed
Adding support for Qwen3Omni (#41025)

committed
Making compute_loss_func always take priority in Trainer (#40632)

Flakes342
authored and
ArthurZucker
committed
Modify Qwen3Omni parameter name since VL changed it (#41045)

authored and
ArthurZucker
committed
Fix Qwen video tests (#41049)

zucchini-nlp
authored and
ArthurZucker
committed
[testing] Fix `qwen2_audio` (#41018)

ydshieh
authored and
ArthurZucker
committed
Fix typing of tuples (#41028)

cyyever
authored and
ArthurZucker
committed
Remove optax (#41030)

cyyever
authored and
ArthurZucker
committed
Fix typos in English/Chinese documentation (#41031)

cyyever
authored and
ArthurZucker
committed
Use torch.autocast (#40975)

cyyever
authored and
ArthurZucker
committed
docs: improved RoPE function Docstrings (#41004)

authored and
ArthurZucker
committed
Fix condition for emitting warning when generation exceeds max model length (#40775)

yannicks1
authored and
ArthurZucker
committed
Fix outdated torch version check (#40925)

cyyever
authored and
ArthurZucker
committed
Add Whole Word Masking and Padding Strategy to DataCollatorForLanguageModeling (#39485)

rjgleaton
authored and
ArthurZucker
committed
[testing] Fix `seed_oss` (#41052)

authored and
ArthurZucker
committed
Remove repeated import (#40937)

cyyever
authored and
ArthurZucker
committed
Simplify unnecessary Optional typing (#40839)

cyyever
authored and
ArthurZucker
committed
Add write token for uploading benchmark results to the Hub (#41047)

ahadnagy
authored and
ArthurZucker
committed
Ci utils (#40978)

remi-or
authored and
ArthurZucker
committed
Fix CI jobs being all red 🔴 (false positive) (#41059)

ydshieh
authored and
ArthurZucker
committed
Update quantization CI (#41068)

SunMarc
authored and
ArthurZucker
committed
[i18n-bn] Add Bengali language README file (#40935)

saidurpulok
authored and
ArthurZucker
committed
Improve documentation and errors in Mamba2-based models (#41063)

mapmeld
authored and
ArthurZucker
committed
Update team member list for some CI workflows (#41094)

ydshieh
authored and
ArthurZucker
committed
fix crash when using chat to send 2+ request to gptoss (#40536)

sywangyi
authored and
ArthurZucker
committed
Minor addition, no split modules for VideoMAEE (#41051)

authored and
ArthurZucker
committed
Switch to `python:3.10-slim` for CircleCI docker images (#41067)

ydshieh
authored and
ArthurZucker
committed
Fix argument name in benchmarking script (#41086)

ahadnagy
authored and
ArthurZucker
committed
Fix typos in documentation (#41087)

cyyever
authored and
ArthurZucker
committed
Fix typing (#40788)

cyyever
authored and
ArthurZucker
committed
Remove unused arguments (#40916)

cyyever
authored and
ArthurZucker
committed
fix wrong height and width when read video use torchvision (#41091)

Juude
authored and
ArthurZucker
committed
docs: Fix Tool Use links and remove dead RAG links (#41104)

RyanMullins
authored and
ArthurZucker
committed
[tests] gpt2 + `CausalLMModelTester` (#41003)

gante
authored and
ArthurZucker
committed
Fix `_get_test_info` for inherited tests (#41106)

ydshieh
authored and
ArthurZucker
committed
Remove bad test skips (#41109)

Cyrilvallez
authored and
ArthurZucker
committed
Format empty lines and white space in markdown files. (#41100)

cyyever
authored and
ArthurZucker
committed
Update ruff to 0.13.1 + target Python 3.10 + apply fixes (#37809)

authored and
ArthurZucker
committed
Support loading LFM2 GGUF (#41111)

authored and
ArthurZucker
committed
[torchao safetensors] integrate torchao safetensors support with transformers (#40735)

liangel-02
authored and
ArthurZucker
committed
[Qwen3-next] Fix dimension mismatch in torch_chunk_gated_delta_rule and torch_recurrent_gated_delta_rule (#40963) (#41036)

authored and
ArthurZucker
committed
Fix the error where a keyword argument appearing before *args (#41099)

cyyever
authored and
ArthurZucker
committed
Fix broken `` expressions in markdown files (#41113)

cyyever
authored and
ArthurZucker
committed
Remove self-assignment (#41062)

authored and
ArthurZucker
committed
Fixed MXFP4 model storage issue (#41118)

YangKai0616
authored and
ArthurZucker
committed
Fixed loading LongT5 from legacy checkpoints (#40724)

Szustarol
authored and
ArthurZucker
committed
dummy commit (#41133)

ydshieh
authored and
ArthurZucker
committed
Fix loading logic flaw with regards to unexpected and missing keys (#40850)

authored and
ArthurZucker
committed
Fix: align Qwen2.5-VL inference rope index with training by passing s… (#41153)

Xqle
authored and
ArthurZucker
committed
Fix single quotes in markdown (#41154)

cyyever
authored and
ArthurZucker
committed
extend gemma3n integration ut cases on XPU (#41071)

yao-matrix
authored and
ArthurZucker
committed
Add Parakeet (#39062)

authored and
ArthurZucker
committed
Fix format of compressed_tensors.md (#41155)

cyyever
authored and
ArthurZucker
committed
Simplify and improve model loading logic (#41103)

Cyrilvallez
authored and
ArthurZucker
committed
Force new vision models addition to include a fast image processor (#40802)

yonigozlan
authored and
ArthurZucker
committed
Add language specifiers to code blocks of markdown files (#41114)

authored and
ArthurZucker
committed
Improve `add_dates` script (#41167)

yonigozlan
authored and
ArthurZucker
committed
Fix flash-attn for paged_attention when no kernels (#41078)

authored and
ArthurZucker
committed
Remove data from examples (#41168)

LysandreJik
authored and
ArthurZucker
committed
Enable fa in amd docker (#41069)

authored and
ArthurZucker
committed
handle flash slow tests (#41072)

authored and
ArthurZucker
committed
Modernbert fix (#41056)

remi-or
authored and
ArthurZucker
committed
CI Runners - move amd runners mi355 and 325 to runner group (#41193)

glegendre01
authored and
ArthurZucker
committed
[XPU] Add MXFP4 support for XPU (#41117)

authored and
ArthurZucker
committed
[tests] `CausalLMTester` automatically infers other test classes from `base_model_class` 🐛 🔫 (#41066)

gante
authored and
ArthurZucker
committed
More typing fixes (#41102)

cyyever
authored and
ArthurZucker
committed
enable flex attention ut cases on XPU (#40989)

authored and
ArthurZucker
committed
fix(trainer): Avoid moving model with device_map (#41032)

authored and
ArthurZucker
committed
Fix attention sink implementation in flex attention (#41083)

SamuelBarryCS
and
ArthurZucker
committed
Separate docker images for Nvidia and AMD in benchmarking (#41119)

ahadnagy
authored and
ArthurZucker
committed
Make quantizers good citizens loading-wise (#41138)

Cyrilvallez
authored and
ArthurZucker
committed
[`Kernels Attention`] Change fallback logic to error out on explicit kernels request and include FA3 (#41010)

vasqu
authored and
ArthurZucker
committed
Add EdgeTAM (#39800)

authored and
ArthurZucker
committed
Fix EXAONE-4.0 dummy id (#41089)

authored and
ArthurZucker
committed
Fix 8bit bnb loading (#41200)

SunMarc
authored and
ArthurZucker
committed
Fix docker quantization (#41201)

SunMarc
authored and
ArthurZucker
committed
Embed interactive timeline in docs (#41015)

yonigozlan
authored and
ArthurZucker
committed
[docs] Fix links (#41110)

stevhliu
authored and
ArthurZucker
committed
Remove unnecessary Optional typing (#41198)

cyyever
authored and
ArthurZucker
committed
docs/examples(speech): pin CTC commands to Hub datasets; add Windows notes (#41027)

authored and
ArthurZucker
committed
Fix Qwen3-Omni audio_token_id serialization issue (#41192)

aug6th
authored and
ArthurZucker
committed
Wait for main process in _save_checkpoint to ensure best checkpoint exists (#40923)

ssharpe42
authored and
ArthurZucker
committed
Avoid assumption that model has config attribute in deepspeed (#41207)

tomaarsen
authored and
ArthurZucker
committed
Trainer: Pass `num_items_in_batch` to `compute_loss` in `prediction_step` (#41183)

authored and
ArthurZucker
committed
[ESM] add accepts_loss_kwargs=False to EsmPreTrainedModel (#41006)

authored and
ArthurZucker
committed
Align pull request template to bug report template (#41220)

tomaarsen
authored and
ArthurZucker
committed
[generate] cache missing custom generate file (#41216)

gante
authored and
ArthurZucker
committed
Remove old Python code (#41226)

cyyever
authored and
ArthurZucker
committed
Adapt to the SDPA interface to enable the NPU to call FlashAttentionScore (#41143)

authored and
ArthurZucker
committed
update code owners (#41221)

ydshieh
authored and
ArthurZucker
committed
Unify is_torchvision_v2_available with is_torchvision_available (#41227)

cyyever
authored and
ArthurZucker
committed
Fix typing of train_args (#41142)

cyyever
authored and
ArthurZucker
committed
Fix sliding window attn mask (#41228)

authored and
ArthurZucker
committed
Revert "Fix DeepSpeed mixed precision precedence over Accelerate defaults" (#41124)

SunMarc
authored and
ArthurZucker
committed
[docs] Fix tp_plan (#41205)

stevhliu
authored and
ArthurZucker
committed
Fix white space in documentation (#41157)

cyyever
authored and
ArthurZucker
committed
fix qwen text config (#41158)

zucchini-nlp
authored and
ArthurZucker
committed
Video processor accepts single frames on cuda (#41218)

zucchini-nlp
authored and
ArthurZucker
committed
Use math.log2 (#41241)

cyyever
authored and
ArthurZucker
committed
fix TrainerIntegrationDeepSpeed UT failures (#41236)

yao-matrix
authored and
ArthurZucker
committed
[repo utils] Update `models_to_deprecate.py` (#41231)

gante
authored and
ArthurZucker
committed
Use removeprefix and removesuffix (#41240)

cyyever
authored and
ArthurZucker
committed
Fix pylint warnings (#41222)

cyyever
authored and
ArthurZucker
committed
Remove all instances of `is_safetensors_available` (#41233)

authored and
ArthurZucker
committed
FP-Quant NVFP4 and Python 3.9 support (#39876)

authored and
ArthurZucker
committed
[`FA3`] Fix masking and loading logic in same process (#41217)

vasqu
authored and
ArthurZucker
committed
[t5gemma] fix `get_text_config` and related fixes (#40939)

gante
authored and
ArthurZucker
committed
Don't convert to `safetensors` on the fly if the call is from testing (#41194)

authored and
ArthurZucker
committed
Resolve remote custom module path warnings (#41243)

XuehaiPan
authored and
ArthurZucker
committed
add peft team members to issue/pr template (#41262)

authored and
ArthurZucker
committed
docs: update bitsandbytes platform support (#41266)

matthewdouglas
authored and
ArthurZucker
committed
add more activation kernels, follow up (#40944)

MekkCyber
authored and
ArthurZucker
committed
fix asr pipeline ut failures (#41275)

yao-matrix
authored and
ArthurZucker
committed
Use regex defailed flags (#41264)

cyyever
authored and
ArthurZucker
committed
Fix multi-video timestamp bug in Qwen-3-VL and GLM4V (#41229)

authored and
ArthurZucker
committed
Fix binding of video frames to video placeholder in `InternVL` model (#41237)

daskol
authored and
ArthurZucker
committed
Deprecate Trackio environment variables and deploy to Spaces by default (#40950)

qgallouedec
authored and
ArthurZucker
committed
Allow private Space id for Trackio (#40948)

qgallouedec
authored and
ArthurZucker
committed
fix async client for transformers chat (#41255)

SunMarc
authored and
ArthurZucker
committed
Unify is_torchvision_v2_available with is_torchvision_available (#41259)

cyyever
authored and
ArthurZucker
committed
Use max/min (#41280)

cyyever
authored and
ArthurZucker
committed
Biogptlogits (#41270)

0x-avi
authored and
ArthurZucker
committed
Fix unnecessary single-item container checks (#41279)

cyyever
authored and
ArthurZucker
committed
Fix pylint generator warnings (#41258)

cyyever
authored and
ArthurZucker
committed
feat: use `aws-highcpu-32-priv` for amd docker img build (#41285)

McPatate
authored and
ArthurZucker
committed
Add processor and intergration test for qwen3vl (#41277)

JJJYmmm
authored and
ArthurZucker
committed
Remove `test_initialization` (#41261)

Cyrilvallez
authored and
ArthurZucker
committed
Remove some previous team members from allow list of triggering Github Actions (#41263)

ydshieh
authored and
ArthurZucker
committed
Build doc in 2 jobs: `en` and `other languages` (#41290)

ydshieh
authored and
ArthurZucker
committed
Fix mxfp4 dequantization (#41292)

Cyrilvallez
authored and
ArthurZucker
committed
[`Flex Attn`] Fix lse x attention sinks logic (#41249)

vasqu
authored and
ArthurZucker
committed
FIX: Bug in PEFT integration delete_adapter method (#41252)

BenjaminBossan
authored and
ArthurZucker
committed
Italian translation for README.md (#41269)

fedtti
authored and
ArthurZucker
committed
Fix README.md error when installing from source (#41303)

TKONIY
authored and
ArthurZucker
committed
download and use HF Hub Cache (#41181)

ydshieh
authored and
ArthurZucker
committed
fix some merge issues
ArthurZucker
committed
[test_all]
ArthurZucker
committed
[test-all]
ArthurZucker
committed