NielsRogge/transformers

Pull Requests Commits

DeepSpeed: hardcode `torch.arange` dtype on `float` usage to avoid incorrect initialization (#28760)

gante committed 2 years ago

Verified beb2a096

Flax mistral (#26943)

kiansierra committed 2 years ago

Verified f7076cd3

Wrap Keras methods to support BatchEncoding (#28734)

Rocketknight1 committed 2 years ago

Verified 7a496100

canonical repos moves (#28795)

julien-c committed 2 years ago

Verified 721e2d94

Resolve DeepSpeed cannot resume training with PeftModel (#28746)

lh0x00 committed 2 years ago

Verified bebeeee0

[Whisper] Refactor forced_decoder_ids & prompt ids (#28687)

patrickvonplaten committed 2 years ago

Verified 65a926e8

[`HFQuantizer`] Remove `check_packages_compatibility` logic (#28789)

younesbelkada committed 2 years ago

Verified f9f1f2ac

don't initialize the output embeddings if we're going to tie them to input embeddings (#28192)

tom-p-reichel committed 2 years ago

Verified ae0c27ad

Prevent MLflow exception from disrupting training (#28779)

codiceSpaghetti committed 2 years ago

Verified a937425e

[`bnb`] Fix bnb slow tests (#28788)

younesbelkada committed 2 years ago

Verified d703eaae

Pin Torch to <2.2.0 (#28785)

Rocketknight1 committed 2 years ago

Verified 74c9cfea

Add tf_keras imports to prepare for Keras 3 (#28588)

Rocketknight1 committed 2 years ago

Verified 415e9a09

Task-specific pipeline init args (#28439)

amyeroberts committed 2 years ago

Verified 1d489b3e

[`Backbone`] Use `load_backbone` instead of `AutoBackbone.from_config` (#28661)

amyeroberts committed 2 years ago

Verified 2fa1c808

Further pin pytest version (in a temporary way) (#28780)

ydshieh committed 2 years ago

Verified c24c5245

Fix transformers.utils.fx compatibility with torch<2.0 (#28774)

fxmarty committed 2 years ago

Verified 6f7d5db5

Use Conv1d for TDNN (#25728)

gau-nernst committed 2 years ago

Verified 5c8d941d

[`HfQuantizer`] Move it to "Developper guides" (#28768)

younesbelkada committed 2 years ago

Verified 866253f8

`HfQuantizer` class for quantization-related stuff in `modeling_utils.py` (#26610)

poedator committed 2 years ago

Verified d78e78a0

Move CLIP _no_split_modules to CLIPPreTrainedModel (#27841)

lz1oceani committed 2 years ago

Verified 1f5590d3

Don't allow passing `load_in_8bit` and `load_in_4bit` at the same time (#28266)

osanseviero committed 2 years ago

Verified a989c6c6

Add French translation: french README.md (#28696)

ThibaultLengagne committed 2 years ago

Verified cd2eb8cb

Support saving only PEFT adapter in checkpoints when using PEFT + FSDP (#28297)

AjayP13 committed 2 years ago

Verified a055d09e

[Whisper] Make tokenizer normalization public (#28136)

sanchit-gandhi committed 2 years ago

Verified da3c79b2

Fix typo of `Block`. (#28727)

xkszltl committed 2 years ago

Verified e694e985

Mark test_constrained_beam_search_generate as flaky (#28757)

amyeroberts committed 2 years ago

Verified 9e8f35fa

Pin pytest version <8.0.0 (#28758)

amyeroberts committed 2 years ago

Verified 0f8d015a

small doc update for CamemBERT (#28644)

julien-c committed 2 years ago

Verified 26aa03a2

Enable Gradient Checkpointing in Deformable DETR (#28686)

FoamoftheSea committed 2 years ago

Verified 0548af54

PatchtTST and PatchTSMixer fixes (#28083)

wgifford committed 2 years ago

Verified f72c7c22

Newer Older