NielsRogge/transformers

Pull Requests Commits

NielsRogge committed 2 years ago

eecacc59

Use `weights_only` only if torch >= 1.13 (#28506)

ydshieh committed 2 years ago

Verified a1668cc7

Save `Processor` (#27761)

ydshieh committed 2 years ago

Verified 3005f965

Fix Switch Transformers When sparse_step = 1 (#28564)

agemagician committed 2 years ago

Verified 98dda8ed

Allow to train dinov2 with different dtypes like bf16 (#28504)

StarCycle committed 2 years ago

Verified fa6d12f7

Fix SDPA tests (#28552)

fxmarty committed 2 years ago

Verified 2c1eebc1

Add qwen2 (#28436)

JustinLin610 committed 2 years ago

Verified d6ffe74d

Fixes default value of `softmax_scale` in `PhiFlashAttention2`. (#28537)

gugarosa committed 2 years ago

Verified d93ef7d7

symbolic_trace: add past_key_values, llama, sdpa support (#28447)

fxmarty committed 2 years ago

Verified a6adc05e

[Makefile] Exclude research projects from format (#28551)

patrickvonplaten committed 2 years ago

Verified 09eb11a1

Config: warning when saving generation kwargs in the model config (#28514)

gante committed 2 years ago

Verified f4f57f9d

Add is_model_supported for fx (#28521)

inisis committed 2 years ago

Verified 7142bdfa

Clearer error for SDPA when explicitely requested (#28006)

fxmarty committed 2 years ago

Verified 02f8738e

[`SpeechT5Tokenization`] Add copied from and fix the `convert_tokens_to_string` to match the fast decoding scheme (#28522)

ArthurZucker committed 2 years ago

Verified fe23256b

[`TokenizationRoformerFast`] Fix the save and loading (#28527)

ArthurZucker committed 2 years ago

Verified 96d08831

[ `TokenizationUtils`] Fix `add_special_tokens` when the token is already there (#28520)

ArthurZucker committed 2 years ago

Verified 716df5fb

Fix/speecht5 bug (#28481)

NimaYaqmuri committed 2 years ago

Verified 07ae53e6

Fix mismatching loading in from_pretrained with/without accelerate (#28414)

fxmarty committed 2 years ago

Verified 66db33dd

Improving Training Performance and Scalability Documentation (#28497)

HamzaFB committed 2 years ago

Verified 002566f3

Remove `task` arg in `load_dataset` in image-classification example (#28408)

regisss committed 2 years ago

Verified 0cdcd7a2

SiLU activation wrapper for safe importing (#28509)

amyeroberts committed 2 years ago

Verified edb17023

improve dev setup comments and hints (#28495)

4imothy committed 2 years ago

Verified ff86bc36

fix: sampling in flax keeps EOS (#28378)

borisdayma committed 2 years ago

Verified 735968b6

Generate: consolidate output classes (#28494)

gante committed 2 years ago

Verified 7e0ddf89

Add a use_safetensors arg to TFPreTrainedModel.from_pretrained() (#28511)

Rocketknight1 committed 2 years ago

Verified 72db39c0

Fixed minor typos (#28489)

rishit5 committed 2 years ago

Verified 78d767e3

[GPTQ] Fix test (#28018)

SunMarc committed 2 years ago

Verified 7c8dd88d

Tokenizer kwargs in textgeneration pipe (#28362)

farzadsbakhtiar committed 2 years ago

Verified 366c0327

Add the XPU device check for pipeline mode (#28326)

yuanwu2017 committed 2 years ago

Verified a573ac74

[`core`/ FEAT] Add the possibility to push custom tags using `PreTrainedModel` itself (#28405)

younesbelkada committed 2 years ago

Verified 1b9a2e4c

Older