huggingface/accelerate

Pull Requests Commits

Document, document, document

muellerzr committed 2 years ago

3c464369

Make note about recursion loop

muellerzr committed 2 years ago

694bb8a4

Allow for users to pass in max_meory

muellerzr committed 2 years ago

e3cb2888

Rm typing literal, only pippy for pippy

muellerzr committed 2 years ago

e974624c

Do it at tracing too

muellerzr committed 2 years ago

6e1e02f6

All tests passing!

muellerzr committed 2 years ago

5970e8e3

Use pad_input_tensor

muellerzr committed 2 years ago

136f495c

Much cleaner implementation

muellerzr committed 2 years ago

a45e7f96

Almost working version

muellerzr committed 2 years ago

6864314c

add some failing test

SunMarc committed 2 years ago

57277673

muellerzr committed 2 years ago

7e958024

muellerzr committed 2 years ago

6167d0ba

muellerzr committed 2 years ago

7045a290

Update the source

muellerzr committed 2 years ago

fcc72a37

muellerzr committed 2 years ago

fed86c4c

Refactor to utils

muellerzr committed 2 years ago

1a9181c4

Use dataloader-like logic

muellerzr committed 2 years ago

c7800f50

Start, need to test

muellerzr committed 2 years ago

2e23e1e9

muellerzr committed 2 years ago

8a452c18

Less slicy-dicy

muellerzr committed 2 years ago

a98e51b0

Break early after the first valid bs is found

muellerzr committed 2 years ago

b1c565d0

Update src/accelerate/inference.py

muellerzr committed 2 years ago

Verified 0af31ff0

muellerzr committed 2 years ago

3534a342

Allow for dynamic batch paddign

muellerzr committed 2 years ago

73e64a1e

fix case num_process=1

SunMarc committed 2 years ago

7ca4bccf

Store split points in hf_split_points

muellerzr committed 2 years ago

8792a8c5

Put split_points in pipelien

muellerzr committed 2 years ago

e3f6b99b

Use no split module classes explicitly

muellerzr committed 2 years ago

df7779aa

muellerzr committed 2 years ago

77f8e92b

muellerzr committed 2 years ago

449eb8d9

Older