huggingface/accelerate

Pull Requests Commits

Bookmark, not working yet

muellerzr committed 1 year ago

810dd387

Should pass the trainer tests

muellerzr committed 1 year ago

c73b9e9c

Upstream to Accelerator

muellerzr committed 1 year ago

0d4c0a1c

Need to add in len for dl

muellerzr committed 1 year ago

e8461cce

Co-authored-by: byi8220 <byi8220@gmail.com>

muellerzr committed 1 year ago

a92bab36

muellerzr committed 1 year ago

9af91af3

muellerzr committed 1 year ago

a6e192c0

Fix bug of clip_grad_norm_ for xla fsdp (#2941)

append-only committed 1 year ago

Verified 288accc0

remove .md to allow proper linking (#2977)

nbroad1881 committed 1 year ago

Verified 83b06101

add MLU devices for rng state saving and loading. (#2940)

huismiling committed 1 year ago

Verified 386f7d28

chore: Update runs-on configuration for CI workflows (#2981)

XciD committed 1 year ago

Verified 308a8e96

Enable Unwrapping for Model State Dicts (FSDP) (#2959)

alex-jw-brooks committed 1 year ago

Verified f35cbd1f

Fix torchvision to be compatible with torch version in CI (#2982)

SunMarc committed 1 year ago

Verified a14260c9

Require safetensors>=0.4.3 (#2957)

byi8220 committed 1 year ago

Verified 32f368ec

feat(ci): add `pip` caching in CI (#2952)

SauravMaheshkar committed 1 year ago

Verified 415eddf1

Properly handle Params4bit in set_module_tensor_to_device (#2934)

matthewdouglas committed 1 year ago

Verified 23085769

Add `torch.float8_e4m3fn` format `dtype_byte_size` (#2945)

SunMarc committed 1 year ago

Verified a5a3e571

delete CCL env var setting (#2927)

Liangliang-Ma committed 1 year ago

Verified 0af1d8b8

Improve test reliability for Accelerator.free_memory() (#2935)

matthewdouglas committed 1 year ago

Verified d16d7371

Consider pynvml available when installed through the nvidia-ml-py distribution (#2936)

matthewdouglas committed 1 year ago

Verified 7a5c231b

Fix import test (#2931)

muellerzr committed 1 year ago

Verified 4f02bb76

Hotfix PyTorch Version Installation in CI Workflow for Minimum Version Matrix (#2889)

yhna940 committed 1 year ago

Verified 709fd1e4

Correct loading of models with shared tensors when using accelerator.load_state() (#2875)

jkuntzer committed 1 year ago

Verified f4f1260a

Allow multiple process per device (#2916)

cifkao committed 1 year ago

Verified c6da9f86

Add huggingface_hub version to setup.py (#2932)

nullquant committed 1 year ago

Verified 3ebbe573

add xpu device check before moving tensor directly to xpu device (#2928)

faaany committed 1 year ago

Verified 24bf5ec5

Better error when a bad directory is given for weight merging (#2852)

muellerzr committed 1 year ago

Verified e1247de0

Support MUSA (Moore Threads GPU) backend in accelerate (#2917)

fmo-mt committed 1 year ago

Verified 12a007d5

fix: bug where mulit_gpu was being set and warning being printed even with num_processes=1 (#2921)

HarikrishnanBalagopal committed 1 year ago

Verified 5bdcd7e1

Fix slowdown on init with `device_map="auto"` (#2914)

muellerzr committed 1 year ago

Verified 2471eacd

Older