huggingface/accelerate

Pull Requests Commits

muellerzr committed 2 years ago

31f0f3a0

muellerzr committed 2 years ago

da9a67ad

muellerzr committed 2 years ago

5433ef9e

Expose a bit of args/docstring fixup (#1824)

muellerzr committed 2 years ago

Verified 7cb19ae6

Update docs and docstrings to match `load_and_quantize_model` arg (#1822)

JonathanRayner committed 2 years ago

Verified 39897a06

Fix bnb import (#1813)

muellerzr committed 2 years ago

Verified aa71bb81

add warning when using to and cuda (#1790)

SunMarc committed 2 years ago

Verified f43a08a9

Improve docs on grad accumulation (#1817)

vwxyzjn committed 2 years ago

Verified b42c65b7

muellerzr committed 2 years ago

Verified 7bad7269

Expand device-map warning (#1819)

muellerzr committed 2 years ago

Verified 29ff7c39

Typo fix (#1812)

muellerzr committed 2 years ago

Verified 30eff605

Detect device map auto and raise a helpful error (#1810)

muellerzr committed 2 years ago

Verified fc95663e

More specific logging in `gather_for_metrics` (#1784)

dleve123 committed 2 years ago

Verified 49cb83a4

Fix pytest import (#1808)

muellerzr committed 2 years ago

Verified d2b159ea

Add FSDP for NPU (#1806)

ji-huazhong committed 2 years ago

Verified 40056c69

Add FSDP for XPU (#1803)

abhilash1910 committed 2 years ago

Verified 505b5be0

Changed allow_val_change param (#1796)

SumanthRH committed 2 years ago

Verified a6333f2e

add support of float memory size in convert_file_size_to_int (#1799)

ranchlai committed 2 years ago

Verified 0dec4779

reserve 10% GPU in `get_balanced_memory` to avoid OOM (#1798)

ranchlai committed 2 years ago

Verified a24189db

Fix import error when torch>=2.0.1 and torch.distributed is disabled (#1800)

natsukium committed 2 years ago

Verified a9aee447

Set ipex default (#1776)

muellerzr committed 2 years ago

Verified d5894ab4

simplify and correct the deepspeed example (#1775)

pacman100 committed 2 years ago

Verified 6f14928e

[FSDP] Fix `load_fsdp_optimizer` (#1755)

awgu committed 2 years ago

Verified 777334a8

Contigous on gather (#1771)

muellerzr committed 2 years ago

Verified c3d82d24

Support wrapping multiple models in Accelerator.accumulate() (#1708)

yuxinyuan committed 2 years ago

Verified 6e70e79e

Introduce an experimental distributed operations framework (#1756)

muellerzr committed 2 years ago

Verified b3fc3c90

Change is_aim_available() function to not match aim >= 4.0.0 (#1769)

alberttorosyan committed 2 years ago

Verified a9d79163

Fix offload on disk when executing on CPU (#1762)

sgugger committed 2 years ago

Verified 0b36ca6e

Fix error when max_memory argument is in unexpected order (#1759)

ranchlai committed 2 years ago

Verified f3b7f9cf

Fix check failure in Accelerator.save_state using multi-gpu (#1760)

CyCle1024 committed 2 years ago

Verified b909bfac

Older