github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-04 10:34:41 +00:00
Code Issues Packages Projects Releases Wiki Activity

[npu] add npu support for gemini and zero (#5067)

Browse Source 
* [npu] setup device utils (#5047)

* [npu] add npu device support

* [npu] support low level zero

* [test] update npu zero plugin test

* [hotfix] fix import

* [test] recover tests

* [npu] gemini support npu (#5052)

* [npu] refactor device utils

* [gemini] support npu

* [example] llama2+gemini support npu

* [kernel] add arm cpu adam kernel (#5065)

* [kernel] add arm cpu adam

* [optim] update adam optimizer

* [kernel] arm cpu adam remove bf16 support
This commit is contained in:
Hongxin Liu
2023-11-20 16:12:41 +08:00
committed by GitHub
parent 8d56c9c389
commit e5ce4c8ea6
46 changed files with 994 additions and 233 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
colossalai/zero/gemini/chunk/chunk.py
View File

@@ -7,6 +7,7 @@ import torch.distributed as dist
from torch.distributed import ProcessGroup
from colossalai.utils import get_current_device
from colossalai.utils.device import IS_NPU_AVAILABLE
class TensorState(Enum):
@@ -172,7 +173,7 @@ class Chunk:
if self.chunk_temp is not None:
# this chunk is not closed
if self.chunk_temp.device.type == "cuda":
if self.chunk_temp.device.type == "cuda" or self.chunk_temp.device.type == "npu":
cuda_memory += self.chunk_mem
else:
cpu_memory += self.chunk_mem
@@ -191,10 +192,8 @@ class Chunk:
if self.chunk_temp is not None:
return self.chunk_temp.device.type
else:
if self.is_gathered:
return "cuda"
elif self.cuda_shard is not None:
return "cuda"
if self.is_gathered or self.cuda_shard is not None:
return "npu" if IS_NPU_AVAILABLE else "cuda"
else:
return "cpu"
@@ -329,12 +328,12 @@ class Chunk:
# when the current chunk is not synchronized with the optimizer
# just use another way for the movement
if not self.optim_sync_flag:
assert device.type == "cuda", "each chunk should first be moved to CUDA"
assert device.type == "cuda" or device.type == "npu", "each chunk should first be moved to CUDA"
self.__paired_shard_move()
self.optim_sync_flag = True
return
if device.type == "cuda":
if device.type == "cuda" or device.type == "npu":
assert device == get_current_device(), "can't move chunk to another device"
if self.cuda_shard:
@@ -484,7 +483,7 @@ class Chunk:
assert friend_chunk.is_gathered is True
self.cuda_global_chunk.copy_(friend_chunk.cuda_global_chunk)
self.optim_sync_flag = True
elif friend_chunk.device_type == "cuda" and self.device_type == "cuda":
elif friend_chunk.device_type in ("cuda", "npu") and self.device_type in ("cuda", "npu"):
self.cuda_shard.copy_(friend_chunk.cuda_shard)
self.optim_sync_flag = True
self.cpu_vis_flag = False