github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-04 02:26:51 +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

5
examples/language/llama2/benchmark.py
View File

@@ -13,6 +13,7 @@ from transformers.models.llama.configuration_llama import LlamaConfig
from transformers.models.llama.modeling_llama import LlamaForCausalLM
import colossalai
import colossalai.utils.device as device_utils
from colossalai.booster import Booster
from colossalai.booster.plugin import GeminiPlugin, HybridParallelPlugin, TorchFSDPPlugin
from colossalai.cluster import DistCoordinator
@@ -194,7 +195,7 @@ def main():
torch.set_default_dtype(torch.bfloat16)
model, optimizer, _, dataloader, _ = booster.boost(model, optimizer, dataloader=dataloader)
torch.set_default_dtype(torch.float)
coordinator.print_on_master(f"Booster init max CUDA memory: {torch.cuda.max_memory_allocated()/1024**2:.2f} MB")
coordinator.print_on_master(f"Booster init max CUDA memory: {device_utils.max_memory_allocated()/1024**2:.2f} MB")
coordinator.print_on_master(
f"Booster init max CPU memory: {resource.getrusage(resource.RUSAGE_SELF).ru_maxrss/1024:.2f} MB"
)
@@ -220,7 +221,7 @@ def main():
performance_evaluator.on_step_end(**batch)
performance_evaluator.on_fit_end()
coordinator.print_on_master(f"Max CUDA memory usage: {torch.cuda.max_memory_allocated()/1024**2:.2f} MB")
coordinator.print_on_master(f"Max CUDA memory usage: {device_utils.max_memory_allocated()/1024**2:.2f} MB")
if __name__ == "__main__":

8
examples/language/llama2/performance_evaluator.py
View File

@@ -5,7 +5,9 @@ import torch
import torch.distributed as dist
from torch import Tensor
import colossalai.utils.device as device_utils
from colossalai.cluster import DistCoordinator
from colossalai.utils.device import get_current_device
def divide(x: float, y: float) -> float:
@@ -20,7 +22,7 @@ def divide(x: float, y: float) -> float:
def all_reduce_mean(x: float, world_size: int) -> float:
if world_size == 1:
return x
tensor = torch.tensor([x], device=torch.cuda.current_device())
tensor = torch.tensor([x], device=get_current_device())
dist.all_reduce(tensor)
tensor = tensor / world_size
return tensor.item()
@@ -84,13 +86,13 @@ class PerformanceEvaluator:
self.disable = self.ignore_steps > 0 and step < self.ignore_steps
if self.disable:
return
torch.cuda.synchronize()
device_utils.synchronize()
self.timer.start()
def on_step_end(self, input_ids: Tensor, **kwargs) -> None:
if self.disable:
return
torch.cuda.synchronize()
device_utils.synchronize()
self.timer.end()
batch_size, seq_len = input_ids.shape