速度对比

Safetensors 的速度非常快。让我们通过加载 gpt2 权重，将其与 PyTorch 进行对比。要运行 GPU 基准测试，请确保您的机器配备了 GPU，或者如果您使用的是 Google Colab，请已选择 GPU runtime。

在开始之前，请确保已安装所有必要的库：

pip install safetensors huggingface_hub torch

让我们从导入所有将要使用的包开始：

>>> import os
>>> import datetime
>>> from huggingface_hub import hf_hub_download
>>> from safetensors.torch import load_file
>>> import torch

下载 gpt2 的 safetensors & torch 权重：

>>> sf_filename = hf_hub_download("gpt2", filename="model.safetensors")
>>> pt_filename = hf_hub_download("gpt2", filename="pytorch_model.bin")

CPU 基准测试

>>> start_st = datetime.datetime.now()
>>> weights = load_file(sf_filename, device="cpu")
>>> load_time_st = datetime.datetime.now() - start_st
>>> print(f"Loaded safetensors {load_time_st}")

>>> start_pt = datetime.datetime.now()
>>> weights = torch.load(pt_filename, map_location="cpu")
>>> load_time_pt = datetime.datetime.now() - start_pt
>>> print(f"Loaded pytorch {load_time_pt}")

>>> print(f"on CPU, safetensors is faster than pytorch by: {load_time_pt/load_time_st:.1f} X")
Loaded safetensors 0:00:00.004015
Loaded pytorch 0:00:00.307460
on CPU, safetensors is faster than pytorch by: 76.6 X

这种加速效果源于该库通过直接映射文件来避免不必要的复制。实际上，在纯 PyTorch中也可以实现。当前展示的加速效果是在以下环境中获得的：

操作系统：Ubuntu 18.04.6 LTS
CPU: Intel(R) Xeon(R) CPU @ 2.00GHz

GPU 基准测试

>>> # This is required because this feature hasn't been fully verified yet, but 
>>> # it's been tested on many different environments
>>> os.environ["SAFETENSORS_FAST_GPU"] = "1"

>>> # CUDA startup out of the measurement
>>> torch.zeros((2, 2)).cuda()

>>> start_st = datetime.datetime.now()
>>> weights = load_file(sf_filename, device="cuda:0")
>>> load_time_st = datetime.datetime.now() - start_st
>>> print(f"Loaded safetensors {load_time_st}")

>>> start_pt = datetime.datetime.now()
>>> weights = torch.load(pt_filename, map_location="cuda:0")
>>> load_time_pt = datetime.datetime.now() - start_pt
>>> print(f"Loaded pytorch {load_time_pt}")

>>> print(f"on GPU, safetensors is faster than pytorch by: {load_time_pt/load_time_st:.1f} X")
Loaded safetensors 0:00:00.165206
Loaded pytorch 0:00:00.353889
on GPU, safetensors is faster than pytorch by: 2.1 X

该加速之所以有效，是因为此库能够跳过不必要的 CPU 内存分配。据我们所知，这在纯 PyTorch 中无法复现。该库的工作原理是：通过内存映射文件，使用 PyTorch 创建一个空张量，然后直接调用 cudaMemcpy 将张量直接传输到 GPU 上。当前展示的加速效果是在以下环境中获得的：

操作系统：Ubuntu 18.04.6 LTS。
GPU：Tesla T4
驱动程序版本：460.32.03
CUDA 版本：11.2

Update on GitHub