safetensors 是一种用于安全存储张量(与 pickle 相反)的新型简单格式,并且仍然很快(零拷贝)。

safetensors 真的很快。

一、安装

1.1 pip 安装

pip install safetensors

1.2 conda 安装

conda install -c huggingface safetensors

二、加载张量

from safetensors import safe_opentensors = {}with safe_open("model.safetensors", framework="pt", device=0) as f:for k in f.keys():tensors[k] = f.get_tensor(k)

仅加载部分张量(在多个GPU上运行时很有趣):

from safetensors import safe_opentensors = {}with safe_open("model.safetensors", framework="pt", device=0) as f:tensor_slice = f.get_slice("embedding")vocab_size, hidden_dim = tensor_slice.get_shape()tensor = tensor_slice[:, :hidden_dim]

三、保存张量

import torchfrom safetensors.torch import save_filetensors = {"embedding": torch.zeros((2, 2)),"attention": torch.zeros((2, 3))}save_file(tensors, "model.safetensors")

四、速度比较

4.1 下载 gpt2 的文件

safetensors 真的很快。让我们通过加载 gpt2 权重将其进行比较。要运行 GPU 基准测试,请确保您的机器具有 GPU,或者您已选择是否使用的是 Google Colab。

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使用的包开始:

import osimport datetimefrom huggingface_hub import hf_hub_downloadfrom safetensors.torch import load_fileimport torch

Download safetensors & torch weights for gpt2:

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

4.2 CPU 基准测试

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

输出结果为:

Loaded safetensors 0:00:00.026842
start_pt = datetime.datetime.now()weights = torch.load(pt_filename, map_location="cpu")load_time_pt = datetime.datetime.now() - start_ptprint(f"Loaded pytorch {load_time_pt}")

输出结果为:

Loaded pytorch 0:00:00.182266
print(f"on CPU, safetensors is faster than pytorch by: {load_time_pt/load_time_st:.1f} X")

输出结果为:

on CPU, safetensors is faster than pytorch by: 6.8 X

这种加速是由于该库通过直接映射文件来避免不必要的副本。实际上可以在 torch 上完成。 当前显示的加速比已打开:

  • 操作系统: Windows
  • 处理器: 英特尔® 至强® CPU @ 2.00GHz

4.3 GPU 基准测试

os.environ["SAFETENSORS_FAST_GPU"] = "1"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_stprint(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_ptprint(f"Loaded pytorch {load_time_pt}")print(f"on GPU, safetensors is faster than pytorch by: {load_time_pt/load_time_st:.1f} X")

输出结果为:

Loaded safetensors 0:00:00.497415Loaded pytorch 0:00:00.250602on GPU, safetensors is faster than pytorch by: 0.5 X

加速有效是因为此库能够跳过不必要的 CPU 分配。不幸的是,据我们所知,它无法在纯 pytorch 中复制。该库的工作原理是内存映射文件,使用 pytorch 创建空张量,并直接调用以直接在 GPU 上移动张量。

显卡:GTX 3060