safetensors格式文件打开和读取＜未解决＞_safetensors文件怎么打开

转 https://blog.csdn.net/wzk4869/article/details/130668642
显卡：GTX 3060

safetensors

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）。

1.1 pip 安装

pip install safetensors
conda install -c huggingface safetensors
1
2

1.2 加载张量

from safetensors import safe_open

tensors = {}
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_open

tensors = {}
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]

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

1.3 保存张量

import torch
from safetensors.torch import save_file

tensors = {
    "embedding": torch.zeros((2, 2)),
    "attention": torch.zeros((2, 3))
}
save_file(tensors, "model.safetensors")
1
2
3
4
5
6
7
8

1.4 速度比较
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
1

让我们从导入所有将使用的包开始：

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

# 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")

# 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}")
# 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_pt
print(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
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

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。

# 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_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")
# 输出结果为：
# Loaded safetensors 0:00:00.497415
# Loaded pytorch 0:00:00.250602
# on GPU, safetensors is faster than pytorch by: 0.5 X
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

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。

==========================================================
实际操作中，不知道为什么上面的代码是跑不通的
打开safetensors格式文件要换safetensors库的safe_open

from safetensors import safe_open

tensors1 = {}
with safe_open(filex, framework="pt", device="cuda:0") as f:
    for k in f.keys():
        tensors1[k] = f.get_tensor(k)
1
2
3
4
5
6

safe_open下面也没有modules属性
safetensors库运行时发现modules()函数不存在
torch.nn.Module的children()函数也不存在，named_children()也不存在
pytorch版本 pytorch-2.1.0-py3.9_cpu_0

具体统计卷积层数量，池化层数量，全连接层数量的代码（会报错）

import torch
import torch.nn as nn

# 加载模型
filex = 'D:/Program Files/sdxl-webui-1.0/sdxl-webui-1.0/models/Stable-diffusion/sd_xl_base_1.0.safetensors'
filexs = 'C:/Users/admin/Desktop/model.pth'

# torch.save(filex.state_dict(), filexs)
# model.load_state_dict(torch.load(filexs))
# model = torch.load('model.pth')

# 统计各层数量
conv_count = 0
pool_count = 0
fc_count = 0

# 遍历模型各层
for layer in model.modules():
    if isinstance(layer, nn.Conv2d):
        conv_count += 1
    elif isinstance(layer, nn.MaxPool2d):
        pool_count += 1
    elif isinstance(layer, nn.Linear):
        fc_count += 1

# 输出各层数量
print("卷积层数量：", conv_count)
print("池化层数量：", pool_count)
print("全连接层数量：", fc_count)
# ===================================================


import torch.nn as nn
from safetensors import safe_open

# 加载模型
filex = 'D:/Program Files/sdxl-webui-1.0/sdxl-webui-1.0/models/Stable-diffusion/sd_xl_base_1.0.safetensors'
filexs = 'C:/Users/admin/Desktop/model2.safetensors'
model = safe_open(filex, framework="pt", device="cuda:0")  # 加载模型的完整路径

def get_modules(module):
    modules = [module]
    for name, child in module.nn.named_children():
        modules += get_modules(child)
    return modules

# 统计各层数量  
conv_count = 0
pool_count = 0
fc_count = 0

modules = get_modules(model)

for module in modules:
    if isinstance(module, nn.Conv2d):
        conv_count += 1
    elif isinstance(module, nn.MaxPool2d):
        pool_count += 1
    elif isinstance(module, nn.Linear):
        fc_count += 1

# 输出各层数量
print("卷积层数量：", conv_count)
print("池化层数量：", pool_count)
print("全连接层数量：", fc_count)


# ===================================================

import torch.nn as nn
from safetensors import safe_open

# 加载模型
filex = 'D:/Program Files/sdxl-webui-1.0/sdxl-webui-1.0/models/Stable-diffusion/sd_xl_base_1.0.safetensors'
filexs = 'C:/Users/admin/Desktop/model2.safetensors'
model = safe_open(filex, framework="pt", device="cuda:0")  # 加载模型的完整路径

# 保存模型的状态字典  
# torch.save(model.state_dict(), filexs)

# 重新加载模型的状态字典  
# model_state_dict = torch.load(filexs)

def get_modules(module):
    modules = [module]
    for name, child in module.nn.named_children():
        modules += get_modules(child)
    return modules

# 统计各层数量  
conv_count = 0
pool_count = 0
fc_count = 0

modules = get_modules(model)

for module in modules:
    if isinstance(module, nn.Conv2d):
        conv_count += 1
    elif isinstance(module, nn.MaxPool2d):
        pool_count += 1
    elif isinstance(module, nn.Linear):
        fc_count += 1

# 输出各层数量
print("卷积层数量：", conv_count)
print("池化层数量：", pool_count)
print("全连接层数量：", fc_count)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109

暂时还没找到原因