从Hugging Face中下载数据集、模型到本地

1、查看本地代理

2、下载数据集

将https://huggingface.co/datasets/YeungNLP/firefly-train-1.1M数据集下载到本地

import os
import json
from datasets import load_dataset

###设置代理，本地vpn
os.environ["http_proxy"] = "http://127.0.0.1:21882"
os.environ["https_proxy"] = "http://127.0.0.1:21882"

dataset = load_dataset("YeungNLP/firefly-train-1.1M")
dataset.save_to_disk("dataset/Salesforce/dialogstudio") # 保存到该目录下
print(len(dataset['train']))
print(dataset['train'][3])

with open('./dataset/data.json', 'w', encoding='utf-8') as fp:
    num = 0
    for i in range(len(dataset['train'])):
        if dataset['train'][i]['kind'] == 'Couplet':
            fp.write(json.dumps({'input': dataset['train'][i]['input'], 'output': dataset['train'][i]['target']},
                                ensure_ascii=False))
            fp.write('\n')
            num += 1
            print(f"已写入{num}条")

数据类型如下：

上述代码只读取对联数据，结果如下：

3、下载模型

3.1、直接下载到本地并加载

from transformers import T5Tokenizer, T5ForConditionalGeneration

os.environ["http_proxy"] = "http://127.0.0.1:21882"
os.environ["https_proxy"] = "http://127.0.0.1:21882"

# 首先，下载并保存tokenizer和模型
tokenizer = T5Tokenizer.from_pretrained("t5-small", cache_dir="./t5_model_v1")
model = T5ForConditionalGeneration.from_pretrained("t5-small", cache_dir="./t5_model_v1")
input_text = "translate English to German: How old are you?"
input_ids = tokenizer(input_text, return_tensors="pt").input_ids

outputs = model.generate(input_ids)
print(tokenizer.decode(outputs[0]))

结果如下：

3.2、从Hugging Face中手动下载指定好的模型文件到本地并加载

from transformers import T5Tokenizer, T5ForConditionalGeneration
import os

os.environ["http_proxy"] = "http://127.0.0.1:21882"
os.environ["https_proxy"] = "http://127.0.0.1:21882"

tokenizer = T5Tokenizer.from_pretrained("t5_model_v2")
model = T5ForConditionalGeneration.from_pretrained("t5_model_v2")

input_text = "translate English to German: How old are you?"
input_ids = tokenizer(input_text, return_tensors="pt").input_ids

outputs = model.generate(input_ids)
print(tokenizer.decode(outputs[0]))

下载模型如下配置：

结果如下：