convmodel

convmodel

convmodel提供了基于 Transformers GPT-2 模型的对话模型

特征

• 利用GPT2模型生成响应
• 处理多轮对话
• 提供有用的接口来微调模型并根据给定上下文生成响应

微调 GPT-2 模型并生成响应的简单示例：

convmodel import ConversationModel from convmodel import ConversationExample # Load model on GPU model = ConversationModel.from_pretrained("gpt2") # Define training/validation examples train_iterator = [ ConversationExample(conversation=[ "Hello", "Hi, how are you?", "Good, thank you, how about you?", "Good, thanks!" ]), ConversationExample(conversation=[ "I am hungry", "How about eating pizza?" ]), ] valid_iterator = [ ConversationExample(conversation=[ "Tired...", "Let's have a break!", "Nice idea!" ]), ] # Fine-tune model model.fit(train_iterator=train_iterator, valid_iterator=valid_iterator) # Generate response model.generate(context=["Hello", "How are you"], do_sample=True, top_p=0.95, top_k=50) # Output could be like below if sufficient examples were given. # => ConversationModelOutput(responses=['Good thank you'], context=['Hello', 'How are you'])">

 from convmodel import ConversationModel
from convmodel import ConversationExample

# Load model on GPU
model = ConversationModel . from_pretrained ( "gpt2" )

# Define training/validation examples
train_iterator = [
    ConversationExample ( conversation = [
        "Hello" ,
        "Hi, how are you?" ,
        "Good, thank you, how about you?" ,
        "Good, thanks!"
    ]),
    ConversationExample ( conversation = [
        "I am hungry" ,
        "How about eating pizza?"
    ]),
]
valid_iterator = [
    ConversationExample ( conversation = [
        "Tired..." ,
        "Let's have a break!" ,
        "Nice idea!"
    ]),
]

# Fine-tune model
model . fit ( train_iterator = train_iterator , valid_iterator = valid_iterator )

# Generate response
model . generate ( context = [ "Hello" , "How are you" ], do_sample = True , top_p = 0.95 , top_k = 50 )
# Output could be like below if sufficient examples were given.
# => ConversationModelOutput(responses=['Good thank you'], context=['Hello', 'How are you'])

有关安装、模型架构和使用的更多详细信息，请参阅文档。

• 安装
• 模型架构概述
• 模型训练
• 响应生成
• CLI（实验）

享受与对话式人工智能交谈的乐趣