Gradio的web界面演示与交互机器学习模型，全局状态与会话状态《4》

最新推荐文章于 2024-04-27 15:16:44 发布

寅恪光潜

最新推荐文章于 2024-04-27 15:16:44 发布

阅读量1.7k

点赞数 4

分类专栏： Python 文章标签： gradio状态state transformers DialoGPT-medium GPT2

本文链接：https://blog.csdn.net/weixin_41896770/article/details/130739147

版权

Python 专栏收录该内容

121 篇文章 33 订阅

订阅专栏

全局状态和会话状态，对于程序员来说都是很熟悉的了，开发中会经常遇到，这里看下在Gradio中是怎么使用的，以及对GPT2的一点介绍

一、Global State全局状态

如果定义的函数想要访问外部的数据，可以将变量写在外面成为一个全局变量，这样就可以在函数内部访问它。例如，您可以在函数外部加载一个大型模型，并在函数内部使用它，这样每次函数调用都不需要重新加载模型。
看一个示例，显示排名前三的分数：

import gradio as gr

scores = [] #全局变量

def track_score(score):
    scores.append(score)
    top_scores = sorted(scores, reverse=True)[:3]
    return top_scores

demo = gr.Interface(
    track_score, 
    gr.Number(label="分数"), 
    gr.JSON(label="排名前三的分数")
)
demo.launch()

其中的scores = [] 分数数组会在所有用户之间共享。如果多个用户访问此演示，他们的分数将被添加到同一列表中，并且返回的前三名分数将在这个共享参考中收集。

二、Session State会话状态

数据持久化的另一种类型是会话状态，其中数据在页面会话中的多个提交中持久化。但是，数据不会在模型的不同用户之间共享。要在会话状态中存储数据，需要做三件事：

1、向函数传递一个额外的参数，该参数表示接口的状态。
2、在函数结束时，将状态的更新值作为额外的返回值返回。
3、在创建接口时添加'state'输入和'state'输出组件

聊天机器人就是一个需要会话状态的例子——您希望访问用户以前提交的内容，但是您不能将聊天历史存储在全局变量中，因为这样聊天历史会在不同的用户之间混淆。这个大家也是很熟悉的，做登录的时候就属于这种，不同用户需要区别开来。
来看个简单的机器人聊天的示例(来自微软的DialoGPT)，对官方例子做了点修改：

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

tokenizer = AutoTokenizer.from_pretrained("microsoft/DialoGPT-medium")
model = AutoModelForCausalLM.from_pretrained("microsoft/DialoGPT-medium")

def predict(input, history=[]):
    new_user_input_ids = tokenizer.encode(input + tokenizer.eos_token, return_tensors='pt')
 
    # 添加到聊天历史记录
    bot_input_ids = torch.cat([torch.LongTensor(history), new_user_input_ids], dim=-1)
 
    # 生成回答
    history = model.generate(bot_input_ids, max_length=1000, pad_token_id=tokenizer.eos_token_id).tolist()
 
    # 将标记转换为文本，然后将响应分成几行
    response = tokenizer.decode(history[0]).split("<|endoftext|>")
    response = [(response[i], response[i+1]) for i in range(0, len(response)-1, 2)]
    return response, history

gr.Interface(fn=predict,
             inputs=[gr.Textbox(label="输入"),"state"],
             outputs=[gr.Textbox(label="聊天记录列表",lines=8),"state"]).launch()

如果没有安装transformers模块将会找不到模块的错误，本人依然推荐大家带镜像的安装

ModuleNotFoundError: No module named 'transformers'

pip install transformers -i http://pypi.douban.com/simple/  --trusted-host pypi.douban.com

整体的调用还是比较简单的，就是先将输入做词嵌入(编码)，通过GPT2模型的处理生成回答，然后将词嵌入做解码输出。

然后我们输入信息，就能和机器人进行聊天了，界面如下所示：

这个DialoGPT是基于GPT2的一个初级聊天，首先会将相关的配置和模型给下载下来。然后就是标准的输入和输出，左边输入想要聊天的内容，右边就是跟机器人聊天的历史聊天记录。

三、GPT2的相关

这里我们来了解下这个DialoGPT-medium模型的分词的相关配置等信息，属于transformers.models.gpt2.tokenization_gpt2_fast.GPT2TokenizerFast

打印看下它的信息print(tokenizer)：

GPT2TokenizerFast(name_or_path='microsoft/DialoGPT-medium', vocab_size=50257, model_max_length=1024, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'bos_token': AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True), 'eos_token': AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True), 'unk_token': AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True)}, clean_up_tokenization_spaces=True)

看下这个模型的结构print(model)：

GPT2LMHeadModel(
  (transformer): GPT2Model(
    (wte): Embedding(50257, 1024)
    (wpe): Embedding(1024, 1024)
    (drop): Dropout(p=0.1, inplace=False)
    (h): ModuleList(
      (0): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (1): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (2): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (3): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (4): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (5): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (6): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (7): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (8): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (9): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (10): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (11): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (12): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (13): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (14): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (15): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (16): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (17): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (18): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (19): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (20): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (21): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (22): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
      (23): GPT2Block(
        (ln_1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2Attention(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
    )
    (ln_f): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)
  )
  (lm_head): Linear(in_features=1024, out_features=50257, bias=False)
)

有兴趣的可以查阅其余章节：

Gradio的web界面演示与交互机器学习模型，安装和使用《1》

Gradio的web界面演示与交互机器学习模型，主要特征《2》

Gradio的web界面演示与交互机器学习模型，分享应用《3》