| |
| import gradio as gr |
| import openai |
| import tiktoken |
|
|
| |
| openai.api_key = 'none' |
| openai.api_base="http://localhost:8080/v1" |
|
|
| def count_token(prompt,answer): |
| encoding = tiktoken.get_encoding("cl100k_base") |
| prompt_count = len(encoding.encode(prompt)) |
| answer_count = len(encoding.encode(answer)) |
| total_count = prompt_count + answer_count |
| print("Prompt消耗 %d Token, 回答消耗 %d Token，总共消耗 %d Token" % (prompt_count, answer_count, total_count)) |
|
|
| def concatenate_history(history): |
| text = "" |
| for item in history: |
| text += f"User: {item[0]}\nBot: {item[1]}\n" |
| return text |
|
|
| def summarize(text): |
| |
| if text is None: |
| return "" |
| else: |
| response = openai.ChatCompletion.create( |
| model="SoulChat", |
| messages=[{"role": "user", "content": text + "\n\n请总结一下User和Bot分别说了什么，并输出为markdown的格式\n"}], |
| temperature=0, |
| max_tokens=500 |
| ) |
| generated_text = response['choices'][0]['message']['content'] |
| count_token(text,generated_text) |
| print("总结回复：%s"%generated_text) |
| return generated_text |
|
|
| |
| def reply(prompt): |
| |
| response = openai.ChatCompletion.create( |
| model="SoulChat", |
| messages=[{"role": "user", "content": prompt}], |
| temperature=0, |
| max_tokens=4096 |
| ) |
| generated_text = response['choices'][0]['message']['content'] |
| count_token(prompt,generated_text) |
| print(generated_text) |
| return generated_text |
|
|
| |
| def send(user_message, history): |
| if not user_message: |
| return '', history |
| history_text = concatenate_history(history) |
| |
| prp=user_message |
| response = reply(prp) |
| return '', history + [[user_message, response]] |
|
|
| |
| with gr.Blocks() as demo: |
| chatbot = gr.Chatbot() |
| msg = gr.TextArea() |
| send_btn = gr.Button('发送') |
| send_btn.click(send, inputs=[msg,chatbot], outputs=[msg,chatbot], show_progress=True) |
|
|
| demo.launch() |
