# ollama部署
curl -fsSL https://ollama.com/install.sh | sh

ollama run mistral-small

ollama run mistral-small:24b-instruct-2501-q4_K_M

ollama run mistral-small:24b-instruct-2501-q8_0

ollama run mistral-small:24b-instruct-2501-fp16

# vLLM部署
conda create -n mistral_env python=3.10
conda activate mistral_env

pip install --upgrade torchvision vllm xformers

pip install --upgrade vllm
pip install --upgrade mistral_common

huggingface-cli login

vllm serve mistralai/Mistral-Small-24B-Instruct-2501 --tokenizer_mode mistral --config_format mistral --load_format mistral --enable-auto-tool-choice --tool-call-parser mistral

!pip install mistralai gradio requests

import os
from mistralai import Mistral

from google.colab import userdata

api_key = userdata.get('MISTRAL_API_KEY')
model = "mistral-small-latest"

client = Mistral(api_key=api_key)

chat_response = client.chat.complete(
    model = model,
    messages = [
        {
            "role": "user",
            "content": "写一篇200字的科幻小说?",
        },
    ]
)

print(chat_response.choices[0].message.content)

import os
import json
import requests
import gradio as gr
from google.colab import userdata

# 配置API
API_KEY = userdata.get('FIREWORKS_API_KEY')
API_URL = "https://api.fireworks.ai/inference/v1/chat/completions"

def generate_response(message, history):
    """
    使用Fireworks AI API生成响应
    """
    try:
        # 准备请求数据
        payload = {
            "model": "accounts/fireworks/models/mistral-small-24b-instruct-2501",
            "max_tokens": 4096,
            "top_p": 1,
            "top_k": 40,
            "presence_penalty": 0,
            "frequency_penalty": 0,
            "temperature": 0.6,
            "messages": [
                {
                    "role": "user",
                    "content": message
                }
            ]
        }
        
        headers = {
            "Accept": "application/json",
            "Content-Type": "application/json",
            "Authorization": f"Bearer {API_KEY}"
        }

        # 发送请求
        response = requests.post(
            API_URL,
            headers=headers,
            data=json.dumps(payload)
        )
        
        # 检查响应状态
        response.raise_for_status()
        
        # 解析响应
        result = response.json()
        return result['choices'][0]['message']['content']
    except Exception as e:
        return f"发生错误: {str(e)}"

# 自定义CSS样式
custom_css = """
#component-0 {
    position: fixed;
    top: 0;
    left: 0;
    right: 0;
    bottom: 0;
    display: flex;
    flex-direction: column;
}

#chatbot {
    flex-grow: 1;
    overflow-y: auto;
    margin: 20px;
    padding: 20px;
}

#input-container {
    position: fixed;
    bottom: 0;
    left: 0;
    right: 0;
    padding: 20px;
    background-color: white;
    box-shadow: 0 -2px 5px rgba(0,0,0,0.1);
}

.message {
    padding: 15px;
    margin: 10px 0;
    border-radius: 10px;
}

.user-message {
    background-color: #f0f0f0;
}

.bot-message {
    background-color: #e6f3ff;
}
"""

# 创建Gradio界面
with gr.Blocks(css=custom_css) as demo:
    gr.Markdown("# Fireworks AI 聊天界面")
    
    chatbot = gr.Chatbot(
        elem_id="chatbot",
        bubble_full_width=False,
        show_label=False,
        height=600
    )
    
    with gr.Row(elem_id="input-container"):
        with gr.Column(scale=20):
            input_text = gr.Textbox(
                placeholder="输入消息...",
                label="输入",
                lines=1,
                max_lines=5,
                show_label=False
            )
        with gr.Column(scale=1, min_width=50):
            submit_btn = gr.Button("发送", variant="primary")

    # 清空对话按钮
    clear_btn = gr.Button("清空对话", variant="secondary")
    
    # 事件处理
    def user_input(user_message, history):
        return "", history + [[user_message, None]]
    
    def bot_response(history):
        user_message = history[-1][0]
        bot_message = generate_response(user_message, history)
        history[-1][1] = bot_message
        return history
    
    def clear_history():
        return []
    
    input_text.submit(
        user_input,
        [input_text, chatbot],
        [input_text, chatbot],
        queue=False
    ).then(
        bot_response,
        chatbot,
        chatbot
    )
    
    submit_btn.click(
        user_input,
        [input_text, chatbot],
        [input_text, chatbot],
        queue=False
    ).then(
        bot_response,
        chatbot,
        chatbot
    )
    
    clear_btn.click(clear_history, None, chatbot)

# 启动界面
demo.launch(
    share=True,
    height=800,
)