sudo apt update && sudo apt upgrade -y
sudo apt install -y git python3-pip python3-dev python3-venv build-essential cmake libopenblas-dev libsodium-dev

mkdir ~/mistral_ai && cd ~/mistral_ai
python3 -m venv venv
source venv/bin/activate

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

python -c "import torch; print(torch.__version__); print(torch.cuda.is_available())"

pip install transformers accelerate bitsandbytes sentencepiece protobuf scipy safetensors ninja flash-attn --no-cache-dir

git lfs install
git clone https://huggingface.co/mistralai/Mistral-7B-v0.1 model_weights/
cd model_weights && git lfs pull && cd ..

from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

# 加载模型和分词器（首次运行会较慢）
model_path = "./model_weights"
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    device_map="auto",   # 自动选择设备(GPU/CPU)
    torch_dtype="auto",   # 自动选择精度(FP16/FP32)
    trust_remote_code=True,
)

# 创建文本生成管道 
pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
)

# 示例生成文本 
prompt = "Explain the concept of machine learning in simple terms:"
outputs = pipe(
    prompt,
    max_new_tokens=256,
    do_sample=True,
    temperature=0.7,
    top_k=50,
    top_p=0.95, 
)

print(outputs[0]["generated_text"])

python test_mistral.py

Explain the concept of machine learning in simple terms:
Machine learning is a type of artificial intelligence that allows computers to learn from data without being explicitly programmed. Imagine teaching a child by showing examples instead of giving strict rules - that's essentially what machine learning does...

model = AutoModelForCausalLM.from_pretrained(
    model_path,
    device_map="cpu",
    torch_dtype=torch.float32,   # CPU上建议使用FP32精度 
)

#!/bin/bash 

source ~/mistral_ai/venv/bin/activate 

MODEL_PATH="./model_weights"

if [ "$1" == "--interactive" ]; then 
    python -c "
from transformers import pipeline; import sys; 

pipe = pipeline('text-generation', model='$MODEL_PATH', device_map='auto') 

print('Enter your prompt (Ctrl+D to exit):') 

for line in sys.stdin: print(pipe(line.strip(), max_new_tokens=256)[0]['generated_text'])"
else  
    python test_mistral.py "$@"
fi  

chmod +x mistral.sh 

#交互模式使用: ./mistral.sh --interactive  
#批量模式使用: ./mistral.sh "你的提示词"  

FROM nvidia/cuda:12.1-base 

RUN apt update && apt install -y python3-pip git && rm -rf /var/lib/apt/lists/* 

WORKDIR /app 

COPY requirements.txt ./
RUN pip install --no-cache-dir -r requirements.txt 

COPY . .  

CMD ["python", "test_mistral.py"]  

docker build -t mistral .
docker run --gpus all mistral  

pip install gradio  

import gradio as gr  
from transformers import pipeline  

def generate_text(prompt):  
    pipe = pipeline("text-generation", model="./model_weights")  
    return pipe(prompt, max_new_tokens=256)[0]["generated_text"]  

iface = gr.Interface(  
    fn=generate_text,  
    inputs="textbox",  
    outputs="textbox",  
)  

iface.launch(server_name="0.0.0.0")   

docker run --gpus all \   
           -p8080:80 \   
           -v $PWD/model_weights:/data \   
           ghcr.io/huggingface/text-generation-inference \   
           --model-id /data \   
           --quantize bitsandbytes   

curl http://localhost:8080/generate \    
     -X POST \    
     -d '{"inputs":"What is AI?","parameters":{"max_new_tokens":20}}' \    
     -H "Content-Type: application/json"    

from datasets import load_dataset    
from transformers import (AutoModelForCausalLM,     
                         AutoTokenizer,     
                         TrainingArguments,     
                         Trainer)    

#加载预训练模型和分词器    
model = AutoModelForCausalLM.from_pretrained("./model_weights")    
tokenizer = AutoTokenizer.from_pretrained("./model_weights")    

#准备数据集(示例)    
dataset = load_dataset("your_dataset")    

def tokenize_function(examples):    
    return tokenizer(examples["text"], truncation=True)    

tokenized_datasets = dataset.map(tokenize_function, batched=True)    

#配置训练参数    
training_args = TrainingArguments(    
    output_dir="./results",    
    per_device_train_batch_size=4,    
)    

trainer = Trainer(    
    model=model,    
    args=training_args,    
    train_dataset=tokenized_datasets["train"],    
)    

trainer.train()     
trainer.save_model("./fine_tuned_model")     

accelerate launch train.py     

import timeit   

def benchmark():     
start_time = timeit.default_timer()     
outputs = pipe("Benchmark test", max_new_tokens=100)     
duration = timeit.default_timer() - start_time     
print(f"Tokens/sec: {len(outputs[0]['generated_text'].split()) / duration}")       

benchmark()      

docker run --read-only --cap-drop ALL ...      

systemd-run --scope -p MemoryLimit=16G ./mistral.sh