简介

AI模型部署将训练好的模型转化为可调用的服务，涵盖API服务化、容器化部署、负载均衡和模型量化等技术。

FastAPI部署模型

# FastAPI部署模型服务
from fastapi import FastAPI
from pydantic import BaseModel
import joblib
import numpy as np

app = FastAPI(title="AI Programming Classifier")
model = joblib.load("model.pkl")

class PredictionRequest(BaseModel):
    features: list[float]

class PredictionResponse(BaseModel):
    prediction: int
    confidence: float

@app.post("/predict", response_model=PredictionResponse)
async def predict(req: PredictionRequest):
    features = np.array(req.features).reshape(1, -1)
    prediction = model.predict(features)[0]
    proba = model.predict_proba(features)[0]
    return PredictionResponse(
        prediction=int(prediction),
        confidence=float(max(proba))
    )

@app.get("/health")
async def health():
    return {"status": "healthy"}

Docker容器化

# Dockerfile
FROM python:3.11-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY . .
EXPOSE 8000
CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]

# docker-compose.yml
version: "3.8"
services:
  model-api:
    build: .
    ports: ["8000:8000"]
    environment:
      - MODEL_PATH=/app/model.pkl
    deploy:
      resources:
        limits: {memory: 2G, cpus: "2"}
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
      interval: 30s
      retries: 3

模型量化

# 模型量化减小体积和加速推理
import torch
from transformers import AutoModelForSequenceClassification

# 加载原始模型
model = AutoModelForSequenceClassification.from_pretrained("model_dir")

# 动态量化
quantized_model = torch.quantization.quantize_dynamic(
    model, {torch.nn.Linear}, dtype=torch.qint8
)

# 比较大小
original_size = sum(p.nelement() * p.element_size() for p in model.parameters())
quantized_size = sum(p.nelement() * p.element_size() for p in quantized_model.parameters())
print(f"原始: {original_size/1024/1024:.1f}MB")
print(f"量化: {quantized_size/1024/1024:.1f}MB")
print(f"压缩率: {quantized_size/original_size:.1%}")

# 保存量化模型
quantized_model.save_pretrained("model_quantized")

负载均衡

# Nginx负载均衡配置
upstream model_backend {
    least_conn;
    server model-api-1:8000 weight=3;
    server model-api-2:8000 weight=2;
    server model-api-3:8000 weight=1;
}

server {
    listen 80;
    location /predict {
        proxy_pass http://model_backend;
        proxy_set_header Host $host;
        proxy_read_timeout 30s;
    }
    location /health {
        access_log off;
        proxy_pass http://model_backend;
    }
}

小结

AI模型部署的核心是服务化和容器化。FastAPI适合中小规模部署，大规模场景需配合负载均衡和自动扩缩容。模型量化能显著减小体积和加速推理，但需验证精度损失是否在可接受范围内。

家庭AI编程入门