enginex-mr_series-vc-cnn/model_test_caltech_http_mr100.py

import torch
import time
import os
from PIL import Image
from transformers import AutoImageProcessor, AutoModelForImageClassification
from flask import Flask, request, jsonify
from io import BytesIO

# 设置CPU核心数（仅用于可能的底层优化，不影响GPU推理）
os.environ["OMP_NUM_THREADS"] = "4"
os.environ["MKL_NUM_THREADS"] = "4"
os.environ["NUMEXPR_NUM_THREADS"] = "4"
os.environ["OPENBLAS_NUM_THREADS"] = "4"
os.environ["VECLIB_MAXIMUM_THREADS"] = "4"
torch.set_num_threads(4)

# 设备配置 - 只关注GPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"当前设备: {device}")
print(f"CPU核心数设置: {torch.get_num_threads()}")

class ImageClassifier:
    def __init__(self, model_path: str):
        self.processor = AutoImageProcessor.from_pretrained(model_path)
        
        # 仅加载GPU模型
        if device.type == "cuda":
            self.model = AutoModelForImageClassification.from_pretrained(model_path).to(device)
        else:
            self.model = None  # 无GPU时模型为None
        
        # 保存id2label映射（从模型配置获取）
        if self.model:
            self.id2label = self.model.config.id2label
        else:
            self.id2label = None

    def _predict_with_model(self, image) -> dict:
        """使用GPU执行预测"""
        try:
            # 检查GPU模型是否可用
            if not self.model or device.type != "cuda":
                return {
                    "class_id": -1,
                    "class_name": "error",
                    "confidence": 0.0,
                    "device_used": str(device),
                    "processing_time": 0.0,
                    "error": "CUDA设备不可用或模型未加载"
                }
            
            # 记录开始时间
            start_time = time.perf_counter()
            
            # 处理图片并移动到GPU
            inputs = self.processor(images=image, return_tensors="pt").to(device)
            
            with torch.no_grad():
                ts = time.time()
                outputs = self.model(** inputs)
                print('mr100 T1', time.time() - ts, flush=True)
                
                ts = time.time()
                for i in range(1000):
                    outputs = self.model(**inputs)
                print('mr100 T2', time.time() - ts, flush=True)
                
            logits = outputs.logits
            probs = torch.nn.functional.softmax(logits, dim=1)
            max_prob, max_idx = probs.max(dim=1)
            class_idx = max_idx.item()
            
            # 计算处理时间
            processing_time = round(time.perf_counter() - start_time, 6)
            
            return {
                "class_id": class_idx,
                "class_name": self.id2label[class_idx],
                "confidence": float(max_prob.item()),
                "device_used": str(device),
                "processing_time": processing_time
            }
        except Exception as e:
            return {
                "class_id": -1,
                "class_name": "error",
                "confidence": 0.0,
                "device_used": str(device),
                "processing_time": 0.0,
                "error": str(e)
            }

    def predict_single_image(self, image) -> dict:
        """预测单张图片，仅使用GPU"""
        results = {"status": "success"}
        results["prediction"] = self._predict_with_model(image)
        return results

# 初始化服务
app = Flask(__name__)
MODEL_PATH = os.environ.get("MODEL_PATH", "/model")  # 模型路径（环境变量或默认路径）
classifier = ImageClassifier(MODEL_PATH)

@app.route('/v1/private/s782b4996', methods=['POST'])
def predict_single():
    """接收单张图片并返回GPU预测结果"""
    if 'image' not in request.files:
        return jsonify({
            "status": "error",
            "prediction": {
                "class_id": -1,
                "class_name": "error",
                "confidence": 0.0,
                "device_used": str(device),
                "processing_time": 0.0,
                "error": "请求中未包含图片"
            }
        }), 400
    
    image_file = request.files['image']
    try:
        image = Image.open(BytesIO(image_file.read())).convert("RGB")
        result = classifier.predict_single_image(image)
        return jsonify(result)
    except Exception as e:
        return jsonify({
            "status": "error",
            "prediction": {
                "class_id": -1,
                "class_name": "error",
                "confidence": 0.0,
                "device_used": str(device),
                "processing_time": 0.0,
                "error": str(e)
            }
        }), 500

@app.route('/health', methods=['GET'])
def health_check():
    return jsonify({
        "status": "healthy", 
        "cuda_available": device.type == "cuda",
        "device_used": str(device),
        "cpu_threads": torch.get_num_threads()
    }), 200

if __name__ == "__main__":
    app.run(host='0.0.0.0', port=80, debug=False)