AI开发平台MODELARTS-Open-Clip基于Lite Server适配PyTorch NPU训练指导:Step7 推理验证

时间：2025-02-25 19:54:35

AI开发平台MODELARTS

Step7 推理验证

首先将上面训练的最终模型文件epoch_xx.pt（xx取决于训练的epoch个数）复制到/home/ma-user/open_clip目录下，然后在/home/ma-user/open_clip下，执行如下命令。

vi inference.py

将下面的代码复制进去后保存。代码中的epoch_29.pt请替换成实际值。

import os
import torch
from PIL import Image
import open_clip

if 'DEVICE_ID' in os.environ:
   print("DEVICE_ID:", os.environ['DEVICE_ID'])
else:
   os.environ['DEVICE_ID'] = "0"

model, _, preprocess = open_clip.create_model_and_transforms('ViT-B-32', pretrained='/home/ma-user/open_clip/epoch_29.pt')
model = model.to("npu")
tokenizer = open_clip.get_tokenizer('ViT-B-32')

image = preprocess(Image.open("./docs/CLIP.png")).unsqueeze(0)
text = tokenizer(["a diagram", "a dog", "a cat"])

print("input image shape:", image.shape)
print("input text shape:", text.shape)

with torch.no_grad(), torch.cuda.amp.autocast():
    image = image.to("npu")
    text = text.to("npu")
    image_features = model.encode_image(image)
    text_features = model.encode_text(text)

    print("output image shape:", image_features.shape)
    print("output text shape:", text_features.shape)

    image_features /= image_features.norm(dim=-1, keepdim=True)
    text_features /= text_features.norm(dim=-1, keepdim=True)

    text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1)

print("Label probs:", text_probs)  # prints: [[1., 0., 0.]]

运行推理脚本。