Create handler.py
Browse files- handler.py +38 -0
handler.py
ADDED
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Dict, List, Any
|
2 |
+
from PIL import Image
|
3 |
+
import torch
|
4 |
+
import base64
|
5 |
+
from io import BytesIO
|
6 |
+
from transformers import CLIPProcessor, CLIPModel
|
7 |
+
|
8 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
9 |
+
|
10 |
+
class EndpointHandler():
|
11 |
+
def __init__(self, path=""):
|
12 |
+
self.model = CLIPModel.from_pretrained("laion/CLIP-ViT-L-14-laion2B-s32B-b82K").to(device)
|
13 |
+
self.processor = CLIPProcessor.from_pretrained("laion/CLIP-ViT-L-14-laion2B-s32B-b82K")
|
14 |
+
|
15 |
+
def __call__(self, data: Any) -> List[float]:
|
16 |
+
inputs = data.pop("inputs", data)
|
17 |
+
|
18 |
+
if "image" in inputs:
|
19 |
+
# decode base64 image to PIL
|
20 |
+
image = Image.open(BytesIO(base64.b64decode(inputs['image'])))
|
21 |
+
inputs = self.processor(images=image, text=None, return_tensors="pt", padding=True).to(device)
|
22 |
+
|
23 |
+
image_embeds = self.model.get_image_features(
|
24 |
+
pixel_values=inputs["pixel_values"]
|
25 |
+
)
|
26 |
+
|
27 |
+
return image_embeds[0].tolist()
|
28 |
+
if "text" in inputs:
|
29 |
+
text = inputs['text']
|
30 |
+
inputs = self.processor(images=None, text=text, return_tensors="pt", padding=True).to(device)
|
31 |
+
|
32 |
+
text_embeds = self.model.get_text_features(
|
33 |
+
input_ids=inputs["input_ids"], attention_mask=inputs["attention_mask"]
|
34 |
+
)
|
35 |
+
|
36 |
+
return text_embeds[0].tolist()
|
37 |
+
|
38 |
+
raise Exception("No 'image' or 'text' provided")
|