|
from typing import Dict, List, Any |
|
import torch |
|
from transformers import AutoProcessor, Pix2StructVisionModel |
|
from PIL import Image |
|
import pdb |
|
import requests |
|
|
|
class EndpointHandler(): |
|
def __init__(self, path=""): |
|
|
|
|
|
self.processor = AutoProcessor.from_pretrained("google/pix2struct-textcaps-base") |
|
self.model = Pix2StructVisionModel.from_pretrained("google/pix2struct-textcaps-base") |
|
|
|
def __call__(self, data: Any) -> List[List[Dict[str, float]]]: |
|
url = data.pop("inputs", data) |
|
image = Image.open(requests.get(url, stream=True).raw) |
|
inputs = self.processor(images=image, return_tensors="pt") |
|
|
|
with torch.no_grad(): |
|
outputs = self.model(**inputs) |
|
|
|
last_hidden_state = outputs['last_hidden_state'] |
|
embedding = torch.mean(last_hidden_state, dim=1).flatten().tolist() |
|
return {"embedding": embedding} |
|
|
|
handler = EndpointHandler() |
|
output = handler({"inputs": "https://figma-staging-api.s3.us-west-2.amazonaws.com/images/a8c6a0cc-c022-4f3a-9fc5-ac8582c964dd"}) |
|
print(output) |
|
|
