File size: 1,468 Bytes
fdc844c
8df5e7a
3077814
c67db7a
3077814
fdc844c
 
 
3077814
fdc844c
3077814
 
 
 
 
 
 
 
 
 
3fac44f
 
 
 
 
 
3077814
70c64cd
 
3077814
 
 
 
 
70c64cd
8df5e7a
fdc844c
 
 
 
3077814
 
 
 
 
 
 
8df5e7a
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
import os
from typing import Dict, List, Any
from PIL import Image
import jax
from transformers import ViTFeatureExtractor, AutoTokenizer, FlaxVisionEncoderDecoderModel


class PreTrainedPipeline():

    def __init__(self, path=""):

        model_dir = os.path.join(path, "ckpt_epoch_3_step_6900")

        self.model = FlaxVisionEncoderDecoderModel.from_pretrained(model_dir)
        self.feature_extractor = ViTFeatureExtractor.from_pretrained(model_dir)
        self.tokenizer = AutoTokenizer.from_pretrained(model_dir)

        max_length = 16
        num_beams = 4
        self.gen_kwargs = {"max_length": max_length, "num_beams": num_beams}
        
        @jax.jit
        def _generate(pixel_values):

            output_ids = self.model.generate(pixel_values, **self.gen_kwargs).sequences
            return output_ids

        self.generate = _generate

        # compile the model
        image_path = os.path.join(path, 'val_000000039769.jpg')
        image = Image.open(image_path)
        self(image)
        image.close()
        
    def __call__(self, inputs: "Image.Image") -> List[str]:
        """
        Args:
        Return:
        """

        pixel_values = self.feature_extractor(images=inputs, return_tensors="np").pixel_values
        
        output_ids = self.generate(pixel_values)
        preds = self.tokenizer.batch_decode(output_ids, skip_special_tokens=True)
        preds = [pred.strip() for pred in preds]

        return preds