codezakh

outputs = blip2.generate(
    pixel_values=pixel_values,
    input_ids=input_ids,
    attention_mask=attention_mask,
    do_sample=False,
    num_beams=5,
    max_new_tokens=10,
    min_length=1,
    length_penalty=-1,
    return_dict_in_generate=True,

codezakh

class CustomBatchSampler(Sampler):
    def __init__(self, batch_size, dataset):
        self.sampler = SequentialSampler(dataset)
        self.batch_size = batch_size
        self.drop_last = False
        self.dataset = dataset
    def __iter__(self):
        batch = []
        for idx in self.sampler:
            batch.append(idx)

codezakh

import pandas as pd
import numpy as np

val_pairs = pd.read_csv('val_pairs.csv').set_index('index')

# Randomly predict {0, 1} for each pair.
random_predictions = np.random.randint(0,2, size=len(val_pairs))

# Add the predictions as a column named "label".
val_pairs['label'] = pd.Series(random_predictions)

codezakh

{
  "language": "en",
  "textAngle": 0.0,
  "orientation": "Up",
  "regions": [
    {
      "boundingBox": "22,25,1389,1863",
      "lines": [
        {
          "boundingBox": "22,25,1389,49",