metadata
license: mit
import re
import transformers
from PIL import Image
from transformers import DonutProcessor, VisionEncoderDecoderModel
import torch
import random
import numpy as np
fine_tuned_model = VisionEncoderDecoderModel.from_pretrained("aravind-selvam/donut_finetuned_chart")
processor = DonutProcessor.from_pretrained("aravind-selvam/donut_finetuned_chart")
# Move model to GPU
device = "cuda" if torch.cuda.is_available() else "cpu"
fine_tuned_model.to(device)
# Load random document image from the test set
dataset = load_dataset("hf-internal-testing/example-documents", split="test")
sample_image = dataset[1]
def run_prediction(sample, model=fine_tuned_model, processor=processor):
# pixel values
pixel_values = processor(image, return_tensors="pt").pixel_values
# prepare inputs
task_prompt = "<s>"
decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
# run inference
outputs = model.generate(
pixel_values.to(device),
decoder_input_ids=decoder_input_ids.to(device),
max_length=model.decoder.config.max_position_embeddings,
early_stopping=True,
pad_token_id=processor.tokenizer.pad_token_id,
eos_token_id=processor.tokenizer.eos_token_id,
use_cache=True,
num_beams=2,
# bad_words_ids=[[processor.tokenizer.unk_token_id]],
return_dict_in_generate=True,
)
# process output
prediction = processor.batch_decode(outputs.sequences)[0]
prediction = re.sub(r"<one>", "1", prediction)
prediction = processor.token2json(prediction)
# load reference target
target = processor.token2json(test_sample["target_sequence"])
return prediction, target
prediction, target = run_prediction(sample_image)
print(f"Reference:\n {target}")
print(f"Prediction:\n {prediction}")