caption.py

from transformers import AutoProcessor, AutoModelForCausalLM
import torch
from PIL import Image

# Load model and processor
model = AutoModelForCausalLM.from_pretrained("Chesscorner/git-chess-v3")
processor = AutoProcessor.from_pretrained("Chesscorner/git-chess-v3")

# Set up device and move model to it
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# Enable mixed precision if on GPU
use_fp16 = device.type == "cuda"
if use_fp16:
    model.half()

max_length = 50
num_beams = 4
gen_kwargs = {'max_length': max_length, 'num_beams': num_beams}


# Prediction function
def predict_step(image):
    # Preprocess the image
    pixel_values = processor(images=image, return_tensors="pt").pixel_values.to(device)

    # Generate predictions with no_grad for efficiency
    with torch.no_grad():
        output_ids = model.generate(pixel_values=pixel_values, **gen_kwargs)

    # Decode predictions
    preds = processor.batch_decode(output_ids, skip_special_tokens=True)
    return preds[0].strip()