app.py

from fastapi import FastAPI, UploadFile, File
from transformers import AutoModelForCausalLM, AutoTokenizer
from PIL import Image
import torch
import io

app = FastAPI()

# Load model (optimized for CPU/Low RAM)
model_id = "THUDM/glm-4v-9b" # Or the specific 0.9B GLM-OCR variant
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = model_id.from_pretrained(model_id, trust_remote_code=True).cpu().eval()

@app.post("/convert")
async def convert_image(file: UploadFile = File(...)):
    image_data = await file.read()
    image = Image.open(io.BytesIO(image_data)).convert("RGB")
    
    # Specific prompt to trigger Chart-to-HTML
    prompt = "Read this chart and output the data as a clean HTML table with headers."
    
    inputs = tokenizer.apply_chat_template([{"role": "user", "image": image, "content": prompt}], 
                                          add_generation_prompt=True, tokenize=True, return_tensors="pt")
    
    with torch.no_grad():
        outputs = model.generate(**inputs, max_new_tokens=1000)
    
    response = tokenizer.decode(outputs[0])
    return {"html_result": response}