from fastapi import FastAPI, File, UploadFile
from fastapi.responses import JSONResponse
from transformers import AutoTokenizer, AutoProcessor, AutoModelForCausalLM
import torch
from PIL import Image
import io

app = FastAPI()

model_path = "lmms-lab/LLaVA-OneVision-1.5-8B-Instruct"
model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype="auto", device_map="auto", trust_remote_code=True)
processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)

def process_vision_info(messages):
    # Dummy implementation, replace with actual from qwen_vl_utils
    image_inputs = [msg['content'][0]['image'] for msg in messages]
    video_inputs = None
    return image_inputs, video_inputs

@app.post("/analyze-image")
async def analyze_image(file: UploadFile = File(...)):
    image_bytes = await file.read()
    image = Image.open(io.BytesIO(image_bytes))
    messages = [
        {
            "role": "user",
            "content": [
                {"type": "image", "image": image},
                {"type": "text", "text": "Describe this image."},
            ],
        }
    ]
    text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    image_inputs, video_inputs = process_vision_info(messages)
    inputs = processor(
        text=[text],
        images=image_inputs,
        videos=video_inputs,
        padding=True,
        return_tensors="pt",
    )
    inputs = inputs.to("cuda")
    generated_ids = model.generate(**inputs, max_new_tokens=1024)
    generated_ids_trimmed = [
        out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
    ]
    output_text = processor.batch_decode(
        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
    )
    return JSONResponse(content={"result": output_text})