from fastapi import FastAPI
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
from qwen_vl_utils import process_vision_info
import torch

app = FastAPI()

model = Qwen2VLForConditionalGeneration.from_pretrained("Qwen/Qwen2-VL-2B-Instruct", torch_dtype="auto", device_map="auto")
processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct")

@app.post("/predict")
async def predict(messages: list):
    # Processamento e inferência
    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("cpu")  # Altere para "cuda" se tiver GPU disponível
    
    generated_ids = model.generate(**inputs, max_new_tokens=128)
    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 {"response": output_text}