Inferencia.py

from transformers import pipeline
import torch
class Inferencia:

    def __init__(self):
        ...

    def extrair_dados_imagem(self, prompt, imagem):
        pipe = pipeline(
            "image-text-to-text",
            model="google/gemma-3-4b-it",
            torch_dtype=torch.bfloat16  # Otimiza para rodar na CPU
        )
        messages = [
            {
                "role": "system",
                "content": [{"type": "text", "text": "You are a helpful assistant."}]
            },
            {
                "role": "user",
                "content": [
                    {"type": "image", "url": imagem},
                    {"type": "text",
                     "text": prompt}
                ]
            }
        ]

        output = pipe(text=messages, max_new_tokens=200)

        return output[0]["generated_text"][-1]["content"]