Upload 2 files

app.py

@@ -23,7 +23,7 @@ def _lazy_load() -> Tuple[LlavaOnevisionForConditionalGeneration, AutoProcessor]
         _processor = AutoProcessor.from_pretrained(MODEL_ID, trust_remote_code=True)
         _model = LlavaOnevisionForConditionalGeneration.from_pretrained(
             MODEL_ID,
-            torch_dtype=DTYPE,
+            dtype=DTYPE,
             low_cpu_mem_usage=True,
             trust_remote_code=True,
             use_safetensors=True,
@@ -58,6 +58,9 @@ def _compose_prompt(user_text: str, context: Optional[Dict] = None) -> List[Dict
 @spaces.GPU  # en HF Spaces usará GPU cuando haya disponibilidad (ZeroGPU)
 def _infer_one(image: Image.Image, text: str, max_new_tokens: int = 256, temperature: float = 0.7,
                context: Optional[Dict] = None) -> str:
+    # Reducir el tamaño de la imagen para ahorrar memoria en la GPU
+    image.thumbnail((1024, 1024))
+
     model, processor = _lazy_load()
     prompt = processor.apply_chat_template(_compose_prompt(text, context), add_generation_prompt=True)
     inputs = processor(images=image, text=prompt, return_tensors="pt").to(DEVICE, dtype=DTYPE)