gpu

app.py

@@ -5,15 +5,13 @@ from deepseek_vl.models import VLChatProcessor, MultiModalityCausalLM
 from deepseek_vl.utils.io import load_pil_images
 from io import BytesIO
 from PIL import Image
-import spaces  # Import spaces for ZeroGPU support
 
 # Load the model and processor
 model_path = "deepseek-ai/deepseek-vl-1.3b-chat"
 vl_chat_processor = VLChatProcessor.from_pretrained(model_path)
 tokenizer = vl_chat_processor.tokenizer
 
-# Define the function for image description with ZeroGPU support
-@spaces.GPU  # Ensures GPU allocation for this function
+# Define the function for image description (CPU-only)
 def describe_image(image, user_question="Solve the problem in the image"):
     try:
         # Convert the PIL Image to a BytesIO object for compatibility
@@ -43,10 +41,10 @@ def describe_image(image, user_question="Solve the problem in the image"):
             conversations=conversation,
             images=pil_images,
             force_batchify=True
-        ).to('cuda')
+        ).to('cpu')  # Move inputs to CPU
 
         # Load and prepare the model
-        vl_gpt = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True).to(torch.bfloat16).cuda().eval()
+        vl_gpt = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True).to(torch.bfloat16).cpu().eval()  # Move model to CPU
 
         # Generate embeddings from the image input
         inputs_embeds = vl_gpt.prepare_inputs_embeds(**prepare_inputs)