revert

app.py

@@ -5,26 +5,17 @@ import torch
 import requests
 
 # ----------------------------
-# MODEL LOADING (MedVLM-R1) - CPU Compatible
+# MODEL LOADING (MedVLM-R1)
 # ----------------------------
 MODEL_PATH = 'JZPeterPan/MedVLM-R1'
 
-# Check if CUDA is available, otherwise use CPU
-device = "cuda" if torch.cuda.is_available() else "cpu"
-print(f"Using device: {device}")
-
 model = Qwen2VLForConditionalGeneration.from_pretrained(
     MODEL_PATH,
-    torch_dtype=torch.float32 if device == "cpu" else torch.bfloat16,
-    device_map="auto" if device == "cuda" else None,
-    low_cpu_mem_usage=True if device == "cpu" else False,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
 )
 processor = AutoProcessor.from_pretrained(MODEL_PATH)
 
-# Move model to device
-if device == "cpu":
-    model = model.to(device)
-
 temp_generation_config = GenerationConfig(
     max_new_tokens=1024,
     do_sample=False,
@@ -80,7 +71,7 @@ def process_pipeline(image, user_question):
         videos=video_inputs,
         padding=True,
         return_tensors="pt",
-    ).to(device)
+    ).to("cuda")
 
     # Generate output from MedVLM
     generated_ids = model.generate(
@@ -162,4 +153,4 @@ with gr.Blocks(title="Brain MRI QA") as demo:
     )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()

requirements.txt

@@ -1,15 +1,13 @@
 gradio==5.42.0
-transformers==4.36.0
---find-links https://download.pytorch.org/whl/torch_stable.html
-torch==2.1.0+cpu
-torchvision==0.16.0+cpu
-torchaudio==2.1.0+cpu
+transformers>=4.40.0
+torch>=2.0.0
+torchvision>=0.15.0
 requests>=2.31.0
 Pillow>=10.0.0
 accelerate>=0.20.0
 safetensors>=0.3.0
 tokenizers>=0.15.0
-numpy<2.0.0
+numpy>=1.24.0
 scipy>=1.10.0
 qwen-vl-utils
 ipython>=8.0.0