Update app.py

app.py

@@ -4,18 +4,10 @@ from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
 from qwen_vl_utils import process_vision_info
 import torch
 from PIL import Image
-import subprocess
 from datetime import datetime
 import numpy as np
 import os
 
-
-# subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
-
-# models = {
-#     "Qwen/Qwen2-VL-7B-Instruct": AutoModelForCausalLM.from_pretrained("Qwen/Qwen2-VL-7B-Instruct", trust_remote_code=True, torch_dtype="auto", _attn_implementation="flash_attention_2").cuda().eval()
-
-# }
 def array_to_image_path(image_array):
     if image_array is None:
         raise ValueError("No image provided. Please upload an image before submitting.")
@@ -35,8 +27,7 @@ def array_to_image_path(image_array):
     return full_path
     
 models = {
-    "Qwen/Qwen2-VL-7B-Instruct": Qwen2VLForConditionalGeneration.from_pretrained("Qwen/Qwen2-VL-7B-Instruct", trust_remote_code=True, torch_dtype="auto").cuda().eval()
-
+    "Qwen/Qwen2-VL-7B-Instruct": Qwen2VLForConditionalGeneration.from_pretrained("Qwen/Qwen2-VL-7B-Instruct", trust_remote_code=True, torch_dtype="auto").eval()
 }
 
 processors = {
@@ -45,14 +36,10 @@ processors = {
 
 DESCRIPTION = "[Qwen2-VL-7B Demo](https://huggingface.co/Qwen/Qwen2-VL-7B-Instruct)"
 
-kwargs = {}
-kwargs['torch_dtype'] = torch.bfloat16
-
 user_prompt = '<|user|>\n'
 assistant_prompt = '<|assistant|>\n'
 prompt_suffix = "<|end|>\n"
 
-@spaces.GPU
 def run_example(image, text_input=None, model_id="Qwen/Qwen2-VL-7B-Instruct"):
     image_path = array_to_image_path(image)
     
@@ -87,7 +74,8 @@ def run_example(image, text_input=None, model_id="Qwen/Qwen2-VL-7B-Instruct"):
         padding=True,
         return_tensors="pt",
     )
-    inputs = inputs.to("cuda")
+    # Move inputs to CPU
+    inputs = inputs.to("cpu")
     
     # Inference: Generation of the output
     generated_ids = model.generate(**inputs, max_new_tokens=1024)
@@ -123,4 +111,4 @@ with gr.Blocks(css=css) as demo:
         submit_btn.click(run_example, [input_img, text_input, model_selector], [output_text])
 
 demo.queue(api_open=False)
-demo.launch(debug=True)
+demo.launch(debug=True)