Update app.py to change torch data type to float16 and comment out attention implementation; modify requirements.txt to reorder dependencies and remove flash-attn.

app.py

@@ -8,14 +8,14 @@ import spaces
 
 # Initialize the model and processor
 device = "cuda" if torch.cuda.is_available() else "cpu"
-torch_dtype = torch.bfloat16 if torch.cuda.is_available() else torch.float32
+torch_dtype = torch.bfloat16 if torch.cuda.is_available() else torch.float16
 
 model = Qwen2_5OmniModel.from_pretrained(
     "Qwen/Qwen2.5-Omni-7B",
     torch_dtype=torch_dtype,
     device_map="auto",
     enable_audio_output=True,
-    attn_implementation="flash_attention_2" if torch.cuda.is_available() else None
+    # attn_implementation="flash_attention_2" if torch.cuda.is_available() else None
 )
 
 processor = Qwen2_5OmniProcessor.from_pretrained("Qwen/Qwen2.5-Omni-7B")

requirements.txt

@@ -2,5 +2,4 @@ transformers @ git+https://github.com/huggingface/transformers@3a1ead0aabed473ea
 qwen-omni-utils[decord]
 soundfile
 torch
-gradio
-flash-attn
+gradio