updated reqquirements for flash

app.py

@@ -7,38 +7,29 @@ import os
 import random
 import re
 import subprocess
-import torch._dynamo
-
 
 # Define the model name
-
 OUTPUT_DIR = "output"
 os.makedirs(OUTPUT_DIR, exist_ok=True)
 
 model_name = "TheBloke/Amethyst-13B-Mistral-AWQ"
-attn_mode = "flash_attention_2" if is_flash_attn_2_available() else "default"
+
 # Load the tokenizer
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 
-def install_flash_attn():
-    try:
-        import flash_attn
-        print("✅ Flash Attention 2 is already installed.")
-    except ImportError:
-        print("🚀 Installing Flash Attention 2...")
-        subprocess.run(["pip", "install", "flash-attn", "--no-build-isolation"], check=True)
-
-install_flash_attn()
-
-# Load the model
+# Load the model (without Flash Attention)
 model = AutoModelForCausalLM.from_pretrained(
     model_name,
-    torch_dtype=torch.float16,
-    device_map="auto",
-    attn_implementation=attn_mode  # Use Flash Attention 2 if available
+    torch_dtype=torch.float16,  # Use float16 for performance
+    device_map="auto"
 )
 
-model = torch.compile(model, mode="max-autotune")  # Optimize inference performance
+# Use torch.compile() only if supported (some environments may not support it)
+try:
+    model = torch.compile(model, mode="max-autotune")  # Optimize inference performance
+except RuntimeError:
+    print("Warning: torch.compile() is not supported on this system. Skipping optimization.")
+
 
 # Define the base prompt
 base_prompt = """