Update app.py

app.py

@@ -1,11 +1,12 @@
 import subprocess 
 
+# Install required dependencies
 subprocess.run(
     'pip install flash-attn --no-build-isolation',
     env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"},
     shell=True
 )
-subprocess.run('pip install googletrans==4.0.0-rc1', shell=True)
+subprocess.run('pip install googletrans==4.0.0-rc1 httpx>=0.24.1 gradio>=5.9.1 gradio-client>=1.5.2', shell=True)
 
 import os
 import re
@@ -23,7 +24,7 @@ from transformers import (
 )
 
 # Configuration Constants
-MODEL_ID = "CohereForAI/aya-expanse-32b"
+MODEL_ID = "CohereForAI/aya-expanse-8b"
 DEFAULT_SYSTEM_PROMPT = """ 
 You are a bilingual translator specializing in Arabic and English translations. Your objective is to create precise, contextually appropriate, and stylistically consistent translations that adhere to the following guidelines:
 
@@ -146,7 +147,7 @@ def initialize_model():
         torch_dtype=torch.float16,
         device_map="cuda",
         attn_implementation="flash_attention_2",
-        quantization_config=quantization_config
+        #quantization_config=quantization_config
 
     )