updated model

app.py

@@ -1,25 +1,52 @@
 import gradio as gr
-from transformers import AutoTokenizer, AutoModelForCausalLM
+from transformers import AutoTokenizer, AutoModelForCausalLM, LlamaConfig
 import os
 from dotenv import load_dotenv
 import logging
 import sys  # Ensure sys is imported
-from huggingface_hub import login
+from huggingface_hub import login, HfApi
 
 # Load environment variables
 load_dotenv()
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s', stream=sys.stdout)
 
 # Authenticate with Hugging Face
-hf_token = os.environ.get("HUGGING_FACE_TOKEN")
-if not hf_token:
-    raise ValueError("HUGGING_FACE_TOKEN not found in environment variables")
-login(token=hf_token)
-
-# Load the Llama-3.1-8B model and tokenizer
+hf_token = os.environ.get("HUGGINGFACE_TOKEN")
 model_name = "meta-llama/Llama-3.1-8B"
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(model_name)
+fallback_model = "facebook/opt-350m"
+
+if hf_token:
+    try:
+        login(token=hf_token)
+        api = HfApi()
+        api.whoami()
+        logging.info("Successfully logged in to Hugging Face")
+    except Exception as e:
+        logging.error(f"Error authenticating with Hugging Face: {str(e)}")
+        logging.warning("Proceeding without authentication. Will use fallback model.")
+        model_name = fallback_model
+else:
+    logging.warning("HUGGINGFACE_TOKEN not found in environment variables. Proceeding without authentication.")
+    model_name = fallback_model
+
+# Load the model and tokenizer
+try:
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    
+    # Custom configuration to handle the RoPE scaling issue
+    if model_name == "meta-llama/Llama-3.1-8B":
+        config = LlamaConfig.from_pretrained(model_name)
+        config.rope_scaling = {"type": "linear", "factor": 8.0}  # Adjust as needed
+        model = AutoModelForCausalLM.from_pretrained(model_name, config=config)
+    else:
+        model = AutoModelForCausalLM.from_pretrained(model_name)
+    
+    logging.info(f"Successfully loaded {model_name}")
+except Exception as e:
+    logging.error(f"Error loading {model_name}: {str(e)}")
+    logging.info(f"Falling back to {fallback_model}")
+    tokenizer = AutoTokenizer.from_pretrained(fallback_model)
+    model = AutoModelForCausalLM.from_pretrained(fallback_model)
 
 # Function to generate a formatted email
 def generate_email(recipient_name, recipient_email, industry, recipient_role, details):