Update app.py

app.py

@@ -16,24 +16,22 @@ MODEL_ID = "microsoft/phi-2"
 
 @st.cache_resource
 def load_model():
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
-    # Set pad_token_id for the tokenizer if it's not already set, using eos_token_id as a fallback
-    if tokenizer.pad_token_id is None:
-        tokenizer.pad_token_id = tokenizer.eos_token_id
-
-    # Load the model configuration
-    config = AutoConfig.from_pretrained(MODEL_ID)
+    tokenizer = AutoTokenizer.from_pretrained(
+        MODEL_ID,
+        trust_remote_code=True
+    )
 
-    # Check if pad_token_id exists in config and set it if not, for Phi models this often needs to be explicitly added
-    if not hasattr(config, 'pad_token_id') or config.pad_token_id is None:
-        config.pad_token_id = tokenizer.pad_token_id
+    if tokenizer.pad_token is None:
+        tokenizer.pad_token = tokenizer.eos_token
 
     model = AutoModelForCausalLM.from_pretrained(
         MODEL_ID,
-        config=config, # Pass the modified config to the model
-        torch_dtype=torch.float32,
-        device_map="auto"
+        trust_remote_code=True,
+        torch_dtype=torch.float32   # CPU safe
+        # ❌ removed device_map="auto"
     )
+
+    model.eval()
     return tokenizer, model
 
 tokenizer, model = load_model()