Update app.py

app.py

@@ -95,6 +95,12 @@ llama_model = AutoModelForCausalLM.from_pretrained(
 )
 llama_model.eval()
 
+# --- IMPORTANT: Set Pad Token ---
+# Llama3 does *not* have a default pad token.  We *must* set one.
+# Using the EOS token as the PAD token is a common and recommended practice.
+if llama_tokenizer.pad_token is None:
+    llama_tokenizer.pad_token = llama_tokenizer.eos_token
+
 
 @spaces.GPU(duration=90)
 def llama_generate(