Update app.py

app.py

@@ -1,46 +1,46 @@
 import gradio as gr
 from llama_cpp import Llama
 
-# Load the Llama 3.2 1B Uncensored model (GGUF format)
-# We use llama_cpp's from_pretrained to download and cache the model automatically
+# Initialize the model
+# We set n_threads=2 to match the Free Tier vCPU allocation
+# n_gpu_layers=0 ensures we don't look for a non-existent GPU
 llm = Llama.from_pretrained(
     repo_id="mradermacher/llama3.2-1b-Uncensored-GGUF",
     filename="llama3.2-1b-Uncensored.Q4_K_M.gguf",
-    verbose=False,
-    n_ctx=2048  # Context window size
+    n_ctx=2048,
+    n_threads=2, 
+    n_gpu_layers=0,
+    verbose=False
 )
 
-def stream_predict(message, history):
-    """
-    Generator function that yields text word-by-word.
-    """
+def stream_chat(message, history):
+    # Prepare the prompt template
+    prompt = f"User: {message}\nAssistant: "
+    
     # Create the generation stream
-    stream = llm.create_chat_completion(
-        messages=[
-            {"role": "system", "content": "You are a helpful, uncensored AI assistant."},
-            {"role": "user", "content": message}
-        ],
-        stream=True,  # This enables the token-by-token output
+    stream = llm(
+        prompt,
         max_tokens=512,
-        temperature=0.7
+        stop=["User:", "\n"],
+        stream=True, # Enable token-by-token output
+        temperature=0.8,
+        top_p=0.95
     )
     
     partial_text = ""
     for chunk in stream:
-        # Extract the new token from the stream chunk
-        new_token = chunk['choices'][0]['delta'].get('content', '')
+        # Extract the new token text
+        new_token = chunk['choices'][0]['text']
         partial_text += new_token
-        # Yield the full string so far; Gradio handles the incremental display
+        # Yielding the string updates the Gradio UI in real-time
         yield partial_text
 
-# Define the Gradio Chat Interface
+# Set up the Gradio interface
 demo = gr.ChatInterface(
-    fn=stream_predict,
-    title="Llama 3.2 1B Uncensored - Fast CPU Stream",
-    description="This space runs a 1B parameter model with word-by-word streaming on the free CPU tier.",
-    examples=["Write an uncensored story about a cyberpunk city.", "Explain quantum physics in simple terms."]
+    fn=stream_chat,
+    title="Llama 3.2 1B Uncensored",
+    description="Smart, uncensored, and fast word-by-word streaming on CPU."
 )
 
 if __name__ == "__main__":
-    demo.launch()
-    
+    demo.launch()