Update app.py

app.py

@@ -52,7 +52,7 @@ def retrieve_from_memory(query, top_k=2):
 
 # --- Swarm Agent Function with Shared Memory (RAG) - DECORATED with @spaces.GPU ---
 @spaces.GPU  # <----  GPU DECORATOR ADDED HERE!
-def swarm_agent_sequential_rag(user_prompt):
+def swarm_agent_sequential_rag(user_prompt, temperature=0.7, top_p=0.9, max_new_tokens=300): # Added settings as arguments
     global shared_memory
     shared_memory = [] # Clear memory for each new request
 
@@ -64,7 +64,13 @@ def swarm_agent_sequential_rag(user_prompt):
     context_1_5b = "\n".join([f"- {mem}" for mem in retrieved_memory_1_5b]) if retrieved_memory_1_5b else "No relevant context found in memory."
     prompt_1_5b = f"Context from Shared Memory:\n{context_1_5b}\n\nYou are a quick idea generator. Generate an initial response to the following user request, considering the context above:\n\nUser Request: {user_prompt}\n\nInitial Response:"
     input_ids_1_5b = tokenizers["1.5B"].encode(prompt_1_5b, return_tensors="pt").to(models["1.5B"].device)
-    output_1_5b = models["1.5B"].generate(input_ids_1_5b, max_new_tokens=200, temperature=0.7, do_sample=True) # Reverted to original max_new_tokens (can adjust)
+    output_1_5b = models["1.5B"].generate(
+        input_ids_1_5b,
+        max_new_tokens=max_new_tokens, # Use user-defined max_new_tokens
+        temperature=temperature, # Use user-defined temperature
+        top_p=top_p,           # Use user-defined top_p
+        do_sample=True
+    )
     response_1_5b = tokenizers["1.5B"].decode(output_1_5b[0], skip_special_tokens=True)
     print(f"1.5B Response:\n{response_1_5b}")
     store_in_memory(f"1.5B Model Initial Response: {response_1_5b[:200]}...")
@@ -75,27 +81,41 @@ def swarm_agent_sequential_rag(user_prompt):
     context_7b = "\n".join([f"- {mem}" for mem in retrieved_memory_7b]) if retrieved_memory_7b else "No relevant context found in memory."
     prompt_7b = f"Context from Shared Memory:\n{context_7b}\n\nYou are a detailed elaborator. Take the following initial response and elaborate on it, adding more detail and reasoning, considering the context above. \n\nInitial Response:\n{response_1_5b}\n\nElaborated Response:"
     input_ids_7b = tokenizers["7B"].encode(prompt_7b, return_tensors="pt").to(models["7B"].device)
-    output_7b = models["7B"].generate(input_ids_7b, max_new_tokens=300, temperature=0.7, do_sample=True) # Reverted to original max_new_tokens
+    output_7b = models["7B"].generate(
+        input_ids_7b,
+        max_new_tokens=max_new_tokens + 100, # Slightly more tokens for 7B
+        temperature=temperature, # Use user-defined temperature
+        top_p=top_p,           # Use user-defined top_p
+        do_sample=True
+    )
     response_7b = tokenizers["7B"].decode(output_7b[0], skip_special_tokens=True)
     print(f"7B Response:\n{response_7b}")
     store_in_memory(f"7B Model Elaborated Response: {response_7b[:200]}...")
 
-    # No 14B Model Stage anymore
-
     return response_7b # Now returns the 7B model's response as final
 
 
-# --- Gradio Interface --- (Modified to reflect 2-model output)
-def gradio_interface(user_prompt):
-    final_response = swarm_agent_sequential_rag(user_prompt) # Get the final response (from 7B now)
-    return final_response
+# --- Gradio ChatInterface ---
+def gradio_interface(message, history, temperature, top_p, max_tokens): # Accept settings from interface
+    # history is automatically managed by ChatInterface
+    response = swarm_agent_sequential_rag(
+        message,
+        temperature=temperature,
+        top_p=top_p,
+        max_new_tokens=int(max_tokens) # Ensure max_tokens is an integer
+    )
+    return response
 
-iface = gr.Interface(
+iface = gr.ChatInterface( # Using ChatInterface now
     fn=gradio_interface,
-    inputs=gr.Textbox(lines=5, placeholder="Enter your task here..."),
-    outputs=gr.Textbox(lines=10, placeholder="Agent Swarm Output will appear here..."),
-    title="DeepSeek Agent Swarm (ZeroGPU Demo - 2 Models)", # Updated title
-    description="Agent swarm using DeepSeek-R1-Distill models (1.5B, 7B) with shared memory. **GPU accelerated using ZeroGPU!** (Requires Pro Space)", # Updated description
+    # Define additional inputs for settings
+    additional_inputs=[
+        gr.Slider(minimum=0.1, maximum=1.0, step=0.1, value=0.7, label="Temperature"),
+        gr.Slider(minimum=0.01, maximum=1.0, step=0.05, value=0.9, label="Top P"),
+        gr.Number(value=300, label="Max Tokens", precision=0), # Use Number for integer tokens
+    ],
+    title="DeepSeek Agent Swarm Chat (ZeroGPU Demo - 2 Models)", # Updated title
+    description="Chat with a DeepSeek agent swarm (1.5B, 7B) with shared memory and adjustable settings. **GPU accelerated using ZeroGPU!** (Requires Pro Space)", # Updated description
 )
 
 if __name__ == "__main__":