Update app.py

app.py

@@ -1,59 +1,53 @@
+import time
 import gradio as gr
 from witness.witness_rzero import WitnessRZero
 import app_math as app_math
 
-# Instantiate WitnessRZero – change device to "cuda" if GPU is available
-wrz = WitnessRZero(device="cpu")
+wrz = WitnessRZero(device="cpu")  # remote inference speed dominates anyway
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    # Build conversation history in OpenAI-style message format
+HISTORY_TURNS = 3          # keep only last N turns
+PROMPT_CHAR_BUDGET = 6000  # trim long contexts
+GENERATION_TIME_CAP_S = 20 # stop streaming after N seconds
+
+def build_prompt(message, history, system_message):
+    # keep only the last HISTORY_TURNS exchanges
+    short_hist = history[-HISTORY_TURNS:] if history else []
     messages = [{"role": "system", "content": system_message}]
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+    for u, a in short_hist:
+        if u: messages.append({"role": "user", "content": u})
+        if a: messages.append({"role": "assistant", "content": a})
     messages.append({"role": "user", "content": message})
-
-    # Concatenate all into a single prompt for WitnessRZero
     prompt = ""
     for m in messages:
         prompt += f"{m['role'].capitalize()}: {m['content']}\n"
+    # hard trim to avoid token explosion
+    return prompt[-PROMPT_CHAR_BUDGET:]
 
-    response = ""
-    # Stream the output from WitnessRZero.generate()
-    for token in wrz.client.text_generation(
+def respond(message, history, system_message, max_tokens, temperature, top_p):
+    prompt = build_prompt(message, history, system_message)
+    response, start = "", time.time()
+    stream = wrz.client.text_generation(
         prompt,
         max_new_tokens=max_tokens,
         stream=True,
         temperature=temperature,
         top_p=top_p,
-    ):
-        part = token  # huggingface_hub’s stream yields token text chunks
+        # return_full_text=False  # set this in WitnessRZero if available
+    )
+    for part in stream:
         response += part
         yield response
+        if time.time() - start > GENERATION_TIME_CAP_S:
+            yield response + "\n\n[stopped for speed; try 'Max new tokens' higher or ask a smaller question]"
+            return
 
-# Build the Gradio ChatInterface
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
         gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+        gr.Slider(minimum=1, maximum=2048, value=128, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.9, step=0.05, label="Top-p (nucleus sampling)"),
     ],
 )
 
@@ -62,3 +56,4 @@ if __name__ == "__main__":
 
 
 
+