sda

app.py

@@ -60,10 +60,11 @@ def respond(
     if system_message and system_message.strip():
         # How a system message is used can vary. For this model, prepending it might work.
         # Or, it could be part of the initial "USER:" turn if the model expects that.
-        # The example prompt is `USER: {prompt}
-ASSISTANT:`
+        # The example prompt format for some models is:
+        # USER: {prompt}
+        # ASSISTANT: {response}
         # We will integrate system_message as part of the first user turn or as general context.
-        # For now, let's prepend it simply.
+        # For now, let's prepend it simply to the overall prompt.
         prompt_parts.append(system_message)
 
     for user_msg, assistant_msg in history:
@@ -100,8 +101,8 @@ ASSISTANT:`
         # top_k=50, # Another sampling param
     )
     
-    # Ensure temperature is valid
-    if generation_kwargs['temperature'] <= 0:
+    # Ensure temperature is valid for sampling
+    if generation_kwargs['temperature'] <= 1e-4: # Using a small epsilon for float comparison
         generation_kwargs['temperature'] = 0.01 # A very small value for near-deterministic
         generation_kwargs['do_sample'] = False
     else: