Update app.py

app.py

@@ -6,7 +6,7 @@ import torch
 physician = pipeline("text-generation", model="HuggingFaceTB/SmolLM2-1.7B-Instruct")
 patient = pipeline("text-generation", model="HuggingFaceTB/SmolLM2-1.7B-Instruct")
 
-# System prompts to define roles
+# System prompts to define roles (not included in the input prompt)
 patient_system_prompt = "You are a patient describing your symptoms to a physician."
 physician_system_prompt = "You are a physician responding to a patient's symptoms."
 
@@ -17,10 +17,10 @@ def generate_conversation(topic, turns):
     patient_tokens = 0
 
     # Initial prompt for the patient
-    patient_prompt = f"I'm here to talk about {topic}."
-    print(f"Patient Initial Prompt: {patient_prompt}")  # Debugging
+    patient_input = f"Patient: I'm here to talk about {topic}."
+    print(f"Patient Initial Input: {patient_input}")  # Debugging
     patient_response = patient(
-        f"{patient_system_prompt} {patient_prompt}",
+        patient_input,  # Only the patient's message, without the system prompt
         max_length=50,  # Reduce max_length for faster responses
         num_return_sequences=1,
         truncation=True,  # Explicitly enable truncation
@@ -35,7 +35,7 @@ def generate_conversation(topic, turns):
         # Physician's turn
         print(f"Physician Turn {turn} Prompt: {patient_response}")  # Debugging
         physician_response = physician(
-            f"{physician_system_prompt} Patient says: {patient_response}",
+            f"Physician: {patient_response}",  # Only the physician's response to the patient's message
             max_length=50,  # Reduce max_length for faster responses
             num_return_sequences=1,
             truncation=True,  # Explicitly enable truncation
@@ -49,7 +49,7 @@ def generate_conversation(topic, turns):
         # Patient's turn
         print(f"Patient Turn {turn} Prompt: {physician_response}")  # Debugging
         patient_response = patient(
-            f"{patient_system_prompt} Physician says: {physician_response}",
+            f"Patient: {physician_response}",  # Only the patient's response to the physician's message
             max_length=50,  # Reduce max_length for faster responses
             num_return_sequences=1,
             truncation=True,  # Explicitly enable truncation