Update app.py

app.py

@@ -1,46 +1,38 @@
+# build on your original chatbot from the previous lesson
+# a basic chatbot from the previous lesson is below -- edit it to incorporate the changes described above
+
 import gradio as gr
-from huggingface_hub import InferenceClient
-import re
+from huggingface_hub import InferenceClient #imports huggingface models
 
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+client = InferenceClient("google/gemma-2-2b-it")
 
 def respond(message, history):
-    messages = [{"role": "system", "content": "You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe."}]
+    messages = [{"role": "system", "content": "I am a kind chatbot."}]
 
-    # Add all previous messages to the messages list
-    for human, assistant in history:
-        messages.append({"role": "user", "content": human})
-        messages.append({"role": "assistant", "content": assistant})
+# add all previous messages to the messages list
+    if history:
+        messages.extend(history)
 
-    # Add the current user\"s message to the messages list
+    # add the current user's message to the messages list
     messages.append({"role": "user", "content": message})
 
+    # makes the chat completion API call,
+    # sending the messages and other parameters to the model
+    # implements streaming, where one word/token appears at a time
     response = ""
-    for chunk in client.chat_completion(
+
+    # iterate through each message in the method
+    for message in client.chat_completion(
         messages,
-        max_tokens=500, 
-        temperature=0.5, 
-        stream=True
-    ):
-        token = chunk.choices[0].delta.content
-        response += token
-        
-        # Enhanced post-processing to handle unwanted patterns
-        cleaned_response = response
-        
-        # Remove special tokens and unwanted labels more aggressively
-        # Added \s* to catch spaces around the tags, and made the regex more comprehensive
-        cleaned_response = re.sub(r"<\|endoftext\|>|<\|user\|>|<\|assistant\|>|<\|system\>|\s*\[USER\]\s*|\s*\[/ASS\]\s*|\s*\[ASS\]\s*|\s*\[/USER\]\s*|\s*\[INST\]\s*|\s*\[/INST\]\s*|\s*\[ASSIST\]\s*|\s*\[/ASSIST\]\s*", "", cleaned_response)
-        
-        # Remove any remaining HTML-like tags (e.g., <p>, <div>) that might appear
-        cleaned_response = re.sub(r"<[^>]*>", "", cleaned_response)
-        
-        # Clean up extra whitespace and newlines, ensuring single newlines between paragraphs
-        cleaned_response = re.sub(r"\n+", "\n", cleaned_response)
-        cleaned_response = cleaned_response.strip()
-        
-        yield cleaned_response
+        max_tokens=100,
+        temperature=.1,
+        stream=True):
+
+        # add the tokens to the output content
+          token = message.choices[0].delta.content # capture the most recent toke
+          response += token # Add it to the response
+          yield response # yield the response:
 
 chatbot = gr.ChatInterface(respond)
 
-chatbot.launch()
+chatbot.launch()