Building conversations

Working with Llama 3

Imtihan Ahmed

Machine Learning Engineer

Maintaining context

Maintaining context workflow - step 1

Maintaining context

Maintaining context workflow - step 2

Maintaining context

Maintaining context workflow - step 3

Maintaining context

Maintaining context workflow - step 4

Maintaining context

Maintaining context workflow - step 5

Track a chat history with a Conversation class

Conversation class

Can store a history of prior messages

class Conversation:

    def __init__(self, llm: Llama, system_prompt='', history=[]):
        self.llm = llm
        self.system_prompt = system_prompt

        self.history = [{"role": "system", "content": self.system_prompt}] + history

Conversation class

Can store a history of prior messages

class Conversation:
    def __init__(self, llm: Llama, system_prompt='', history=[]):
        self.llm = llm
        self.system_prompt = system_prompt
        self.history = [{"role": "system", "content": self.system_prompt}] + history


    def create_completion(self, user_prompt=''):

        self.history.append({"role": "user", "content": user_prompt}) # Append input

        output = self.llm.create_chat_completion(messages=self.history)

        conversation_result = output['choices'][0]['message']
        self.history.append(conversation_result) # Append output
        return conversation_result['content'] # Return output

Running a multi-turn conversation

conversation = Conversation(llm, system_prompt="You are a virtual travel assistant 
                                                helping with planning trips.")


response1 = conversation.create_completion("What are some destinations in France for a 
                                            short weekend break?")

print(f"Response 1: {response1}")


response2 = conversation.create_completion("How about Spain?")


print(f"Response 2: {response2}")

Running a multi-turn conversation

print(f"Response 1: {response1}")

print(f"Response 2: {response2}")

Output

Running a multi-turn conversation

print(f"Response 1: {response1}")

print(f"Response 2: {response2}")

Output with highlighted parts

Let's practice!

Working with Llama 3