Assigning chat roles

Working with Llama 3

Imtihan Ahmed

Machine Learning Engineer

Defining roles

Refining model's tone
Example: customer support chatbot

Conversation with a chatbot - 1

Defining roles

Refining model's tone
Example: customer support chatbot

Conversation with a chatbot - 2

Defining roles

Refining model's tone
Example: customer support chatbot

Conversation with a chatbot - 3

Using roles in chat completion

Chat roles to guide Llama's responses

Screenshot 2025-02-26 at 15.32.28.png

Sets the personality and style

Screenshot 2025-02-26 at 15.31.50.png

Represents the person asking the question

Using roles in chat completion

Sending a structured conversation
create_chat_completion() function

from llama_cpp import Llama

llm = Llama(model_path="path/to/model.gguf")


message_list = [...] # This list includes roles


response = llm.create_chat_completion(

        messages = message_list

)

The system role

System message: instructions about how model should behave

system_message = "You are a business consultant who gives data-driven answers."



message_list = [{

                 "role": "system",

                 "content": system_message

                 }]

The user role

User message: the prompt being asked to the model

system_message = "You are a business consultant who gives data-driven answers."

user_message = "What are the key factors in a successful marketing strategy?"


message_list = [{"role": "system", "content": system_message},

                {
                 "role": "user",
                 "content": user_message
                 }

                ]

Generating the response

from llama_cpp import Llama
llm = Llama(model_path="path/to/model.gguf")

system_message = "You are a business consultant who gives data-driven answers."
user_message = "What are the key factors in a successful marketing strategy?"

message_list = [{"role": "system", "content": system_message},
               {"role": "user", "content": user_message}]


response = llm.create_chat_completion(messages = message_list)
print(response)

{'id': ..., 'object': ..., 'created': ..., 'model': ..., 'choices': [...], ...}

The assistant role

response["choices"][0]

Response output

The assistant role

response["choices"][0]

Response output with message highlighted

The assistant role

response["choices"][0]

Response output with role highlighted

The assistant role

response["choices"][0]

Response output with content highlighted

result['choices'][0]['message']['content']

'A successful marketing strategy relies on clear objectives, established 
through specific, measurable goals. Understanding the target audience ...'

Let's practice!

Working with Llama 3