Intro to Chatbots with HuggingFace

In this tutorial, we will show you how to use the Transformers library from HuggingFace to build chatbot pipelines. Let’s start by installing the transformers library:

pip install transformers

Once we install the library, we can move on. We will work with the ‘blenderbot-400M-distill’ model from Meta. This is a small open-source model (700 MB) that performs relatively well. Let’s start with the pipeline.

from transformers import pipeline

chatbot = pipeline(task="conversational",
                   model="./models/facebook/blenderbot-400M-distill")
 

At this point, I will initiate the conversation by passing the following message:

My name is George and I’m from Greece. Have you ever been to my country?

user_message = """
My name is George and I'm from Greece. Have you ever been to my country?
"""

from transformers import Conversation

conversation = Conversation(user_message)

print(conversation)
 

Output:

Conversation id: b1f14acb-7d8c-47c8-9fd4-4c587acbd6ad
user: 
My name is George and I am from Greece. Have you ever been to my country?

Let’s see what answer we get.

conversation = chatbot(conversation)
print(conversation)
 

Output:

Conversation id: b1f14acb-7d8c-47c8-9fd4-4c587acbd6ad
user: 
My name is George and I am from Greece. Have you ever been to my country?

assistant:  I have not, but I would love to go one day. I hear it's beautiful there.
 

The chatbot responded:

assistant: I have not, but I would love to go one day. I hear it’s beautiful there.

Now, we will ask the chatbot the following question:

What is my name?

print(chatbot(Conversation("What is my name?")))

Output:

Conversation id: 0d716331-6af0-47a5-8c8a-30c40a4af8ee
user: What is my name?
assistant:  I don't know. What do you do for a living? I'm an accountant.
 

As we can see, the chatbot gave an unrelated response because it did not remember the previous conversation. To include prior conversations in the LLM’s context, we should add a ‘message’ to the chat history. Let’s see how to do it.

conversation.add_message(
    {"role": "user",
     "content": """
What is my name?
"""
    })

print(conversation)
 

Output:

Conversation id: dc38a6d7-41f9-4eb3-adff-a63451d07199
user: 
My name is George and I'm from Greece. Have you ever been to my country?

assistant:  I have not, but I would love to go one day. I hear it's beautiful there.
user: 
What is my name?
 

Now, we can pass the conversation to the chatbot:

conversation = chatbot(conversation)

print(conversation)
 

Output:

Conversation id: dc38a6d7-41f9-4eb3-adff-a63451d07199
user: 
My name is George and I'm from Greece. Have you ever been to my country?

assistant:  I have not, but I would love to go one day. I hear it's beautiful there.
user: 
What is my name?

assistant:  George is a beautiful name. I love Greek food. What is your favorite food?
 

As we can see, we managed to pass the history to the model and got a proper answer!

References

[1] Deeplearning.ai