gpt2-mi-reflector / README.md
andrewbrown's picture
Update README - New widget to test for users
0c5077b
metadata
license: openrail
language:
  - en
pipeline_tag: text-generation
tags:
  - medical
widget:
  - text: >
      Prompt: To start, could you tell me the things that you like about
      smoking?

      Response: It helps me stay focused when I'm working
    example_title: What do you like about smoking?
  - text: >
      Prompt: Now, what is one thing about your smoking habit that you would
      like to change?

      Response: I would love to decrease the number of cigarettes per day.
    example_title: What would you change about your about smoking?

GPT2-MI-Reflector

A gpt2 based motivational interviewing reflector

Given a prompt and a response, this GPT2 based transformer will generate a reflection. It was fine-tuned on prompt/response/reflection triplets so it should also be inferenced with the same input style. Examples are shown below.

The triplets this reflector was fine-tuned on focus on smoking-cessation. Given this, its best performance is when inferencing about smoking.

To properly inference model (as it was fine-tuned), use the following format:


Prompt: ...Question from bot...

Response: ...Response from user...


Example Prompt: paste two lines below and generate a reflection

Prompt: What do you want to change in your smoking?

Response: I want to smoke less often in my life, its become quite stressful for me.


Model should generate a motivational interviewing adherent statement

This is the first model (of many!) that I'll upload to huggingface!