Free response support

Copy-pasting idea from Discord user

> INSPIRATION:
I was going through the "Deep RL Course" and for as much as I love huggingface and its courses it seemed ironic that the "free response" questions (https://huggingface.co/deep-rl-course/unit1/quiz?fw=pt#q5-what-is-the-explorationexploitation-tradeoff) didn't leverage Huggingface and include a gradio input (or something along those lines) that would analyze sentence similarity with the input and the answer.

> FEATURE:
Create an input that takes in the correct answer and user input and returns how correct it is leveraging models already in Huggingface Hub. Similar to (https://github.com/huggingface/deep-rl-class/blob/07008786541ad96e143f95be309f2a89a80ff988/units/en/unit1/quiz.mdx?plain=1#L28)

> SEEKING ADVICE:
I really wanted to make the contribution but when researching how the tag worked I realized here https://github.com/huggingface/deep-rl-class/blob/07008786541ad96e143f95be309f2a89a80ff988/units/en/unit1/quiz.mdx?plain=1#L28 that there is some additional processing happening at some other step that does not exist within this repository. Where else should I look to see how the tag works?
Any additional advice on how to create this feature and contribute it back to Hugginface?