Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Ballpark Trivia: Size XL
|
| 2 |
+
Are you frequently asked google-able Trivia questions and annoyed by it? Well, this is the model for you! Ballpark Trivia Bot answers any trivia question with something that sounds plausible but is probably not 100% correct. It should be _more_ credible or at least able to handle more kinds of questions
|
| 3 |
+
## Training
|
| 4 |
+
This text gen model is a GPT-2 ~1.5 B Parameter Size XL Model, first trained on [Wizard of Wikipedia](https://parl.ai/projects/wizard_of_wikipedia/) for 40k steps (**33**/36 layers frozen for the fine-tuning), and then subsequently trained for 40k steps on a parsed variant of [Natural Questions](https://ai.google.com/research/NaturalQuestions)(then **34**/36 layers frozen for the second fine-tuning) to accidentally create this model.
|
| 5 |
+
|
| 6 |
+
Note that because the model was originally trained for use in a [chatbot application](https://github.com/pszemraj/ai-msgbot), it uses a named conversation dialogue structure, _i.e. the questions are asked by person alpha, and responded to by person beta_. Even if you don't specify person alpha in the prompt, it hopefully responds to any question.
|