metadata

language:
  - en
tags:
  - conversational
co2_eq_emissions:
  emissions: '840'
  source: mlco2.github.io
  training_type: fine-tuning
  geographical_location: West Java, Indonesia
  hardware_used: 1 Tesla P100
license: cc-by-nc-sa-4.0
widget:
  - text: Hey kekbot! What's up?
    example_title: Asking what's up
  - text: Hey kekbot! How r u?
    example_title: Asking how he is

THIS MODEL IS IN PUBLIC BETA, PLEASE DO NOT EXPECT ANY FORM OF STABILITY IN ITS CURRENT STATE.

Art Union server chatbot

Based on a DialoGPT-medium (kekbot-beta-3-medium) model, fine-tuned to a select subset (65k<= messages) of Art Union's general-chat channel chat history.

Current issues

(Which hopefully will be fixed in future iterations) Include, but not limited to:

Limited turns, after ~20 turns output may break for no apparent reason.
Inconsistent variance, acts like an overfitted model from time to time for no reason whatsoever.