Description

This is my first attempt to simulate the Filian's Twitch stream chat based on scraped data. The model is a fine-tuning of GPT-Neo for 125M parameters. The model was able to be trained within 30 minutes on a GTX 1650 with Pytorch's memory pinning feature to share the local RAM with the GPU.

Usage

Make sure to include the prefix "STREAMER: " at the end of the prompt or anywhere the response of the streamer is supposed to go. "Chat messages" are prepend with the "CHAT: " keyword. In case the Hosted Inference API (text box on the side) gets too little text, just keep pressing "Compute" until the sentence ends.

Prompt examples

CHAT: You're cringe
CHAT: KEKW
STREAMER:
CHAT: Are you a Filipino boy?
STREAMER: 
CHAT: Hello
CHAT: KEKW Filian is cringe
STREAMER: 

Using the keyword "STREAM TITLE" will attempt to replicate the name of a stream:

STREAM TITLE: 

You can also induce the streamer to say something by adding a few words of your own at the beginning:

STREAM: Today we will
Downloads last month
14
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train Differentiable/twitch-simulator-filian-neo-125M