48 11 months ago

A more experimental and buggy, but smaller model. A QLOrA finetune created with publicly available information and datasets. Based off of TinyLlama:1.1b. (AS OF RECENT TESTS, THIS MODEL IS ABLE TO RESPOND BETTER AND FASTER THAN GIBBER 2, NOT PERFECTLY.)

cc82da57dc2f ยท 139B
{
"stop": [
"<|system|>",
"<|user|>",
"<|assistant|>",
"</s>"
],
"temperature": 0.7,
"top_k": 40,
"top_p": 0.9
}