This is a joke model trained on actual messages. It sometimes spits tenor links but try to ignore that

QLoRa applied on ArliAI/Gemma-2-2B-ArliAI-RPMax-v1.1

Basically i was bored out of my mind and decided to make a QLoRa, which you now see as this merged model. It can still answer basic questions but doesn't have any kind of reasoning capability and doesn't seem to write code.

I wanted 1 epoch initially, but 103 steps was enough. Might finish it later but idk we'll see.

The recomended penalties are the same as gemma, use rep pen 1.1-1.3 maybe and like 0.3-0.8 temperature depending on how lost it is. Lower Top K is better, like 40.

This model WILL be saying some wild stuff and i do not carry the responisbility for the answers it gives. Personal use only.

settings
Downloads last month
109
GGUF
Model size
3B params
Architecture
gemma2
Hardware compatibility
Log In to add your hardware

4-bit

8-bit

16-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for Timelyyyy/HOS-1