Testish instruct checkpoint of Koto.

Upscale Nemo 12B to 22B

Do a continual pretrain for 1B tokens on creative data (big thanks to Allura Org for doing this)

Train on ~ 17K instruct samples.

Uses alpaca format.

Safetensors

Model size

22B params

Tensor type

BF16

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for ConicCat/Ne-Marvin-22B

Base model

Finetuned

Finetuned

(1)

this model

Quantizations