This is SOLAR-10.7B, but upscaled to 13B, to optimize VRAM usage of typical GPU cards (a 4bit quant fits in 12GB).

Evaluations coming soon!

This is a frankenmerge model created using mergekit.

This model was merged using the passthrough merge method.

The following models were included in the merge:

The following YAML configuration was used to produce this model:

The same as in SOLAR-10.7B:

<s> ### User:
{prompt}

### Assistant:
{response}</s>

Detailed results can be found here