|
--- |
|
license: other |
|
language: |
|
- en |
|
--- |
|
The Llama 2 sequel to my [original experiment](https://huggingface.co/Gryphe/MythoLogic-13b) with gradient merges using [the following script](https://github.com/Gryphe/BlockMerge_Gradient). Its three models ([Hermes](https://huggingface.co/NousResearch/Nous-Hermes-Llama2-13b), [Chronos](https://huggingface.co/elinas/chronos-13b-v2) and [Airoboros](https://huggingface.co/jondurbin/airoboros-l2-13b-gpt4-2.0)) are almost evenly divided over the layer structure this time. Airoboros was the "wildcard model" due to its superior ability to understand complex instructions. |
|
|
|
## Model details |
|
|
|
As before, the main objective was to create an all-round model with improved roleplaying capabilities. MythoLogic-L2 differs from its predecessor in that it focuses primarily on the understanding of instructions and personalities of complex character cards. |
|
|
|
Illustrated below are the gradients used for this specific L2 recipe, with the top of the image representing layer 0 and the bottom layer 40. |
|
|
|
 |
|
|
|
## Prompt Format |
|
|
|
This model primarily uses (and was heavily tested with) Alpaca formatting, so for optimal model performance, use: |
|
``` |
|
### Instruction: |
|
Your instruction or question here. |
|
### Response: |
|
``` |
|
|
|
--- |
|
license: other |
|
--- |
