|
--- |
|
datasets: |
|
- PygmalionAI/PIPPA |
|
- lemonilia/LimaRP |
|
--- |
|
|
|
## Training |
|
|
|
[axolotl](https://github.com/OpenAccess-AI-Collective/axolotl) was used for training |
|
on a 4x nvidia a100 gpu cluster. |
|
|
|
the a100 GPU cluster has been graciously provided by [lloorree](https://huggingface.co/lloorree). |
|
|
|
rank 16 qlora (all modules) tune |
|
|
|
base model alpindale/goliath-120b tuned on koishi commit 6e675d1 for one epoch |
|
|
|
then tuned on pippa 6412b0c for one epoch (metharme completion) |
|
|
|
then tuned on limarp Version 2023-10-19 for 2 epochs in metharme completion format |
|
|
|
## Prompting |
|
|
|
The current model version has been trained on prompts using three different roles, which are denoted by the following tokens: `<|system|>`, `<|user|>` and `<|model|>`. |
|
|
|
The `<|system|>` prompt can be used to inject out-of-channel information behind the scenes, while the `<|user|>` prompt should be used to indicate user input. The `<|model|>` token should then be used to indicate that the model should generate a response. These tokens can happen multiple times and be chained up to form a conversation history. |
