metadata
library_name: transformers
tags:
- mergekit
- merge
- roleplay
- mistral
Information
Attempt at extending context window for an older Mistral-v0.1 model.
It seems to work fine at 16K.
Irene-RP-v4-7B
This is a merge of pre-trained language models created using mergekit.
Merge Details
Merge Method
This model was merged using the SLERP merge method.
Models Merged
The following models were included in the merge:
Configuration
The following YAML configuration was used to produce this model:
slices:
- sources:
- model: Virt-io/Helen-v1_7B
layer_range: [0, 32]
- model: Virt-io/Irene-RP-v3-7B
layer_range: [0, 32]
merge_method: slerp
base_model: Virt-io/Helen-v1_7B
parameters:
t:
- filter: self_attn
value: [0.25, 0.45, 0.50, 0.20, 0.25]
- filter: mlp
value: [0.35, 0.45, 0.55, 0.20, 0.25]
- value: 0.25 # fallback for rest of tensors
dtype: float16