kunoichi-lemon-royale-v2-32K-7B-GGUF

This repository contains GGUF quantizations of kunoichi-lemon-royale-v2-32K-7B.

This merge amounts to the grafting of a model derived from Mistral v0.1 (4K sliding window context, to a maximum of 8K practical context length) onto a model derived from Mistral v0.2 (32K context length). It appears to work, although rope_theta in config.json was lowered from 1000000.0 to 100000.0, which works well enough to 16K.

In light testing, this model appears to follow formatting very well, with temperature 1.0 and minP 0.01, using ChatML prompts, even though the underlying model claims to follow Alpaca prompts.

This is a merge of pre-trained language models created using mergekit.

• Full weights: kunoichi-lemon-royale-v2-32K-7B
• GGUFs: kunoichi-lemon-royale-v2-32K-7B-GGUF

Merge Details

Merge Method

This model was merged using the task arithmetic merge method using grimjim/Mistral-7B-Instruct-demi-merge-v0.2-7B as a base.

Models Merged

The following models were included in the merge:

• grimjim/kunoichi-lemon-royale-7B

Configuration

The following YAML configuration was used to produce this model:

base_model: grimjim/Mistral-7B-Instruct-demi-merge-v0.2-7B
dtype: bfloat16
merge_method: task_arithmetic
slices:
- sources:
  - layer_range: [0, 32]
    model: grimjim/Mistral-7B-Instruct-demi-merge-v0.2-7B
  - layer_range: [0, 32]
    model: grimjim/kunoichi-lemon-royale-7B
    parameters:
      weight: 1.00