File size: 1,775 Bytes
6b8127b 0ea61b0 c496519 9f5fbdd 6b8127b 0ea61b0 3845c8f 0ea61b0 9f5fbdd 3845c8f 0ea61b0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 |
---
base_model:
- grimjim/kukulemon-7B
library_name: transformers
tags:
- mergekit
- merge
license: cc-by-nc-4.0
pipeline_tag: text-generation
---
# kululemon-spiked-9B
This is a frankenmerge of a pre-trained language model created using [mergekit](https://github.com/cg123/mergekit). As an experiment, this appears to be a partial success.
Lightly tested with temperature 1-1.2 and minP 0.01 with ChatML prompts; the model supports Alpaca prompts and has 8K context length, a result of its Mistral v0.1 provenance. The model's output has been coherent and stable during aforementioned testing.
The merge formula for this frankenmerge is below. It is conjectured that the shorter first section is not key to variation, the middle segment is key to balancing reasoning and variation, and that the lengthy final section is required for convergence and eventual stability. The internal instability is probably better suited for narrative involving unstable and/or unhinged characters and situations.
Quants available:
- [GGUF](https://huggingface.co/grimjim/kukulemon-spiked-9B-GGUF)
- [8.0bpw h8 exl2](https://huggingface.co/grimjim/kukulemon-spiked-9B-8.0bpw_h8_exl2)
## Merge Details
### Merge Method
This model was merged using the passthrough merge method.
### Models Merged
The following models were included in the merge:
* [grimjim/kukulemon-7B](https://huggingface.co/grimjim/kukulemon-7B)
### Configuration
The following YAML configuration was used to produce this model:
```yaml
slices:
- sources:
- model: grimjim/kukulemon-7B
layer_range: [0, 12]
- sources:
- model: grimjim/kukulemon-7B
layer_range: [8, 16]
- sources:
- model: grimjim/kukulemon-7B
layer_range: [12, 32]
merge_method: passthrough
dtype: float16
```
|