|
--- |
|
tags: |
|
- mergekit |
|
- merge |
|
|
|
library_name: transformers |
|
--- |
|
<img src="https://huggingface.co/Virt-io/FuseChat-Kunoichi-10.7B/resolve/main/README.png"> |
|
|
|
--- |
|
|
|
# Quantizations |
|
[GGUF by Lewdiculous](https://huggingface.co/Lewdiculous/FuseChat-Kunoichi-10.7B-GGUF-IQ-Imatrix) |
|
|
|
--- |
|
|
|
|
|
Merged [Kunoichi-DPO-v2-7B](https://huggingface.co/SanjiWatsuki/Kunoichi-DPO-v2-7B) into [FuseChat-7B-Varm](https://huggingface.co/FuseAI/FuseChat-7B-VaRM) to fix the GPTism |
|
|
|
The idea was to keep FuseChat's smarts since from my testing it was amazing, just a little stubborn for RP |
|
|
|
|
|
Silly tavern [preset](https://huggingface.co/Virt-io/FuseChat-7B-VaRM-GGUF/tree/main/presets) |
|
|
|
|
|
|
|
Merge template copied from [TheProfessor](https://huggingface.co/abacusai/TheProfessor-155b) |
|
|
|
--- |
|
base_model: |
|
- FuseAI/FuseChat-7B-VaRM |
|
- SanjiWatsuki/Kunoichi-DPO-v2-7B |
|
--- |
|
# FUSECHAT-VaRM-Kunoichi-10.7b.v1 |
|
|
|
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit). |
|
|
|
## Merge Details |
|
### Merge Method |
|
|
|
This model was merged using the [linear](https://arxiv.org/abs/2203.05482) merge method. |
|
|
|
### Models Merged |
|
|
|
The following models were included in the merge: |
|
* [FuseAI/FuseChat-7B-VaRM](https://huggingface.co/FuseAI/FuseChat-7B-VaRM) |
|
* [SanjiWatsuki/Kunoichi-DPO-v2-7B](https://huggingface.co/SanjiWatsuki/Kunoichi-DPO-v2-7B) |
|
|
|
### Configuration |
|
|
|
The following YAML configuration was used to produce this model: |
|
|
|
```yaml |
|
merge_method: linear # use linear so we can include multiple models, albeit at a zero weight |
|
parameters: |
|
weight: 1.0 # weight everything as 1 unless specified otherwise - linear with one model weighted at 1 is a no-op like passthrough |
|
slices: |
|
- sources: |
|
- model: FuseAI/FuseChat-7B-VaRM # embed_tokens comes along with the ride with whatever is the first layer |
|
layer_range: [0, 1] |
|
- model: SanjiWatsuki/Kunoichi-DPO-v2-7B # add dummy second model with 0 weight so tokenizer-based merge routine is invoked for embed_tokens |
|
layer_range: [0, 1] |
|
parameters: |
|
weight: 0 |
|
- sources: |
|
- model: FuseAI/FuseChat-7B-VaRM |
|
layer_range: [1, 5] |
|
- sources: |
|
- model: SanjiWatsuki/Kunoichi-DPO-v2-7B |
|
layer_range: [5, 7] # 2 layers |
|
- sources: |
|
- model: FuseAI/FuseChat-7B-VaRM |
|
layer_range: [5, 15] |
|
- sources: |
|
- model: SanjiWatsuki/Kunoichi-DPO-v2-7B |
|
layer_range: [15, 27] # 12 layers |
|
- sources: |
|
- model: FuseAI/FuseChat-7B-VaRM |
|
layer_range: [15, 27] |
|
- sources: |
|
- model: SanjiWatsuki/Kunoichi-DPO-v2-7B |
|
layer_range: [27, 29] # 2 layers |
|
- sources: |
|
- model: FuseAI/FuseChat-7B-VaRM |
|
layer_range: [27, 31] |
|
- sources: # same as above, but for lm_head with the last layer |
|
- model: FuseAI/FuseChat-7B-VaRM |
|
layer_range: [31, 32] |
|
- model: SanjiWatsuki/Kunoichi-DPO-v2-7B |
|
layer_range: [31, 32] |
|
parameters: |
|
weight: 0 |
|
dtype: float16 |
|
tokenizer_source: model:FuseAI/FuseChat-7B-VaRM # keep exact tokenizer used by dolphin - or you could use `union` if you add all of the input models to the first/last slice |
|
|
|
``` |
