Update README.md
Browse files
README.md
CHANGED
|
@@ -9,28 +9,25 @@ tags:
|
|
| 9 |
- merge
|
| 10 |
- roleplay
|
| 11 |
---
|
| 12 |
-
#
|
| 13 |
|
| 14 |
-
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
|
| 15 |
|
| 16 |
-
## Merge
|
| 17 |
-
### Merge Method
|
| 18 |
|
| 19 |
-
|
|
|
|
| 20 |
|
| 21 |
-
###
|
| 22 |
-
|
| 23 |
-
The following models were included in the merge:
|
| 24 |
-
* [Vortex5/Moonlit-Shadow-12B](https://huggingface.co/Vortex5/Moonlit-Shadow-12B)
|
| 25 |
-
* ./intermediates/First
|
| 26 |
|
| 27 |
-
|
|
|
|
|
|
|
| 28 |
|
| 29 |
-
The following YAML configuration was used to produce this model:
|
| 30 |
<details>
|
| 31 |
-
<summary><b>
|
| 32 |
|
| 33 |
-
```yaml
|
| 34 |
name: First
|
| 35 |
base_model: Vortex5/Vermilion-Sage-12B
|
| 36 |
models:
|
|
@@ -41,19 +38,40 @@ parameters:
|
|
| 41 |
normalize: true
|
| 42 |
t: [0.25, 0.35, 0.45, 0.55, 0.65, 0.75, 0.6, 0.5, 0.6, 0.6]
|
| 43 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
```yaml
|
| 45 |
dtype: bfloat16
|
| 46 |
merge_method: karcher
|
| 47 |
modules:
|
| 48 |
default:
|
| 49 |
slices:
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
parameters:
|
| 56 |
-
max_iter: 9999
|
| 57 |
-
tol:
|
| 58 |
```
|
| 59 |
-
</details>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
- merge
|
| 10 |
- roleplay
|
| 11 |
---
|
| 12 |
+
# 🩵 LunaMaid-12B
|
| 13 |
|
| 14 |
+
This is a multi-stage merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
|
| 15 |
|
| 16 |
+
## 🧬 Merge Overview
|
|
|
|
| 17 |
|
| 18 |
+
**LunaMaid-12B** was produced through a **two-stage multi-model merge** using [MergeKit](https://github.com/arcee-ai/mergekit).
|
| 19 |
+
Each stage fuses models with complementary linguistic and stylistic traits to create a cohesive, emotionally nuanced personality.
|
| 20 |
|
| 21 |
+
### 🩵 **Stage 1 — Slerp Merge (Intermediate Model `First`)**
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
|
| 23 |
+
- **Base Model:** [Vortex5/Vermilion-Sage-12B](https://huggingface.co/Vortex5/Vermilion-Sage-12B)
|
| 24 |
+
- **Merged With:** [yamatazen/NeonMaid-12B-v2](https://huggingface.co/yamatazen/NeonMaid-12B-v2)
|
| 25 |
+
- **Method:** Spherical Linear Interpolation (Slerp)
|
| 26 |
|
|
|
|
| 27 |
<details>
|
| 28 |
+
<summary><b>Stage 1 Configuration</b></summary>
|
| 29 |
|
| 30 |
+
```yaml
|
| 31 |
name: First
|
| 32 |
base_model: Vortex5/Vermilion-Sage-12B
|
| 33 |
models:
|
|
|
|
| 38 |
normalize: true
|
| 39 |
t: [0.25, 0.35, 0.45, 0.55, 0.65, 0.75, 0.6, 0.5, 0.6, 0.6]
|
| 40 |
```
|
| 41 |
+
</details>
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
### 🌑 **Merge Method — Karcher Mean Merge (Final Model)**
|
| 46 |
+
|
| 47 |
+
- **Base Model:** Intermediate output from Stage 1 `./intermediates/First`
|
| 48 |
+
- **Merged With:** [Vortex5/Moonlit-Shadow-12B](https://huggingface.co/Vortex5/Moonlit-Shadow-12B)
|
| 49 |
+
- **Method:** [Karcher Mean](https://en.wikipedia.org/wiki/Karcher_mean) (Riemannian Barycenter)
|
| 50 |
+
|
| 51 |
+
<details>
|
| 52 |
+
<summary><b>Stage 2 Configuration</b></summary>
|
| 53 |
+
|
| 54 |
```yaml
|
| 55 |
dtype: bfloat16
|
| 56 |
merge_method: karcher
|
| 57 |
modules:
|
| 58 |
default:
|
| 59 |
slices:
|
| 60 |
+
- sources:
|
| 61 |
+
- layer_range: [0, 40]
|
| 62 |
+
model: ./intermediates/First
|
| 63 |
+
- layer_range: [0, 40]
|
| 64 |
+
model: Vortex5/Moonlit-Shadow-12B
|
| 65 |
parameters:
|
| 66 |
+
max_iter: 9999
|
| 67 |
+
tol: 1e-9
|
| 68 |
```
|
| 69 |
+
</details>
|
| 70 |
+
|
| 71 |
+
### Models Merged
|
| 72 |
+
|
| 73 |
+
The following models were included in the merge:
|
| 74 |
+
* [Vortex5/Moonlit-Shadow-12B](https://huggingface.co/Vortex5/Moonlit-Shadow-12B)
|
| 75 |
+
* [Vortex5/Vermilion-Sage-12B](https://huggingface.co/Vortex5/Vermilion-Sage-12B)
|
| 76 |
+
* [yamatazen/NeonMaid-12B-v2](https://huggingface.co/yamatazen/NeonMaid-12B-v2)
|
| 77 |
+
* ./intermediates/First
|