MarsupialAI
commited on
Commit
•
cb875f2
1
Parent(s):
69d7466
Update README.md
Browse files
README.md
CHANGED
@@ -1,42 +1,49 @@
|
|
1 |
---
|
2 |
-
|
3 |
-
|
|
|
4 |
tags:
|
5 |
-
-
|
6 |
-
- merge
|
7 |
-
|
8 |
---
|
9 |
-
#
|
|
|
|
|
|
|
|
|
10 |
|
11 |
-
This is a merge of
|
|
|
12 |
|
13 |
-
|
14 |
-
|
|
|
|
|
15 |
|
16 |
-
This model
|
|
|
|
|
|
|
17 |
|
18 |
-
### Models Merged
|
19 |
|
20 |
-
|
21 |
-
* f:\raw\moist14
|
22 |
-
* f:\raw\solstice14
|
23 |
-
* f:\raw\fimbul14
|
24 |
|
25 |
-
### Configuration
|
26 |
|
27 |
-
|
|
|
|
|
|
|
|
|
|
|
28 |
|
29 |
-
|
30 |
-
models:
|
31 |
-
- model: f:\raw\fimbul14
|
32 |
-
parameters:
|
33 |
-
weight: 1.0
|
34 |
-
- model: f:\raw\moist14
|
35 |
-
parameters:
|
36 |
-
weight: 1.0
|
37 |
-
- model: f:\raw\solstice14
|
38 |
-
parameters:
|
39 |
-
weight: 1.0
|
40 |
-
merge_method: linear
|
41 |
-
dtype: float16
|
42 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
license: cc-by-nc-4.0
|
3 |
+
language:
|
4 |
+
- en
|
5 |
tags:
|
6 |
+
- solar
|
|
|
|
|
7 |
---
|
8 |
+
# Skunk Ape 14b
|
9 |
+
|
10 |
+
![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/65a531bc7ec6af0f95c707b1/p9tbuezkb2qvf8kWEnO_2.jpeg)
|
11 |
+
|
12 |
+
This version performs *substantially* better than the 16b version.
|
13 |
|
14 |
+
This model is a merge of three self-merged Solar-based models in a 14b (64 layer) configuration. The result of
|
15 |
+
this "frankenmerge" is a medium-sized model that contains what I consider to be the best of the solar finetunes.
|
16 |
|
17 |
+
Mergefuel:
|
18 |
+
- Sao10K/Fimbulvetr-11B-v2
|
19 |
+
- Sao10K/Solstice-11B-v1
|
20 |
+
- TheDrummer/Moistral-11B-v1
|
21 |
|
22 |
+
This model is uncensored and capable of generating objectionable material. However, it is not an explicitely-NSFW model,
|
23 |
+
and it has never "gone rogue" and tried to insert NSFW content into SFW prompts in my experience. As with any LLM, no
|
24 |
+
factual claims made by the model should be taken at face value. You know that boilerplate safety disclaimer that most
|
25 |
+
professional models have? Assume this has it too. This model is for entertainment purposes only.
|
26 |
|
|
|
27 |
|
28 |
+
iMatrix GGUFs: https://huggingface.co/MarsupialAI/SkunkApe-14b_iMatrix_GGUF
|
|
|
|
|
|
|
29 |
|
|
|
30 |
|
31 |
+
# Sample output
|
32 |
+
|
33 |
+
```
|
34 |
+
{{[INPUT]}}
|
35 |
+
Write a detailed and humorous story about a cute and fluffy bunny that goes to a Gwar concert.
|
36 |
+
{{[OUTPUT]}}
|
37 |
|
38 |
+
<<<This goes on for a while. See sample.txt for full output>>>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
```
|
40 |
+
|
41 |
+
|
42 |
+
# Prompt format
|
43 |
+
Prefers alpaca.
|
44 |
+
|
45 |
+
|
46 |
+
# Weird merge fuckery
|
47 |
+
According to Toasty Pigeon, FM, Akai, and probably others on the KAI discord, this merge method works better than a normal stacked merge.
|
48 |
+
I don't pretend to understand why, but the huge PPL improvement (5.96 for this model vs 7.65 for the 16b @ Q4km) indicates that they're right.
|
49 |
+
See recipe.txt for all the alchemy.
|