Update README.md
Browse files
README.md
CHANGED
@@ -2,14 +2,11 @@
|
|
2 |
datasets:
|
3 |
- ewof/koishi-instruct-metharme
|
4 |
- PygmalionAI/PIPPA
|
5 |
-
- ludis/geepeetee4
|
6 |
---
|
7 |
|
8 |
just testing for now, qlora merge, several things different between this and the 7b
|
9 |
|
10 |
-
|
11 |
-
|
12 |
-
![image/png](https://cdn-uploads.huggingface.co/production/uploads/64f38f3a89203d118da8b477/HkWCy5CQwlBGEpmmLAwuO.png)
|
13 |
|
14 |
NousResearch/Llama-2-13b-hf tuned on koishi data (without code subsets) for 1 epoch
|
15 |
|
@@ -20,3 +17,11 @@ then tuned on gpt4 rp data from whocars proxy for 1 epoch
|
|
20 |
then tuned on limarp (without ponyville, lolicit, all the fallen, and eka's portal subsets) for 2 epochs
|
21 |
|
22 |
all metharme format
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
datasets:
|
3 |
- ewof/koishi-instruct-metharme
|
4 |
- PygmalionAI/PIPPA
|
|
|
5 |
---
|
6 |
|
7 |
just testing for now, qlora merge, several things different between this and the 7b
|
8 |
|
9 |
+
## training
|
|
|
|
|
10 |
|
11 |
NousResearch/Llama-2-13b-hf tuned on koishi data (without code subsets) for 1 epoch
|
12 |
|
|
|
17 |
then tuned on limarp (without ponyville, lolicit, all the fallen, and eka's portal subsets) for 2 epochs
|
18 |
|
19 |
all metharme format
|
20 |
+
|
21 |
+
## prompting
|
22 |
+
|
23 |
+
https://rentry.org/tsukasa13b - reccomended prompts and gen settings
|
24 |
+
|
25 |
+
The current model version has been trained on prompts using three different roles, which are denoted by the following tokens: `<|system|>`, `<|user|>` and `<|model|>`.
|
26 |
+
|
27 |
+
The `<|system|>` prompt can be used to inject out-of-channel information behind the scenes, while the `<|user|>` prompt should be used to indicate user input. The `<|model|>` token should then be used to indicate that the model should generate a response. These tokens can happen multiple times and be chained up to form a conversation history.
|