Update README.md
Browse files
README.md
CHANGED
|
@@ -36,9 +36,15 @@ Three models all working together to code.
|
|
| 36 |
|
| 37 |
Default config is 2 experts (of 3 activated).
|
| 38 |
|
| 39 |
-
|
| 40 |
-
|
| 41 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
|
| 43 |
MODELS in THIS MOE:
|
| 44 |
|
|
|
|
| 36 |
|
| 37 |
Default config is 2 experts (of 3 activated).
|
| 38 |
|
| 39 |
+
Suggest using 3 experts for maximum power.
|
|
|
|
| 40 |
|
| 41 |
+
SETTINGS:
|
| 42 |
+
- Temp .5 to .7 (or lower)
|
| 43 |
+
- Max Context is 32k (final versions will be 32k, 128k and maybe higher).
|
| 44 |
+
- topk: 20, topp: .8, minp: .05
|
| 45 |
+
- rep pen: 1.05-1.1 (can be lower)
|
| 46 |
+
- Jinja Template (embedded) or CHATML template.
|
| 47 |
+
- A System Prompt is not required. (ran tests with blank system prompt)
|
| 48 |
|
| 49 |
MODELS in THIS MOE:
|
| 50 |
|