Update README.md
Browse files
README.md
CHANGED
|
@@ -30,10 +30,14 @@ Note that 128k and 256k versions tends to elongate output too, and add in more d
|
|
| 30 |
|
| 31 |
Longer, more detailed prompts may "contain" the model's output length somewhat.
|
| 32 |
|
| 33 |
-
Also with the 128k/256k you may need to stop the model's generation.
|
| 34 |
|
| 35 |
IE: You ask for a scene of 1000-2000 words, and it may produce multiple scenes (in sequence!) of 1000-2000 words EACH.
|
| 36 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
For the 256k context version, keep prompts as clear as possible otherwise the model will have issues. Also increase rep pen to 1.1
|
| 38 |
and run temps 1.1 to 2.2. I would suggest using this specific model for creative use only or limited general usage.
|
| 39 |
|
|
|
|
| 30 |
|
| 31 |
Longer, more detailed prompts may "contain" the model's output length somewhat.
|
| 32 |
|
| 33 |
+
Also with the 128k/256k you may need to stop the model's generation AND/OR For 128k/256k version I suggest you state clearly the "length of output" and/or set a hard length output limit.
|
| 34 |
|
| 35 |
IE: You ask for a scene of 1000-2000 words, and it may produce multiple scenes (in sequence!) of 1000-2000 words EACH.
|
| 36 |
|
| 37 |
+
OR
|
| 38 |
+
|
| 39 |
+
You ask for 2000 words, and you get 3k (output) in 64K, 5K in 128k and 12k in 256K versions.
|
| 40 |
+
|
| 41 |
For the 256k context version, keep prompts as clear as possible otherwise the model will have issues. Also increase rep pen to 1.1
|
| 42 |
and run temps 1.1 to 2.2. I would suggest using this specific model for creative use only or limited general usage.
|
| 43 |
|