Update README.md
Browse files
README.md
CHANGED
|
@@ -5,4 +5,7 @@ Otw : IQ2_XS
|
|
| 5 |
|
| 6 |
I recommand you folks to try this model, because it's quite an efficient merge of Miqu, WinterGoddess, AuroraNights, and XWin.
|
| 7 |
|
| 8 |
-
The Theta Rope 1,000,000 of Miqu, hence the 32k context, is functional up to 16k accordingly to my tests, and probably above (I need a smaller quant to test, which is otw).
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
|
| 6 |
I recommand you folks to try this model, because it's quite an efficient merge of Miqu, WinterGoddess, AuroraNights, and XWin.
|
| 7 |
|
| 8 |
+
The Theta Rope 1,000,000 of Miqu, hence the 32k context, is functional up to 16k accordingly to my tests, and probably above (I need a smaller quant to test, which is otw).
|
| 9 |
+
|
| 10 |
+
To use it with a quantized KV cache for a higher context, here's KoboldCPP Frankenstein version with several different KV cache quantization lebels to chose from :
|
| 11 |
+
https://github.com/Nexesenex/kobold.cpp/releases
|