Update README.md
Browse files
README.md
CHANGED
|
@@ -79,6 +79,33 @@ The mountain moved:
|
|
| 79 |
|
| 80 |
150 points better: PPL = 8.5850 +/- 0.05881 VS: BASE/ORGINAL: PPL = 8.6012 +/- 0.05900
|
| 81 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 82 |
<B>THE RESULTS ARE IN: </b>
|
| 83 |
|
| 84 |
AS per Jeb Carter, original creator of the model:
|
|
|
|
| 79 |
|
| 80 |
150 points better: PPL = 8.5850 +/- 0.05881 VS: BASE/ORGINAL: PPL = 8.6012 +/- 0.05900
|
| 81 |
|
| 82 |
+
<B>Settings: CHAT / ROLEPLAY and/or SMOOTHER operation of this model:</B>
|
| 83 |
+
|
| 84 |
+
In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;
|
| 85 |
+
|
| 86 |
+
Set the "Smoothing_factor" to 1.5 to 2.5
|
| 87 |
+
|
| 88 |
+
: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"
|
| 89 |
+
|
| 90 |
+
: in text-generation-webui -> parameters -> lower right.
|
| 91 |
+
|
| 92 |
+
: In Silly Tavern this is called: "Smoothing"
|
| 93 |
+
|
| 94 |
+
|
| 95 |
+
NOTE: For "text-generation-webui"
|
| 96 |
+
|
| 97 |
+
-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)
|
| 98 |
+
|
| 99 |
+
Source versions (and config files) of my models are here:
|
| 100 |
+
|
| 101 |
+
https://huggingface.co/collections/DavidAU/d-au-source-files-for-gguf-exl2-awq-gptq-hqq-etc-etc-66b55cb8ba25f914cbf210be
|
| 102 |
+
|
| 103 |
+
OTHER OPTIONS:
|
| 104 |
+
|
| 105 |
+
- Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor")
|
| 106 |
+
|
| 107 |
+
- If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.
|
| 108 |
+
|
| 109 |
<B>THE RESULTS ARE IN: </b>
|
| 110 |
|
| 111 |
AS per Jeb Carter, original creator of the model:
|