Update README.md

README.md

@@ -50,17 +50,20 @@ This version will follow your instructions better, and output will be more nuanc
 
 Roughly, if you are using Q4KM of V2, you will get close to Q5KM of V1 performance.
 
+The model was also remastered from "float16" to "bfloat16" to capture more nuance from the source models.
+
 There are also links to new guides (30+ pages) below to get the most out of this model, regardless of use case(s) you want to use it for.
 
 Templates to use with this model have also been added.
 
 For Grand Horror 16.5B (V1 and/or V2), you may want to check out "generational steering" (section : "Highest Quality Settings..." below.)
 
-I updated the examples with the "v2" versions so you can see some of the contrasts.
+I updated the examples with the "v2" versions ("v1" also shown) so you can see some of the contrasts.
 
-3 new (v2 only) examples showing the full "horror" this model can output are also at the bottom of this page.
+3 new (v2 only) examples showing the full "horror" this model can output are also at the bottom of this page including "dynamic temp".
 
-If you are using V1, you should immediately download the newest quants for maximum performance.
+If you are using V1, you should immediately download the newest quants for maximum performance especially if you are using it
+for role play and/or multi-turn chat.
 
 Note that performance upgrades (via LLAMACPP and requanting the model) affect ALL QUANTS.