Update README.md

README.md

@@ -76,7 +76,7 @@ output\.weight=iq6_k
 ```
 
 
-- Mostly iq5_ks GPU layers to minimize loss cheaply, keep it fast, and minimize the number of quantization types.
+- Mostly iq5_ks GPU layers to minimize loss cheaply, keep it fast (as iqX_ks quantizations are very fast), and minimize the number of quantization types.
 
 - iq3_ks shared experts near the beginning and end, as this seems to be where there are perplexity 'bumps.'