Update README.md

README.md

@@ -6,6 +6,16 @@ tags:
 ---
 ![demo](https://thireus.com/AI/Thireus_Vicuna13B-v1.1-8bit-128g_08.png)
 
+Q. Why quantized in 8bit instead of 4bit?
+A. In theory, a 8bit quantized model should provide slightly better perplexity (maybe not noticeable - To Be Evaluated...) over a 4bit quatized version. If your available GPU VRAM is over 15GB you may want to try this out.
+Note that quatization in 8bit does not mean loading the model in 8bit precision. Loading your model in 8bit precision (--load-in-8bit) definitely comes with a non-linear quality (perplexity) degradation.
+
+Refs:
+- https://github.com/ggerganov/llama.cpp/pull/951
+- https://news.ycombinator.com/item?id=35148542
+- https://arxiv.org/abs/2105.03536
+- https://github.com/IST-DASLab/gptq
+
 **This model is a 8bit quantization of Vicuna 13B.** 
 - 13B parameters
 - Group size: 128