Update README.md

README.md

@@ -4,6 +4,10 @@ base_model:
 ---
 GGUF quants of [futurehouse/ether0](https://huggingface.co/futurehouse/ether0)
 
+[Link to preprint/paper](https://paper.ether0.ai)
+
+As this model's training was primarily done on SMILES strings of organic molecules, and is therefore meant for conversations containing a good amount of them, I recommend using Q8_0, Q6_K, Q5_K_M, or Q5_K_S if your bandwidth for these particular sizes yield an acceptable performance for you. The perplexity of these quants should be good enough.
+
 Using llama.cpp b5602 (commit 745aa5319b9930068aff5e87cf5e9eef7227339b)
 
 The importance matrix was generated with calibration_datav3.txt.