Update README.md

README.md

@@ -3,13 +3,20 @@ license: other
 ---
 Experimental: Created using an unofficial and unsupported method. I have no metrics on how this performs against 13b and I'm not planning on gathering any at this point. Still has weak spots that need work.
 
-A further instruction fine tune of https://huggingface.co/nkpz/llama2-22b-blocktriangular-alpaca
+https://huggingface.co/nkpz/llama2-22b-blocktriangular-alpaca with further instruction and chat fine tuning
 
 First, I trained it on an epoch of https://huggingface.co/datasets/Adapting/empathetic_dialogues_v2 to give it a decent base knowledge of a casual chat style. The result was conversational, but not very smart.
 
 Then I trained it on an epoch of https://huggingface.co/datasets/vicgalle/alpaca-gpt4 and landed here, a model that is capable of chatting but very focused on following instructions.
 
-This strongly prefers the alpaca prompt format and will try to autocomplete it if you don't provide it. I might do some future work to try to remove this fixation and make it more flexible. Also would like to filter the rows with phrases "AI assistant" and "virtual assistant" from all future runs.
+**Future Ideas**
+- This strongly prefers the alpaca prompt format and will try to autocomplete it if you don't provide it. I'd like to work on removing this fixation and making it more flexible. 
+- Also would like to filter the rows with phrases "AI assistant" and "virtual assistant" from all future runs.
+- Thinking it might also help to do a short run on a dataset focused on character impersonation
+
+
+**Prompting**
+
 
 Standard prompt format examples: