Update README.md

README.md

@@ -1,3 +1,32 @@
 ---
 license: mit
+datasets:
+- daswer123/sentence-danbooru-tags-dataset
 ---
+
+Qlora for openchat-3.5-1210 which allows you to convert a given model sentence into a set of danbooru tags
+format: openChat 
+train format: Alpaca
+
+This is the first time I have had this lora done and I am quite happy with the result. This qlora was trained on 4bit full model openchat-3.5-1210. with rank of 64 for 3 epochs
+There were 11k instructions in the dataset that had examples of how to convert from sentences to tags.
+
+I use a this prompt, I'm not sure if it's optimal but maybe you can do better
+```
+You must follow this rules:
+- If a character is alone, add the "solo" tag.
+- Describe physical attributes such as hair color or clothing (e.g., "pink hair," "gothic lolita fashion").
+- Describe actions of characters if any (e.g., “playing violin,” “studying”).
+- Mention setting details like location and background objects (e.g.,”music room”, “classical instruments”).
+- Keep each tag separated by a comma.
+- All words should be tags that can be tagged to danboororu, they should be as informative as possible and represent a single entity or action.
+- From the verbs, select words that can denote place
+- Always start with the number of characters and their gender (e.g., "1girl," "3boys").
+- If a person is described by a noun, e.g. "Beautiful girl", then we should distinguish as follows: 1girl, beautiful.
+- In output can be only girl(s) or boy(s) there's a human there, if the scenery can be left out.
+- Answer only on English
+
+You have to convert user input into danbooru tags
+```
+
+I use with the [openchat-3.5-1210-4.0bwp model](https://huggingface.co/LoneStriker/openchat-3.5-1210-4.0bpw-h6-exl2-2).