Update README.md

README.md

@@ -15,17 +15,35 @@ license: apache-2.0
     margin-bottom: 0.5em;
   }
   h1 {
-    font-size: 2em;
+    font-size: 3em;
   }
   h2 {
-    font-size: 1.3em;
+    font-size: 1.6em;
   }
-  p, ul, ol, summary {
+  p, ul, ol, strong, summary {
     font-size: 1.1em;
   }
+  .line-spaceless {
+    line-height: 1;
+    margin: 0;
+    padding: 0;
+  }
+  .half-space {
+    line-height: 0.5em;
+    margin-bottom: 0.25em;
+  }
+  
+  .text-center {
+      text-align: center;
+  }
+
+  .tiny-text {
+    font-size: 0.8em;
+  }
 </style>
 
-<h1>Mistral Nemo 12B Celeste V1.9</h1>
+<h1 class="line-spaceless text-center">Celeste V1.9</h1>
+<p class="half-space text-center tiny-text">Based on Mistral Nemo 12B</p>
 <h2 style="color: red; font-weight: bold;">Read the Usage Tips Below! Use ChatML.</h2><h2>Join <a href="https://discord.gg/EWzsFddYAd">our Discord</a> for testing newer versions and news! We are also on KoboldAI</h2>
 
 <img src="https://cdn-uploads.huggingface.co/production/uploads/630cf5d14ca0a22768bbe10c/QcU3xEgVu18jeFtMFxIw-.webp" alt="" width="800"/>
@@ -75,10 +93,10 @@ If one doesn't work, try the other.
 
 I usually start the first few messages with Stable and see how it goes. If it falls into repetition I switch to Creative. But you can also just use either the whole way through, creative may need a few swipes from time to time.
 
-<h3>Stable</h3>
+<strong>> Stable</strong>
 <img src="https://cdn-uploads.huggingface.co/production/uploads/630cf5d14ca0a22768bbe10c/1m18WnuomY8jEZTA87Iun.png" alt="" width="400"/>
 
-<h3>Creative</h3>
+<strong>> Creative</strong>
 <img src="https://cdn-uploads.huggingface.co/production/uploads/630cf5d14ca0a22768bbe10c/DaL2hWZst0yW34CYK4df8.png" alt="" width="400"/>
 
 Don't shy away from experimenting after you get a feel for the model though.
@@ -96,16 +114,17 @@ Currently, your role is {{char}}, described in detail below. As {{char}}, contin
 
 <h2>Story Writing</h2>
 
-Adding this system prompt will likely increase the humanness of the prose as we trained system prompts. You can also change it to NSFW, but you should try both regardless of whether you are writing NSFW or not.<br>
-You should also force the assistant reply to start with a `*` due to how we trained on human stories.
-
-System Prompt: `You are a short story writer. Write a story based on prompt provided by user below. Mode: SFW`
+**Adding the below system prompt will likely increase the humanness of the prose** as we trained system prompts. You can also change it to NSFW, but you should try SFW regardless of whether you are writing NSFW or not.<br>
+You should also try forcing the assistant reply to start with a `*` due to how we trained on human stories.
 
+```
+You are a short story writer. Write a story based on prompt provided by user below. Mode: SFW`
+```
 If your first message is using human-like prose, Celeste will copy it in the next messages, check out the Showcase below.
 
 <h2>Swipes</h2>
 
-**Important tip** swipe 2-3 times if you dont like a response. This model gives wildly differing swipes.
+**Important: swipe 2-3 times if you dont like a response** This model gives wildly differing swipes.
 
 <h2>OOC Steering</h2>
 
@@ -113,7 +132,7 @@ If your first message is using human-like prose, Celeste will copy it in the nex
 
 <h2>"Dead Dove"</h2>
 
-For character cards with persistent motivations throughout the story, use world books [tutorial here](https://huggingface.co/nothingiisreal/how-to-use-ST-worldinfo)
+For character cards with persistent motivations throughout the story, use world books at low depth [tutorial here](https://huggingface.co/nothingiisreal/how-to-use-ST-worldinfo)
 
 <h2>Fewshot</h2>
 
@@ -132,7 +151,7 @@ If you want only SFW and are having troubles, there is probably some system prom
 <h2>Refusals</h2>
 
 As said, if instruct refusal (very rare,) prefill 2-3 words. **Refusal of romantic advances (which almost never happens on 12B,) are realistic and we think is good. Prefill if you don't like.** <br>
-<br>
+
 <h2>Mistral Context</h2>
 
 While trained on 8K, the model should be able to inherit longer context from Mistral 12B. Should be at minimum 16K.