Update README.md

README.md

@@ -14,6 +14,8 @@ language:
 # Pantheon-RP-1.5-12b-Nemo
 Welcome to the next iteration of my Pantheon model series, in which I strive to introduce a whole collection of personas that can be summoned with a simple activation phrase. The huge variety in personalities introduced also serve to enhance the general roleplay experience.
 
+**Disclaimer:** Despite my goal to create the perfect Pantheon finetune I still feel like I've been unable to shave some rougher edges off the Nemo base model. Rather then continue to bash my head against the wall (and as a result not release anything) I've instead decided to release my finest attempt so far as it should already surpass my 1.0 release.
+
 ## Model details
 This time around I went for a multi-stage finetuning process as Mistral Nemo was proving to be somewhat stubborn without a solid base training being performed first;
 
@@ -25,21 +27,22 @@ This time around I went for a multi-stage finetuning process as Mistral Nemo was
 
 ## Inference
 
-Despite Mistral's insistence to use lower temperatures I continue to recommend the following settings for inference:
+Nemo is a somewhat strange model when it comes to temperatures so I highly encourage you to experiment to see which works best.
 ```
-"temperature": 1.0,
+"temperature": 0.3-1.0,
 "repetition_penalty": 1.05,
 "top_p": 0.95
 "top_k": 40
 "min_p": 0.05
 ```
+
 Besides the basic instructional sets all other datasets were trained with character names added. Enable this at all times for an optimal experience.
 
 **Note:** My previous release suffered from a tendency to generate shorter roleplay responses, which I now believe has been mostly resolved.
 
 ## General Roleplay
 
-The second finetune was focused solely on an asterisk-style, no quotes for speech roleplay style (aka Markdown), as that is the style my Pantheon Roleplay dataset uses.
+The second finetune was focused solely on an asterisk-style, no quotes for speech roleplay style (aka Markdown), as that is the style my Pantheon Roleplay dataset uses. I expect there to be a bias inside the model itself towards responding in this style.
 
 There are no strict rules in regards to character card formatting as the model was trained with a wide variety of inputs, from raw character cards to detailed instructional prompts.