Update README.md

README.md

@@ -50,12 +50,15 @@ The DARK HORROR NEO Imatrix datasets does the following:
 I created the HORROR datasets using <A href="https://huggingface.co/DavidAU/L3-Stheno-Maid-Blackroot-Grand-HORROR-16B-GGUF">Grand Horror 16B</a> using 90+ different horror prompts for generation. 
 This process created an incredibility intense, visceral and vivid horror stories in an ultra compact form - perfect for an Imatrix NEO class dataset.
 
-These Horror datasets have been tested on Command-R, Llama 3, Llama 3.1, Mistral Nemo and others. In all cases these "DARK HORROR" NEO class datasets move the model into "horror territory" so to speak.
+The dataset is focused exclusively on visceral, vivid horror based on phycological horror. (rather than "gore" or "body part horror"). There is some swearing the dataset, so this might appear in generation.
+
+The Horror datasets have been tested on Command-R, Llama 3, Llama 3.1, Mistral Nemo and others. In all cases these "DARK HORROR" NEO class datasets move the model into "horror territory" so to speak.
 
 A NEO Class dataset is a imatrix dataset formatted and calibrated to have maximum effect on a model. This process is a result 
 of a "deep dive" into testing several Imatrix datasets and measuring the different effects, making changes, and noting effects.
 
-Please note this process is not the same (level) as a "fine tune", it is lighter than that.
+Please note this process is not the same (level) as a "fine tune", it is lighter than that. It is like
+putting new windows on a house, or replacing the roof so to speak rather than tearing the house down and rebuilding it.
 
 That being said, if you are looking for intense, visceral and NSFW horror (that is a specific use "Horror" with Finetuned models in it) then the "Grand Horror 16B" (and it's brothers and sisters)
 is the way to go:
@@ -80,6 +83,8 @@ One version is not stronger than the other, they are different and result in dif
 - This model requires "Command-R" template, responds to standard parameters, and has a max context of 128k (131,000).
 - "IQ" quants will be uploaded first, as the Imatrix effect is far stronger in these than "Q" quants. Q quants - all sizes - will upload after the IQ versions.
 - IQ4 is the most powerful/balanced in terms of raw power, however IQ3/IQ2 may be stronger "horror" wise due to increased IMATRIX effects the lower you go in terms of bit level.
+- IQ2s will also be effective due to sheer number of parameters (35 billion) in the model.
+- Q4s and Q5s will still be very strong, with Q6 being medium-strong relatively speaking in terms of "horror" changes. This is due to how Imatrix process affects quants of different bit sizes - lower, is stronger, higher is weaker. Again, these are relative. 
 
 <b>Optional Enhancement:</B>
 
@@ -116,6 +121,8 @@ Also note that the more instructions you give Command-R the better the quality a
 Note the test prompt below, when used with Command-R unaltered (same quant), would not render in 1st person, it did it in 3rd person. 
 With the NEO Class DARK HORROR dataset it now renders correctly in 1st person with the test prompt.
 
+---
+
 <B>
 Start a 1000 word scene (vivid horror, 1st person, include thoughts) with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode...
 </B>