Major improvements for V1, V2 and related parameters

Also merged V3 with V2, and added ASSISTANT MODE for V2

README.md

@@ -138,7 +138,7 @@ Avoid with strict settings.
 Values below 0.96 are not recommended.
 </details>
 <details><summary>Adaptive-P (new):</summary>
-Does not work with high Temperature; Avoid at any case.
+Can't work with high Temperature; Avoid at any case.
 </details>
 <details><summary>DRY:</summary>
 Avoid in any case.
@@ -157,16 +157,16 @@ Probability - the chance for Threshold to cutoff the desired most likely tokens.
 <details><summary>V1 **-CREATIVE-BALANCED-**</summary>
 <img src="https://gitlab.com/Azuro721/trueperfect-ai/-/raw/main/PERF1.png" style="float:right; width:200px; height:300px; padding:10px;">
 
-Very fine with very good creativity, level of detail, emotional connections, "surprising" outcomes and descriptions.
+Very fine with very good creativity, level of detail (might skip some due to higher Top-A compared to V2), emotional connections, "surprising" outcomes and descriptions.
 
-In some cases where Repetition Penalty 1.12082 outputs overly descriptive results, try out 1.121, it might provide better results and fix such issues; but in most cases adjusting the typos and logical inconsistency from instructions or character cards will fix such issues, so 1.121 is generally not recommended.
+<details><summary>Optional Adjustments</summary>
+If Repetition Penalty 1.12082 outputs overly descriptive results, try improving the descriptions for character cards or altering the instructions, it will fix most of such issues; otherwise (if nothing helps), use 1.121.
 
-Might confuse things like character names, species, User actions and etc (due to high Temperature alongside with not low enough Top-A). **In such cases lower Repetition Penalty to 1.02612; this will fix most issues and alter the outputs in a more focused and less creative way, with attention to finer details**.
-
-In this case, Top-K 206 with Repetition Penalty 1.02612 will improve attention to finer details with less attention to creative things and slower progression. **Top-K 134 (with Repetition Penalty 1.02612) is generally more preferred due to being more subtle, with faster progression (good in most cases) and better creative outcomes**.
+In some cases, tends to confuse things like character names, species (mostly from trained data), pronouns, misuse (confuse) of User's actions, improper pronouns and etc (mostly due to low-probability token picks). **In such cases, lower Repetition Penalty to 1.02612; this will noticeably reduce the frequency of such occurrences and alter the outputs in a more focused and less creative way, with attention to finer details and more subtle transitions**.
 
 In some specific models (most likely), TFS 0.9551 might improve things even further, with more attention, creativity, and performance overall; do not use if you notice overly long descriptions (extremely long after each section).
 </details>
+</details>
 <details><summary>V2 **-INSANE-DETAIL/ATTENTION-**</summary>
 <img src="https://gitlab.com/Azuro721/trueperfect-ai/-/raw/main/PERF2.png" style="float:right; width:200px; height:300px; padding:10px;">
 
@@ -176,22 +176,40 @@ OVERKILL.
 
 Maintains insane amount of details, attention, accuracy, and length: focused outputs, "surprising" outcomes and descriptions (noticeably (in some models) less compared to Top-A 0.07 but still generally good).
 
-Lower Repetition Penalty to get even more insane attention and level of detail, but sacrifice a bit of creativity, "surprising" outcomes and descriptions (especially with 1.02612).
+V2 will be more bland and less descriptive compared to V1 in favor of maximum stability and fixes related to general incoherence with character names (mostly from trained data) and improper pronouns (complete removal of very-low-probability token picks that produce rare misspellings and related issues).
 
-Repetition Penalty 1.02612 is preferred as the lowest point; will output very attentive and detailed descriptions, with other things described earlier.
+<details><summary>ASSISTANT MODE</summary>
+TFS: 0.9551
+Repetition Penalty: 1.02612
 
-In this case, Top-K 206 will increase attention even further.
+Insane for maximum accuracy for ASSISTANT-related tasks (personal assistant); Will be less creative in favor of attention.
+<details><summary>If any issues occur (too detailed with incoherence)</summary>
+Decrease TFS to 0.8413 at cost of lower accuracy and level of detail, but noticeably better stability.
+</details>
+<details><summary>Optional Adjustments (might degrade stability and accuracy)</summary>
+Might provide very good results in specific models, but generally unstable.
+<details><summary>Repetition Penalty 1.12082 / 1.121</summary>
+Will improve creativity with slight reduction of accuracy.
+</details>
+<details><summary>Temperature 4.8 with Top-K 134</summary>
+Will make outputs more lively, creative and "surprising", but might also increase chances of instabilities, such as overly-high description with sudden incoherence and overall degradation over higher amount of tokens.
+</details>
+<details><summary>Temperature 4.8 with Top-K 278 (284 / 296)</summary>
+Will make outputs more lively, creative, "surprising", descriptive and attentive, but tends to have higher chances of instability compared to Top-K134, with even faster degradation.
+</details>
+<details><summary>TFS 0.9551 (Special)</summary>
+Will improve attention to details and overall performance in many aspects.
 
-Use Top-K 134 for faster transitions and attention to more "surprising" moments.
+Special because tends to have more chances to work correctly across different models.
+</details>
+</details>
 </details>
-<details><summary>V2-Alt **-INSANE-DETAIL/ATTENTION-ASSISTANT-**</summary>
-<img src="https://gitlab.com/Azuro721/trueperfect-ai/-/raw/main/PERF3.png" style="float:right; width:200px; height:300px; padding:10px;">
 
-Insane for maximum accuracy for ASSISTANT tasks; can be used with overly complexive RP scenarios (with favor of attention). Will be less creative in favor of attention.
+Lower Repetition Penalty to get even more insane attention and level of detail, but sacrifice a bit of creativity and "surprising" outcomes (especially with 1.02612).
 
-Generally, Top-K 278 might improve attention to details, but not recommended for Temperature 2.4 and other cases, mostly due to increased chance of becoming overly descriptive, which will confuse model and force hallucinations.
+Repetition Penalty 1.02612 is preferred as the lowest point; will output very attentive and detailed descriptions, with other things described earlier.
 
-In rare cases (specific models) with overly descriptive results, decrease TFS to 0.8413 or/and Top-K to 134.
+Use Top-K 134 for faster transitions and attention to more "surprising" moments (works (mostly) only with Top-A 0.07).
 </details>
 
 To preserve the versatility, I would like to describe complete and specific sampler values below **Additional fine-tuning:**, to aim perfection for any case.
@@ -215,31 +233,26 @@ Lower temperature will be output less "exciting" and creative results (like less
 
 Higher Top-K will expand the attention to smaller details, and preserve attention to multiple simultaneous events, and also can fix smaller text-related issues (like with quotation marks, asterisks, hyphens and etc.)
 
-Top-K 278, as described earlier, might cause overly descriptive results, which will most likely cause hallucinations.
-
-Top-K 206 is the more attentive one, which fits more with assistant tasks, as it will take away some of creativity, so might not be suitable for Temperature 4.8.
-
-Top-K 134 is the middle-balanced one, with better creativity, good level of detail and fine transitions. Recommeded one for in-character actions and strong roleplay scenarios.
+Top-K 278, as described earlier, might cause overly descriptive results, which will most likely lead to incoherent results.
 
-Temperature 4.8 with Top-K 206 and TFS 0.8413 will output more creative, emotional and "surprising" results compared to lower ones.
+Top-K 206 is the more attentive one, which fits more with assistant tasks, as it will take away some of creativity, but tends to be more repetitive and might lead to incoherence.
 
-**Temperature 2.4 with Top-K 206 and TFS 0.9551 will output the best possible results as an assistant.**
-
-Temperature 2.4 with Top-K 206 and TFS 0.8413 will output more attentive results, with less variety, emotions and "surprising" moments.
-
-Temperature 1.2 with Top-K 206 and TFS 0.8413 will output slightly more attentive results, with even less variety, emotions and "surprising" moments.
+Top-K 134 is the middle-balanced one, with better creativity, good level of detail and fine transitions. Recommeded one for in-character actions and strong roleplay scenarios **NOT suitable for Top-A 0.0001**.
 
 Further experimentation with Top-K might not be possible, mostly due to logical limit for all settings combined.
 </details>
 <details><summary>Repetition Penalty:</summary>
 Base value: 1.12082, which will output more creative, emotional, varied, smart and "exciting" results. But tends to have issues with asterisks and quotation marks; similar to 1.02612, but with more creativity, less descriptions, faster pace, but prone to issues if input has logical inconsistencies or lots of typos.
 
-1.105 (untuned): specific value I found out during experimentation. Will output less "exciting" results, but fairly better compared to 1.05.
+1.105 (not fine-tuned): specific value I found out during experimentation. Will output less "exciting" results, but fairly better compared to 1.05.
 
-1.05 (untuned, not recommended): base value, which is widely used in various LLMs. Might output focused results with average creativity (better than 1.02612), but prone to be less stable compared to 1.105.
+1.05 (not fine-tuned, not recommended): base value, which is widely used in various LLMs. Might output focused results with average creativity (better than 1.02612), but prone to be less stable compared to 1.105.
 
 1.02612: very specific one; will output very descriptive, attentive and expanded results. Will try to pay attention to noticeably more things compared to other variants. Will preserve character details and much more things as events go by. Great as an **assistant.**. Great for very complex instructions, very complex character cards and complex scenes. Great attention to multiple characters.
 
+1.15 (not fine-tuned): tends to be more creative with shorter descriptions; might be better with Top-A 0.0001 and might be incoherent (might perform well on specific models).
+
+1.23 (not fine-tuned): tends to be even more creative with slight shorter descriptions; might be better with Top-A 0.0001 and tends to have less chances to be coherent, but might perform well in rare cases (with specific models).
 <details><summary>Other values (might output unstable results):</summary>
 
 Feel free to experiment with these variants, and show any good results (if stable enough to be used for at least ~6K tokens).
@@ -263,13 +276,14 @@ Base value: 0.915, which will output very attentive, consistent and stable resul
 
 0.905: pays more attention to specific details, slightly less emotions, and very close to being repetitive.
 
-0.95/0.97: very creative and unpredictable; might be used for better models, but generally less attentive.
+0.95/0.97: very creative and unpredictable; might be used for better models, but generally less attentive (might perform well on higher-quality models (12B+)).
 </details>
 <details><summary>Top-A:</summary>
-Base value: 0.07, which will output very consistent, overall stable results, with smooth transitions and very good attention to most things. Recommeded for all cases.
+Base value: 0.07, which will output very consistent, generally stable results, with smooth transitions and relatively good attention to most details. Recommeded for all cases.
+
 0.0001: will output insane amount of details, attention, accuracy and other things described in **V2**.
 <details><summary>Other values (experimental):</summary>
-0.043725: more attention to anatomy, but unstable and tends to be unpredictable. Works better with Repetition Penalty 1.02612 or **V2** but goes bad soon, after several thousand tokens.
+0.043725: more attention to anatomy, but unstable and tends to be unpredictable. Works better with Repetition Penalty 1.02612 or **V2** but degrades.
 
 0.2025: more creative, descriptive, "exciting" and emotional, but tends to skip some details. Less accurate and have lots of issues with high Temperature; not suitable to be used generally, only to get initial creative inputs.
 </details>