Naphula/Updated_Settings

1## Recommended Settings2> In particular temp 1 and topnsigma 1.25 seems to help a lot with improving quality but sometimes you need to reswipe.3> 4> Some models are more sensitive to Rep Pen settings, 1.12 is another, you may need to adjust higher for smaller models (like 1.4 if overcooked) or lower for smarter ones.5> 6> Not sure about adaptive_p that is a newer setting added to kobold7 8<pre>9Ban em dashes and ellipses (or set to -50 logit bias) to reduce slop:10 11—||$||...||$||…12</pre>13 14**(bolded kobold non-defaults)**15- **Temp 1.0**16- **TopNSigma 1.25**17- **Min-P 0.1**18- **Repetition Penalty 1.08**19- **Top-P 1.0**20- Top-K 10021- Top-A 022- Typical Sampling 123- Tail-Free Sampling 124- Presence Penalty 025- Sampler Seed -126- Rp.Range 36027- Rp.Slope 0.728- Smoothing Factor 029- Smoothing Curve 130- DynaTemp 031- Mirostat Mode OFF ("2" enhances creativity but also errors)32- Mirostat Tau 533- Mirostat Eta 0.134- **DRY Multiplier 0.8**35- DRY Base 1.7536- DRY A.Len 237- DRY L.Len 32038- **XTC Threshold 0.1**39- **XTC Probability 0.08** (The "Anti-Cliche" Shield)40- **DynaTemp ON** (The "Poor Man's Fading Mirostat")41- **Minimum Temperature 0.65**42- **Maximum Temperature 1.35**43- **Temperature 1.0**44- **DynaTemp-Range 0.35**45- **DynaTemp-Exponent 1**46 47## New Settings48I have not tested these yet but some users on reddit have reported the following settings work extremely well49- min_p: 0.0550- adaptive_target: 0.9 (down to 0.65 for more creative)51- adaptive_decay: 0.9952- repetition_penalty_range: 053- temp and topnsigma: 0.75