Models

Infermatic/MN 12B Inferor v0.0

inferor

MN 12B Inferor v0.0: Dynamic and Creative

MN 12B Inferor, also known as Mistral Nemo Inferor, is an impressive model merge created from 4 MN fine-tunes. It uses the following models:

Review

This model stands out for its remarkable creativity, often generating long, contextually adaptive paragraphs. However, it has a notable weakness in stability. To counteract this, you can control its performance by adjusting key settings such as:

  • Temperature
  • Minimum Probability (Min P)
  • Repetition Penalty

The penalty settings are particularly impactful in ensuring high-quality outputs.

Want to Try It?

Experience MN 12B Inferor v0.0 on the following platforms, both offering a 32K context window:

 

 

Recommended Settings for MN Inferor 12B v0.0

For optimal performance, here are the recommended settings:

  Setting Value
  Format ChatML
  Tokenizer Mistral Nemo
  Temperature 0.88
  Top K -1
  Top P 0.77
  Typical P 1
  Min P 0.025
  Top A 0
  Repetition Penalty 1
  Frequency Penalty 0.52
  Presence Penalty 1
  Response Tokens 333

Pro Tips: To make the model more deterministic, decrease the temperature. To avoid incomplete sentences, enable the ‘Trim incomplete sentences’ option (if using Silly Tavern). For better results, reduce the response tokens if the output seems off.

 

Are you using Silly Tavern?

Import the master settings from here: Inferor-creativity-sk

 

Why Mistral Nemo?

Mistral Nemo, a collaboration with NVIDIA, is a robust base model offering a large context window, making it ideal for complex text generation tasks. It brings enhanced creativity and flexibility to your outputs. Also brings Multilingual Capacities to the table with a Benchmark (MMLU) looking like this:

Language Score
French 62.3%
German 62.7%
Spanish 64.6%
Italian 61.3%
Portuguese 63.3%
Russian 59.2%
Chinese 59.0%
Japanese 59.0%
 From: Mistral.ai hugginface