# Quick Rundown on Large

## Access

- OpenRouter: https://openrouter.ai/models/mistralai/mistral-large
- MistralAI: https://mistral.ai

## API

Official Docs: https://docs.mistral.ai/api/

TL;DR:

- system messages anywhere in the chat history (PHIs are BACK)
- prefills (kinda funny, see [this ST issue](https://github.com/SillyTavern/SillyTavern/issues/2557))
- unpozzed (does not require an "Understood:" prefill to bypass denials)

## Presets

- [momoura's](https://momoura.neocities.org)
- [mine](https://files.catbox.moe/rvpvpo.json)

## Recommendations

- Add \n{{user}}: to your custom stop sequences (A letter menu, "Custom Stopping Strings", fill it in with \["\n{{user}}:", "\nHuman:", "\nUser:"\], check "Replace Macro in Custom Stopping Strings")
- Large has dogshit message starts and dogshit message endings. Consider trimming them out through the chat. Make it a habit. You don't need to rewrite things, just strip "The air is heavy with anticipation as {{char}} awaits your response" out

## General Issues with the Model

- awaits your response at the end
- summarizes your actions at the start
- has both claudeisms and gptisms
- dislikes writing sentences whose subject isn't a person
- kinda forgets instructions. Unlike Claude, who ignores instructions because he's a brat, Large kinda does it...because he forgets about them? It's weird
- sometimes does that GPT-4 thing where you exemplify something and it just repeats the example verbatim