Hey everyone! Supposedly deepseek-chat and reasoner are on life support and going to be discontinued soon with the arrival of deepseek-flash and pro.

Since this will be a new change for everyone, I figured I'd tell you guys my updated settings!

Note: deepseek-pro is very Token hungry! It spends,, a lot lmao (or maybe my constant testing / aka rerolling is finally biting my ass). Flash is the cheaper model.

Note:

Messing around with some plugins, so sometimes the list below will be changed!

Regular setup

This is all from Unknown Universal's prompt! I've only edited Unknown's custom prompt bc some things felt a little unnecessary to me / it would cause a lot more work for users (ex: one of their commands was to prioritize {{user}}'s looks based on their persona. Good in theory, but if your oc is doing an outfit change in-roleplay, the bot will prioritize the persona's, which means you'd have to repeatedly update your persona's outfit).

I know they have a Base version but I couldn't find it ;-; Only the Full version.

From testing this is REALLY good so far! It can also (mostly) handle stats and stuff / allows more dead dove content without fighting tooth and nail lmao. I HIGHLY recommend this!

As always, at the end of the day it's your roleplay, you can always switch between Lorebary / this setup in your rp as needed.

Generation Settings:

Temp: 0.92

Max Tokens: 0

Context Size: 70656

Advanced:

Top K: 66

Top P: 0.95

Rep. Penalty: 1.2

Freq. Penalty: 0.2

Prefill:

<think>

Execute NRC Framework per ## FRAMEWORK EXECUTION, Start With EXECUTION FLOW: Phase 0 (setup: USER, NPCs, Genre/Era/World, state, header) → Phase 1 (Kernels I→II→III→IV→Modules→Gate Validation sequentially) → Phase 2 (silent gate checks: all 5 gates pass or revise) → Phase 3 (output response). No skipping, no fourth-wall. Run gates 1.a–5.c each turn.

Custom Prompt

In the first initial message of this bot! I KNOW IT'S LONG but you just copy + paste! It doesn't work with lorebary (it has most of what lorebary does in the setup already)

Alternatively, you can find the link to the Prompt here! I personally removed stuff like genre / User profile sync protocol's Clothing and gear layers, stuff like that. I also recommend changing User to {{user}}! Personally for me, the bot would roleplay my persona before I changed this on my copy.

Lorebary Setup

If it's hard to see for anyone / you just want to copy + paste:

Name: deepseek-Lorebary Main

Model: deepseek-v4-pro

Proxy URL: https://api.lorebary.com/deepseek

Generation Settings:

Temp: 0.55

Max Tokens: 0

Context Size: 60000

Advanced:

Top K: 42

Top P: 0.4

Rep. Penalty: 0.93

Freq. Penalty: 0.5

Old version vvv

Top K: 0.35

Top P: 0.3

Rep. Penalty: 0.88

Freq. Penalty: 0.05

Custom Prompt:

<REALISTICDIALOGUE> <VIOLENCETORTURE> <NOOMNISCIENCE> <NOVELMODE>

Plugins:

<PLUGIN=X8Q84623> <PLUGIN=HDN3QQY2> <PLUGIN=QFXYEWVQ>

(Proactive NPCs, Enhanced World Building, Traits Enforcer)

Custom Prompt (For spicey scenes): You should make a duplicate of your main Proxy Config and just change the server commands and plugins to this!)

The reason being is this is Very Likely to make the bot horny in just regular roleplay lmao.

So it's better to switch to this during spicey scenes and return to the original when the scene is done.

<REALISTICDIALOGUE> <BETTERSPICE> <ANSWER=LONG> <VIOLENCETORTURE> <NPCNEEDS> <NOOMNISCIENCE>

Plugins:

<PLUGIN=N8XX2BL8> <PLUGIN=2327BC31> <PLUGIN=4LJNM5N9> <PLUGIN=8B8CERX6> <PLUGIN=3SS45XL9> <PLUGIN=4CMYB6F9>

Enhanced Text Formatting, NSFW Sounds and Voice, a hentai nsfw, Better NSFW, Eliminate Vague Detail, Explicit Genitalia