feat(docker): Include an optional Docker image for AIPlayerbot & LLM support

[dekuplant]

https://www.youtube.com/watch?v=pa-kLvQEMpk
Nirv posted a cool video about cMangos running with LLM AI Playerbots that react with proper context of speech. Would this be possible to implement with your awesome docker cMangos project?

[Byloth]

Hi, @dekuplant!
AMAZING! I had no idea this was even possible! 🤯

---

Now... Honestly?

The quick-n-dirty solution is to start from 39:13 and follow the video step by step.
Set up and run by yourself KoboldCPP, download the 10GB model and configure the aiplayerbot.conf by adding the line shown in the video using your machine's IP.

ipconfig  # This will print your machine's IP.

From there, it should work the same way, regardless of whether you're running CMaNGOS inside a Docker container.

---

On the other hand, I'll look into a cleaner way to integrate this directly into the CMaNGOS project and provide it as an out-of-the-box solution, including LLM integration.

It might take me some time though... 🥲

[dekuplant]

Awesome! I'll give it a go, thank you.

BTW how do I increase the playerbot count with your CMANGOS Docker setup? Somewhere in the SQL db? Or in a config file? I'm excited to see the out-of-box solution!! :)

EDIT:
Default seems to be ~49 bots online.
aiplayerbot.conf that I sourced from
https://github.com/cmangos/playerbots/blob/master/playerbot/aiplayerbot.conf.dist.in.tbc
Do I have to increase these values? :S

# Random bot count
AiPlayerbot.MinRandomBots = 1000
AiPlayerbot.MaxRandomBots = 1000
AiPlayerbot.RandomBotAccountCount = 200

[dekuplant]

Got the LLM bots working! Usingkoboldcpp-linux-x64-nocuda running airoboros-l2-7B-gpt4-2.0.Q6_K.gguf.
Here is my kobold custom start script that helped cmangos-docker communicate with kobold. I'm a linux and LLM noob so I'm not sure if this is super safe, but only using this on LAN with friends. Is there a better way to do this?

#!/bin/bash
./koboldcpp-linux-x64-nocuda --host 0.0.0.0 --port 5001 --model airoboros-l2-7B-gpt4-2.0.Q6_K.gguf --threads 8

I used the aiplayerbot.conf from github.com/cmangos/playerbots/blob/master/playerbot/aiplayerbot.conf.dist.in.tbc
I had to comment out the header like this #[AiPlayerbotConf]
Scrolled down to "# LLM values" and enabled LLM

AiPlayerbot.LLMEnabled = 1
# The api endpoint that should be called for chat generation.
AiPlayerbot.LLMApiEndpoint = http://localhost:5001/api/v1/generate
# The api key needed to access the endpoint.
# AiPlayerbot.LLMApiKey =
# The default json to send to the endpoint.
AiPlayerbot.LLMApiJson = {"max_length": 100, "prompt": "[<pre prompt>]<context> <prompt> <post prompt>"}

Time to have some fun now!

On the topic of me being a Linux noob... do you think there is anything I can improve with my system or optimize with kobold and docker cmangos? Thanks again for the awesome cmangos docker project.

           OS: CachyOS x86_64
         Host: ROG Zephyrus G14 (GA403UU)
       Kernel: Linux 6.16.8-2-cachyos
    Packages: 1610 (pacman), 9 (flatpak-system), 18 (flatpak-user)
       Shell: fish 4.0.8
     Display: 2880x1800 @ 120 Hz (14")
          DE: KDE Plasma 6.4.5
          WM: KWin (Wayland)
    WM Theme: Breeze
       Theme: Windows (EmeraldSmooth), Breeze-Dark
    Terminal: konsole 25.8.1
         CPU: AMD Ryzen 7 8845HS (16) @ 5.14 GHz
         GPU: NVIDIA RTX 4050 Max-Q / AMD Radeon 780M
      Memory: 12.98 GiB / 14.90 GiB (87%) (Soldered RAM, can't afford 32GB model)
        Swap: 12.10 GiB / 14.90 GiB (81%)
        Disk: 732.56 GiB / 1.00 TiB (71%) - btrfs
     Battery: 100% [AC Connected]

[Byloth]

Great! Amazing! 🚀

Just to point out a small detail about this...

I used the aiplayerbot.conf from github.com/cmangos/playerbots/blob/master/playerbot/aiplayerbot.conf.dist.in.tbc I had to comment out the header like this #[AiPlayerbotConf] Scrolled down to "# LLM values" and enabled LLM

AiPlayerbot.LLMEnabled = 1
# The api endpoint that should be called for chat generation.
AiPlayerbot.LLMApiEndpoint = http://localhost:5001/api/v1/generate
# The api key needed to access the endpoint.
# AiPlayerbot.LLMApiKey =
# The default json to send to the endpoint.
AiPlayerbot.LLMApiJson = {"max_length": 100, "prompt": "[<pre prompt>]<context> <prompt> <post prompt>"}

---

As for how the CMaNGOS Docker configuration works (details here), you only need to add the properties you want to change, rather than copying the entire set of rules.

That’s why you don't need to include [AiPlayerbotConf].

Your runner/config/aiplayerbot.conf can simply be:

###############################################
#  CMaNGOS AI Playerbot - Configuration file  #
###############################################

AiPlayerbot.Enabled = 1

AiPlayerbot.LLMEnabled = 1
AiPlayerbot.LLMApiEndpoint = http://localhost:5001/api/v1/generate
AiPlayerbot.LLMApiJson = {"max_length": 100, "prompt": "[<pre prompt>]<context> <prompt> <post prompt>"}

---

Other than that, I don't really have further suggestions for improving or optimizing the LLM aspect within CMaNGOS Docker...
It's quite a new topic for me too.

I’ll go ahead and rename this issue to "Include an optional Docker image for AIPlayerbot & LLM support".