SA-MP 0.3DL compatibility

[AlexanderFreeman]

Hello! Thank you for such a useful plugin. The plugin loads correctly in 0.3DL, but could you clarify - was it created to work in this version, or are there possible bugs / it is not tested on this version? I haven't find any information on the wiki/in the description. Thank you.

[AlexanderFreeman]

Another question - does it work with groq.com using only LLaMA model or I can use any model from groq.com?

[SimoSbara]

Hi, thanks you for the feedback.

I will answer you in order:

1. This plugin works on both 0.3.7 and 0.3.DL because it is designed to be version independent, I will explicit this in the description.
2. With groq.com you can use other chat bot models even if it is "LLAMA" type, maybe I should change the bot type name to "GROQ".

In the near future, I was thinking to implement a more flexible architecture where you don't need to specify the bot type but just the URL and the associated parameters (like model name, API KEY, etc...), that are different depending on the provider.

I hope this helped you.

[AlexanderFreeman]

Thank you for answer! Last question: does the plugin support self-hosted LLM now?

[SimoSbara]

Thank you for answer! Last question: does the plugin support self-hosted LLM now?

Theoretically yes, but you need to also have a self-hosted endpoint for API call that is compatible with one of the implemented bot types.

However I need to implement a function to set a custom url like: SetCustomBotURL(url).
Also you need to emulate the url/request/response format of one of the bot types, the plugin wouldn't notice.

BUT, there is another possibility: the use of shader memory communication between processes another: https://github.com/SimoSbara/samp-shared-memory

This is a plugin that I made a while ago, and this could an interesting use case.

The idea is:
The samp server process creates a shared memory segment that can be accessed by other processes: in your case would be a program in C++ (for example) that is running the LLM locally and all the requestes between server and LLM would be exchanged inside this shared memory segment, it is a faster solution in terms of performance but it is tricky to achieve