Class LLMPayloadPostprocessor

Namespace

GrindFest.Editor

Assembly

GrindFest.dll

Keeps the experimental local-LLM payload out of shipped builds.

Nothing in the game uses LLMUnity — only LLMUnity's own sample scenes and one test scene do. What used to happen: LLM.cs downloaded llamafile-0.8.6.exe (~35 MB) plus two APE loader binaries into StreamingAssets on every editor load, so every build shipped a local llama.cpp server the game never started. A build later, someone noticed ~165 MB of it (plus .gguf test models) inside a player build.

The download is disabled now (see the commented-out [InitializeOnLoadMethod] in LLM.cs) and the binaries were deleted from the repo. This postprocessor is the belt-and-braces guard so they cannot creep back in silently.

If the local LLM is ever revived it needs a current llamafile build (0.8.6 is from 2024) or a runtime downloader — and that payload should be fetched next to the saves (persistentDataPath), not committed into StreamingAssets.

public class LLMPayloadPostprocessor : IPostprocessBuildWithReport, IOrderedCallback

Inheritance

object

LLMPayloadPostprocessor

Implements

IPostprocessBuildWithReport

IOrderedCallback

Properties

callbackOrder

Returns a numeric value that determines the order in which the build callback is invoked.

Methods

OnPostprocessBuild(BuildReport)

Implement this function to receive a callback after the build is complete.