Some embedding models do not work

[Artem-Viveritsa]

Qwen3-Embedding-0.6B-Q8_0.gguf starts and works as expected, while any other ones I've tried either give an error (LLM/LLM.swift:44: Fatal error: Unexpectedly found nil while unwrapping an Optional value) or an array with only zeros (vector).

Now I do it like this:

guard let modelURL = Bundle.main.url(forResource: "paraphrase-multilingual-mpnet-base-277M-v2-Q8_0", withExtension: "gguf") else {
    print("Model file not found")
    return
}

self.bot = LLM(from: modelURL)

Perhaps there are some initialization features?

[onestardao]

looks like the loader hit a nil, usually because the model file is missing or the file format is unsupported. this maps to ProblemMap No.16 (pre-deploy collapse).

quick checks:

1. confirm the model file exists and is readable at the URL you pass.
2. confirm the model format matches the loader (Q8 vs non-quant, folder layout, metafile names).
3. list the model dir from your app before calling LLM, or guard the optional to see which file is nil.

want a tiny Swift snippet and a one-line shell check to run locally? say "send" and i'll paste them.