How to export to GGUF or Safetensors?

[magnuspri]

In the youtube video: "LLMs Are Databases" (https://www.youtube.com/watch?v=8Ppw8254nLI&t=295s). Hay mentions that there is a way to export the vindex to safetensors or GGUF that can work with model providers. I tried doing it, but I'm stuck. Like I tried this command: COMPILE CURRENT INTO VINDEX "/tmp/test";

Just like in the video, and considering this respons: Compiled gemma3-4b.vindex → /tmp/.test.tmp.95072

It seem to have done something. But then where are the folder with the GGUF file or safe tensors? Or should I make use another command for it. Have tried talking with an llm back and forth. Been searching the documentation manually. Been experimenting till my hair got grey. But I just can't figure this one out.

[magnuspri]

Ok I finally found the answer. Ironically in the documentation:
COMPILE CURRENT INTO MODEL "edited/" FORMAT safetensors;

But now I got this error:
Error: Execution error: failed to load model weights: parse error: vindex is quantised (q4k). load_model_weights only handles float weights. Call VectorIndex::load_attn_kquant + load_interleaved_kquant on the loaded VectorIndex instead.

I don't know. Is it this a bug, or was it because I earlier used this?
// Q4_K/Q6_K inline (Ollama-compatible, smallest disk footprint)
larql extract google/gemma-3-4b-it -o gemma3-4b.vindex --quant q4k

[jterrano]

Maybe you need to extract using WITH ALL;

https://github.com/chrishayuk/larql/blob/main/docs/lql-guide.md

-- Full (~10 GB f16 / ~18 GB f32, enables COMPILE for recompilation)
EXTRACT MODEL "google/gemma-3-4b-it" INTO "gemma3-4b.vindex" WITH ALL;

Hope this help you,