How to use? I just get errors

[Disonantemus]

I was able to compile, but ...

Models downloaded and a wav test file:

$ du -ah
2.2G	./mistralai_Voxtral-Mini-3B-2507-IQ4_XS.gguf
683M	./mmproj-Voxtral-Mini-3B-2507-Q8_0.gguf
2.4G	./mistralai_Voxtral-Mini-3B-2507-Q4_K_M.gguf
5.1M	./1.wav
5.1G	.

Using only test file: doesn't find previously downloaded models:

$ voxtral-cli 1.wav
Select model (↑/↓ + Enter):A2TB/Voxtral/models:
  ▶ q4 - Mini Q4 (2.3GB) - Fast, good quality
    q8 - Mini Q8 (4.0GB) - Better quality
    bf16 - Mini BF16 (7.5GB) - Full precision
    small-q4 - Small Q4 (14GB) - Best quality, slower
    small-q8 - Small Q8 (25GB) - Maximum quality
Selected: q4
Error: Model not found: models/Voxtral-Mini-3B-2507-Q4_K_M.gguf
Download models from HuggingFace:
  https://huggingface.co/bartowski/mistralai_Voxtral-Mini-3B-2507-GGUF

Adding model flag: ask for projector

$ voxtral-cli -m mistralai_Voxtral-Mini-3B-2507-Q4_K_M.gguf 1.wav
Error: Multimodal projector not found: models/mmproj-Voxtral-Mini-3B-2507-Q8_0.gguf

Adding projector file (flag not --help or README.md): still don't find it

$ voxtral-cli -m mistralai_Voxtral-Mini-3B-2507-Q4_K_M.gguf --mmproj mmproj-Voxtral-Mini-3B-2507-Q8_0.gguf 1.wav
Error: Multimodal projector not found: models/mmproj-Voxtral-Mini-3B-2507-Q8_0.gguf

Going one level up in folder tree: surprisingly doesn't show "projector not found", but some strange function error:

$ cd ..
$ voxtral-cli -m models/mistralai_Voxtral-Mini-3B-2507-Q4_K_M.gguf --mmproj models/mmproj-Voxtral-Mini-3B-2507-Q8_0.gguf models/1.wav
Loading model...
llama_model_load_from_file_impl: no backends are loaded. hint: use ggml_backend_load() or ggml_backend_load_all() to load a backend before calling this function
Error: failed to load model

• How it is used this tool?

  I can do transcription with llama-mtmd-cli, but not this, what I'm doing wrong?

System

distro: Arch Linux x86_64
kernel: 6.12.33-1-lts
shell: bash 5.3.9
term: foot
cpu: Intel i7-4790 (8) @ 3.600GHz
gpu: NVIDIA GeForce GTX 1660 SUPER