Running mamba seems to run single threaded on NixOs with AMD CPU

I am running cargo run --release --example mamba -- prompt "Mamba is the" on amd cpu without cuda on a fresh clone of candle. It seems to only use single threading.

Is this expected? I tried setting RAYON_NUM_THREADS but still single threaded. I installed rust toolchain using nixos and ensured I have all the required llvm dependencies, also using nixos packages.

Start by reproducing `cargo run --release --example mamba -- prompt "Mamba is the"` on an AMD CPU under NixOS, with and without `RAYON_NUM_THREADS`. Determine whether the observed single-threaded behavior is expected, then use the result to define whether the issue needs documentation or an actionable fix.

Written by the indexing model from the issue text.