Files
zeekayandHanzo Dev 6a363f7ee3 merge: upstream main ddf1b879 and the aarch64 repack branch 35d7ae7c
Three-way per file through the rename: base 39355c6c91, theirs the upstream
merge of main with 35d7ae7c (via 7ef6a694d), ours origin/main.

Lands the barrier thread pool, runtime CPU feature dispatch, aarch64 and x86
repacked quantized kernels (indexed_gemv, gemv_fused_shared_lhs), quantized
embedding on CPU/Metal/CUDA, paged flash-attn kernels, cutile fused MoE, the
in-place op aliasing API, parking_lot storage locks, Metal buffer builders and
debug labels, safetensors 0.8, hf-hub 1.0, tokenizers 0.23.

Ours keeps ROCm, Vulkan, wgpu, the IQ/MXFP4/NVFP4/TQ dtypes, and the i-quant
weight/activation block split in matmul. Our backends ride InplaceOpN and the
quantized embedding; Metal get_rows gains the IQ kernels. ug leaves, as upstream.

Co-authored-by: Hanzo Dev <[email protected]>
2026-09-15 15:52:28 -07:00
..

hanzo-ml-voxtral: speech recognition

An implementation of Voxtral speech recognition using hanzo-ml.

Running the example

Run with the cuda feature for GPU acceleration:

cargo run --example voxtral --features tekken,symphonia,rubato,cuda --release
# you may also add the `cudnn` feature for extra performance
# cargo run --example voxtral --features tekken,symphonia,rubato,cuda,cudnn --release

Remove the cuda feature to run on the CPU instead:

cargo run --example voxtral --features tekken,symphonia,rubato --release
# or pass the `--cpu` flag to force CPU usage
# cargo run --example voxtral --features tekken,symphonia,rubato,cuda --release -- --cpu

Command line options

  • --cpu: Run on CPU rather than on GPU (default: false, uses GPU if available)
  • --input: Audio file path in wav format. If not provided, a sample file is automatically downloaded from the hub.
  • --model-id: Model to use (default: mistralai/Voxtral-Mini-3B-2507)