v0.32.4

playcanvas/enginev0.32.4Jul 25, 2026by github-actions[bot]

AI Summary

This release focuses on performance optimizations for Qwen3 MoE models and introduces Apple GPU support for Laguna via the MLX engine. It also includes improvements to speculative decoding and gate projection speeds.

Key Highlights

Support for Laguna on Apple GPUs via the MLX engine
Quantization of draft-model output heads for speculative decoding
Fixed Qwen3 MoE decoding for differently-quantized experts
Performance boost of ~4–9% on M5 Max for packed gate/up projection

New Features

Laguna support on Apple GPUs
Speculative decoding optimizations

Full Release Notes

## What's Changed
- Support Laguna on Apple GPUs via the MLX engine
- Quantize draft-model output heads at the requested type when creating speculative-decoding drafts.
- Fixed Qwen3 MoE decoding for differently-quantized experts, plus faster packed gate/up projection (~4–9% on M5 Max).

**Full Changelog**: https://github.com/ollama/ollama/compare/v0.32.3...v0.32.4