v0.32.4
Added
- Support Laguna on Apple GPUs via the MLX engine
Changed
- Quantize draft-model output heads at the requested type when creating speculative-decoding drafts
- Faster packed gate/up projection for Qwen3 MoE decoding on M5 Max
Fixed
- Fixed Qwen3 MoE decoding for differently-quantized experts and improved packed gate/up projection performance
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