v0.34.4-rc0: mlx: speed up Qwen 3.8 prompt processing (#18550)

What moved

mlx: speed up Qwen 3.8 prompt processing Use MLX's gated-delta kernel for long scans and fold dense MLP global scales into SwiGLU. address comments

Why it matters

Filed as a public note from Ollama, 2026-09-23. Tagged Hardware.

On the record

• Filed from the Ollama official RSS on 2026-09-23.
• Primary source host: github.com.
• mlx: speed up Qwen 3.8 prompt processing Use MLX's gated-delta kernel for long scans and fold dense MLP global scales into SwiGLU.