Q2_0 group 64: Metal backend

[khosravipasha]

DRAFT PR for tesintg and reviews.

[Copilot]

Pull request overview

Adds Metal backend support for the Q2_0 quantization format (group size 64) so quantized weights can participate in common Metal paths (copy/dequant, get_rows, mul_mv, mul_mm, and mul_mat dispatch).

Changes:

• Implements Q2_0 quantize/dequant routines in the Metal shader library and wires them into existing generic kernels (cpy/get_rows/mul_mm).
• Adds a Q2_0-specific dot-product routine and a dedicated kernel_mul_mv_q2_0_f32 path.
• Enables GGML_TYPE_Q2_0 in Metal op dispatch / device capability checks and sets up pipeline params for mul_mv/mul_mv_id.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
ggml/src/ggml-metal/ggml-metal.metal	Adds Q2_0 quantize/dequant, dot-product helper, and kernel/template instantiations for copy/get_rows/mul_mm/mul_mv.
ggml/src/ggml-metal/ggml-metal-ops.cpp	Allows Q2_0 to use the small-batch mul-mv-ext path in MUL_MAT when applicable.
ggml/src/ggml-metal/ggml-metal-impl.h	Introduces N_R0_Q2_0 / N_SG_Q2_0 constants for pipeline configuration.
ggml/src/ggml-metal/ggml-metal-device.m	Updates Metal device op support checks to include Q2_0 for relevant copy/dup/cont paths.
ggml/src/ggml-metal/ggml-metal-device.cpp	Adds Q2_0 case for mul_mv and mul_mv_id pipeline configuration (nsg/nr0).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.