sglang_v0.5.2/pytorch_2.8.0/third_party/XNNPACK/test/qp8-f32-qb4w-gemm-minmax.yaml

41 lines
1.5 KiB
YAML

# Copyright 2023 Google LLC
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
# Arm KleidiAI kernels
- name: xnn_qp8_f32_qb4w_gemm_minmax_ukernel_1x4c16s2__aarch64_neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_kai_qb4_weights_and_biases
packed-stride: xnn_packed_stride_kai_qb4_weights_and_biases
k-block: 32
cpp-check: XNN_ENABLE_KLEIDIAI
- name: xnn_qp8_f32_qb4w_gemm_minmax_ukernel_1x8c16s2__aarch64_neondot
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_kai_qb4_weights_and_biases
packed-stride: xnn_packed_stride_kai_qb4_weights_and_biases
k-block: 32
cpp-check: XNN_ENABLE_KLEIDIAI
- name: xnn_qp8_f32_qb4w_gemm_minmax_ukernel_4x8c16s2__neoni8mm
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_kai_qb4_weights_and_biases
packed-stride: xnn_packed_stride_kai_qb4_weights_and_biases
k-block: 32
cpp-check: XNN_ENABLE_KLEIDIAI
- name: xnn_qp8_f32_qb4w_gemm_minmax_ukernel_8x4c16s2__neoni8mm_mstep2
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_kai_qb4_weights_and_biases
packed-stride: xnn_packed_stride_kai_qb4_weights_and_biases
k-block: 32
cpp-check: XNN_ENABLE_KLEIDIAI
- name: xnn_qp8_f32_qb4w_gemm_minmax_ukernel_16x4c16s2__neoni8mm_mstep4
init: xnn_init_f32_qb4w_minmax_scalar_params
pack: xnn_pack_kai_qb4_weights_and_biases
packed-stride: xnn_packed_stride_kai_qb4_weights_and_biases
k-block: 32
cpp-check: XNN_ENABLE_KLEIDIAI