| // SPDX-License-Identifier: Apache-2.0 | |
| void fp4_w4a16_linear_bf16( | |
| torch::Tensor const& a_packed, | |
| torch::Tensor const& b_packed, | |
| torch::Tensor const& sfa, | |
| torch::Tensor const& sfb, | |
| torch::Tensor& out, | |
| double alpha, | |
| int64_t variant); | |
| void quantize_fp4_sfa_fp16( | |
| torch::Tensor const& x, | |
| torch::Tensor& packed, | |
| torch::Tensor& sfa, | |
| bool is_sfb); | |
| void quantize_fp4_sfa_bf16( | |
| torch::Tensor const& x, | |
| torch::Tensor& packed, | |
| torch::Tensor& sfa, | |
| bool is_sfb); | |
| void dequantize_fp4_sfa_fp16( | |
| torch::Tensor const& packed, | |
| torch::Tensor const& sfa, | |
| torch::Tensor& out, | |
| bool is_sfb); | |