THSFP4.cpp.f32_dummy

#include "THSTensor.h"
#include <ATen/ATen.h>
#include <ATen/cuda/CUDAConfig.h>
#include <iostream>
#include <vector>

extern "C"
{
    EXPORT_API(Tensor) THSTensor_scaled_mm(const Tensor mat1, const Tensor mat2, const Tensor scale_a, const Tensor scale_b, const int8_t out_dtype)
    {
        CATCH_TENSOR(
            at::zeros({(*mat1).size(0), (*mat2).size(1)}, (*mat1).options().dtype((c10::ScalarType)out_dtype))
        )
    }

    EXPORT_API(void) THSFP4_quantize(const Tensor input, Tensor* qdata, Tensor* scale)
    {
        CATCH(
            at::Tensor in = *input;
            int64_t K = in.size(-1);
            int64_t M = in.numel() / K;
            // Return Float32 dummies
            *qdata = new torch::Tensor(at::zeros({M, K}, in.options().dtype(at::kFloat)));
            *scale = new torch::Tensor(at::ones({M, 1}, in.options().dtype(at::kFloat)));
        )
    }
}