cwenzi/neuroflow-cpp
1
1#include <iostream>2#include <exception>3#include "../include/neuroflow/memory.hpp"4 5using namespace neuroflow;6 7int main() {8 try {9 std::cout << "MLA debug test..." << std::endl;10 11 LatentKVCache mla(64, 4, 16, 128);12 13 std::cout << "mla.d_model: " << mla.d_model << std::endl;14 std::cout << "mla.n_heads: " << mla.n_heads << std::endl;15 std::cout << "mla.d_latent: " << mla.d_latent << std::endl;16 std::cout << "mla.head_dim: " << mla.head_dim << std::endl;17 std::cout << "mla.cache_len: " << mla.cache_len << std::endl;18 19 Tensor input({1, 64});20 std::cout << "input shape: [" << input.shape_[0] << ", " << input.shape_[1] << "]" << std::endl;21 std::cout << "input numel: " << input.numel() << std::endl;22 23 for (size_t i = 0; i < input.numel(); ++i) input.as_fp32()[i] = 0.1f * i;24 25 std::cout << "Calling mla.forward(input, true)..." << std::endl;26 Tensor output1 = mla.forward(input, true);27 28 std::cout << "output1 shape size: " << output1.shape_.size() << std::endl;29 for (size_t i = 0; i < output1.shape_.size(); ++i) std::cout << " dim " << i << ": " << output1.shape_[i] << std::endl;30 std::cout << "mla.cache_len after 1st forward: " << mla.cache_len << std::endl;31 32 std::cout << "Second forward..." << std::endl;33 Tensor input2({1, 64});34 for (size_t i = 0; i < input2.numel(); ++i) input2.as_fp32()[i] = 0.2f * i;35 Tensor output2 = mla.forward(input2, true);36 37 std::cout << "output2 shape size: " << output2.shape_.size() << std::endl;38 for (size_t i = 0; i < output2.shape_.size(); ++i) std::cout << " dim " << i << ": " << output2.shape_[i] << std::endl;39 std::cout << "mla.cache_len after 2nd forward: " << mla.cache_len << std::endl;40 41 std::cout << "Success!" << std::endl;42 return 0;43 } catch (const std::exception& e) {44 std::cout << "Error: " << e.what() << std::endl;45 return 1;46 }47}48 