Fabio
c36510c21e
Some checks failed
Linux Build / Linux cpu (push) Has been cancelled
Linux Build / Linux cpu-clang (push) Has been cancelled
Linux Build / Linux vulkan (push) Has been cancelled
mac-build / macOS CPU (push) Has been cancelled
Nix Build / Nix (cpu) (push) Has been cancelled
Nix Build / Nix (vulkan) (push) Has been cancelled
windows-build / Windows CPU (push) Has been cancelled
91 lines
2.5 KiB
C++
91 lines
2.5 KiB
C++
#pragma once
|
|
|
|
#include "engine/framework/assets/resource_bundle.h"
|
|
#include "engine/framework/assets/tensor_source.h"
|
|
|
|
#include <filesystem>
|
|
#include <memory>
|
|
#include <string>
|
|
#include <vector>
|
|
|
|
namespace engine::models::demucs {
|
|
|
|
struct HTDemucsManifest {
|
|
std::string model_type;
|
|
std::string name;
|
|
};
|
|
|
|
struct HTDemucsConfig {
|
|
std::string class_name;
|
|
std::string signature;
|
|
std::string checkpoint_file;
|
|
std::vector<std::string> sources;
|
|
int audio_channels = 0;
|
|
int sample_rate = 0;
|
|
float segment_seconds = 0.0f;
|
|
int64_t segment_samples = 0;
|
|
int channels = 0;
|
|
int growth = 0;
|
|
int n_fft = 0;
|
|
int hop_length = 0;
|
|
int wiener_iters = 0;
|
|
bool wiener_residual = false;
|
|
bool cac = true;
|
|
int depth = 0;
|
|
bool rewrite = true;
|
|
int multi_freqs_depth = 0;
|
|
float freq_emb_scale = 0.0f;
|
|
float embedding_scale = 1.0f;
|
|
bool embedding_smooth = false;
|
|
int kernel_size = 0;
|
|
int stride = 0;
|
|
int time_stride = 0;
|
|
int context = 0;
|
|
int context_enc = 0;
|
|
int norm_starts = 0;
|
|
int norm_groups = 0;
|
|
int dconv_mode = 0;
|
|
int dconv_depth = 0;
|
|
int dconv_comp = 0;
|
|
float dconv_init = 0.0f;
|
|
int bottom_channels = 0;
|
|
bool has_channel_sampler = false;
|
|
int transformer_layers = 0;
|
|
float transformer_hidden_scale = 0.0f;
|
|
int transformer_heads = 0;
|
|
float transformer_dropout = 0.0f;
|
|
bool transformer_layer_scale = false;
|
|
bool transformer_gelu = true;
|
|
bool transformer_norm_in = true;
|
|
bool transformer_norm_in_group = false;
|
|
bool transformer_group_norm = false;
|
|
bool transformer_norm_first = true;
|
|
bool transformer_norm_out = true;
|
|
bool transformer_cross_first = false;
|
|
float transformer_max_period = 10000.0f;
|
|
float transformer_weight_pos_embed = 1.0f;
|
|
|
|
int input_freq_channels = 0;
|
|
int output_freq_channels = 0;
|
|
int output_time_channels = 0;
|
|
int stft_freq_bins = 0;
|
|
int stft_frames = 0;
|
|
};
|
|
|
|
struct HTDemucsSubmodelAssets {
|
|
std::shared_ptr<const assets::TensorSource> tensor_source;
|
|
HTDemucsConfig config;
|
|
};
|
|
|
|
struct HTDemucsAssets {
|
|
assets::ResourceBundle resources;
|
|
HTDemucsManifest manifest;
|
|
std::vector<std::shared_ptr<const HTDemucsSubmodelAssets>> submodels;
|
|
};
|
|
|
|
void validate_demucs_weight_storage_type(assets::TensorStorageType storage_type);
|
|
std::shared_ptr<const HTDemucsAssets> load_htdemucs_assets(
|
|
const std::filesystem::path & model_path,
|
|
std::string_view family = "htdemucs");
|
|
|
|
} // namespace engine::models::demucs
|