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
85 lines
3.1 KiB
C++
85 lines
3.1 KiB
C++
#include "engine/community_models/glm_tts/assets.h"
|
|
#include "engine/community_models/glm_tts/llama.h"
|
|
#include "engine/community_models/glm_tts/prompt.h"
|
|
#include "engine/community_models/glm_tts/speech_tokenizer.h"
|
|
#include "engine/community_models/glm_tts/tokenizer_text.h"
|
|
#include "engine/framework/audio/wav_reader.h"
|
|
|
|
#include <exception>
|
|
#include <iostream>
|
|
|
|
int main(int argc, char ** argv) try {
|
|
if (argc < 5) {
|
|
std::cerr
|
|
<< "usage: glm_tts_llama_probe <model-path> <reference.wav> "
|
|
"<reference-text> <text> [--greedy] [--seed n] "
|
|
"[--full-head]\n";
|
|
return 2;
|
|
}
|
|
|
|
auto assets = engine::models::glm_tts::load_glm_tts_assets(argv[1]);
|
|
const auto wav =
|
|
engine::audio::read_wav_f32(std::filesystem::path(argv[2]));
|
|
engine::runtime::AudioBuffer audio;
|
|
audio.sample_rate = wav.sample_rate;
|
|
audio.channels = wav.channels;
|
|
audio.samples = wav.samples;
|
|
|
|
const engine::models::glm_tts::GlmTTSSpeechTokenizer speech_tokenizer(
|
|
assets, {engine::core::BackendType::Cuda, 0, 8});
|
|
const auto reference_tokens = speech_tokenizer.encode(audio);
|
|
const engine::models::glm_tts::GlmTTSTextTokenizer text_tokenizer(
|
|
assets->resources.require_file("tokenizer_vocab"),
|
|
assets->resources.require_file("tokenizer_merges"),
|
|
assets->resources.require_file("tokenizer_config"));
|
|
const auto prompt = engine::models::glm_tts::build_glm_tts_prompt(
|
|
assets->config,
|
|
text_tokenizer,
|
|
argv[3],
|
|
argv[4],
|
|
reference_tokens);
|
|
|
|
const engine::models::glm_tts::GlmTTSLlamaRuntime llama(
|
|
assets,
|
|
engine::core::BackendType::Cuda,
|
|
0,
|
|
8);
|
|
engine::models::glm_tts::GlmTTSGenerateOptions options;
|
|
options.seed = 0;
|
|
for (int index = 5; index < argc; ++index) {
|
|
const std::string option = argv[index];
|
|
if (option == "--greedy") {
|
|
options.temperature = 0.0F;
|
|
} else if (option == "--full-head") {
|
|
options.restrict_output_head = false;
|
|
} else if (option == "--seed" && index + 1 < argc) {
|
|
options.seed = static_cast<uint32_t>(
|
|
std::stoul(argv[++index]));
|
|
} else {
|
|
throw std::runtime_error(
|
|
"unknown or incomplete probe option: " + option);
|
|
}
|
|
}
|
|
const auto generated = llama.generate(prompt, options);
|
|
|
|
std::cout << "{\"prompt_ids\":[";
|
|
for (size_t index = 0; index < prompt.input_ids.size(); ++index) {
|
|
if (index != 0) std::cout << ',';
|
|
std::cout << prompt.input_ids[index];
|
|
}
|
|
std::cout << "],\"speech_tokens\":[";
|
|
for (size_t index = 0; index < generated.speech_tokens.size(); ++index) {
|
|
if (index != 0) std::cout << ',';
|
|
std::cout << generated.speech_tokens[index];
|
|
}
|
|
std::cout
|
|
<< "],\"stopped\":"
|
|
<< (generated.stopped_on_end_of_audio ? "true" : "false")
|
|
<< "}\n";
|
|
return 0;
|
|
} catch (const std::exception & error) {
|
|
std::cerr
|
|
<< "glm_tts_llama_probe failed: "
|
|
<< error.what() << '\n';
|
|
return 1;
|
|
}
|