voice_audio_multi/include/engine/models/yue2/pipeline.h
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
Initial import: audio.cpp with Orange Pi documentation
2026-11-04 19:05:22 +01:00

63 lines
2.1 KiB
C++

#pragma once
#include "engine/framework/assets/tensor_source.h"
#include "engine/framework/codecs/oobleck_audio_vae_runtime.h"
#include "engine/framework/core/attention_fallback.h"
#include "engine/framework/core/execution_context.h"
#include "engine/framework/runtime/session.h"
#include "engine/models/yue2/assets.h"
#include "engine/models/yue2/request.h"
#include "engine/models/yue2/tokenizer_text.h"
#include <memory>
#include <optional>
#include <string>
#include <vector>
namespace engine::models::yue2 {
struct Yue2RunResult {
std::optional<runtime::AudioBuffer> audio;
std::string plan_abc_text;
bool plan_abc_truncated = false;
std::vector<int32_t> semantic_codes;
bool semantic_truncated = false;
};
class Yue2PipelineRuntime {
public:
Yue2PipelineRuntime(
core::ExecutionContext & execution,
std::shared_ptr<const Yue2Assets> assets,
assets::TensorStorageType model_weight_type,
assets::TensorStorageType vae_weight_type,
size_t model_weight_context_bytes,
size_t vae_weight_context_bytes,
size_t ar_prefill_graph_arena_bytes,
size_t ar_decode_graph_arena_bytes,
size_t nar_graph_arena_bytes,
size_t vae_graph_arena_bytes,
core::AttentionPreference attention_preference = core::AttentionPreference::Auto,
int64_t nar_attention_tile_rows = 0,
bool mem_saver = true,
bool packed_projections = false,
int64_t kv_cache_alignment = 1,
int64_t vae_tile_frames = 0);
~Yue2PipelineRuntime();
Yue2Plan plan(const Yue2Request & request);
Yue2SemanticResult generate_semantic(const Yue2Request & request, Yue2Plan plan);
std::vector<float> synthesize_latents(
const Yue2SemanticResult & semantic,
const Yue2GenerationConfig & generation,
uint64_t seed);
runtime::AudioBuffer decode_audio(const std::vector<float> & latents, int64_t frames);
Yue2RunResult run(const Yue2Request & request);
void release_runtime_graphs();
private:
class Impl;
std::unique_ptr<Impl> impl_;
};
} // namespace engine::models::yue2