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
66 lines
2.4 KiB
C++
66 lines
2.4 KiB
C++
#pragma once
|
|
|
|
#include "engine/framework/assets/resource_bundle.h"
|
|
#include "engine/framework/audio/dsp.h"
|
|
#include "engine/framework/audio/nemo_mel_frontend.h"
|
|
#include "engine/framework/core/backend_weight_store.h"
|
|
#include "engine/framework/core/execution_context.h"
|
|
#include "engine/framework/modules/conformer_modules.h"
|
|
#include "engine/framework/modules/transformers/transformer_blocks.h"
|
|
#include "engine/framework/runtime/model.h"
|
|
#include "engine/framework/runtime/cache_slots.h"
|
|
#include "engine/framework/tokenizers/sentencepiece.h"
|
|
|
|
#include <memory>
|
|
#include <string>
|
|
#include <vector>
|
|
|
|
namespace engine::models::cohere_asr {
|
|
|
|
struct CohereAssets {
|
|
assets::ResourceBundle resources;
|
|
std::shared_ptr<const assets::TensorSource> source;
|
|
std::vector<tokenizers::SentencePiecePiece> vocabulary;
|
|
std::vector<float> window;
|
|
audio::SparseMelFilterbank filterbank;
|
|
std::shared_ptr<const audio::NemoMelFrontend> frontend;
|
|
int32_t special_token(const std::string & text) const;
|
|
};
|
|
|
|
struct CohereWeights {
|
|
std::unique_ptr<core::BackendWeightStore> store;
|
|
modules::DepthwiseConvSubsamplingWeights subsampling;
|
|
modules::LinearWeights encoder_out, head;
|
|
std::vector<modules::RelativeConformerBlockWeights> encoder;
|
|
std::vector<modules::TransformerDecoderBlockWeights> decoder;
|
|
core::TensorValue embedding, positions;
|
|
modules::NormWeights embedding_norm, decoder_norm;
|
|
};
|
|
|
|
std::shared_ptr<const CohereAssets> load_cohere_assets(const std::filesystem::path & path);
|
|
std::unique_ptr<CohereWeights> load_cohere_weights(
|
|
const CohereAssets & assets, core::ExecutionContext & execution, assets::TensorStorageType type);
|
|
|
|
audio::AudioTensor extract_cohere_frontend(
|
|
const std::vector<float> & samples,
|
|
const CohereAssets & assets,
|
|
size_t threads);
|
|
|
|
class CohereRuntime {
|
|
public:
|
|
CohereRuntime(const CohereAssets & assets, const CohereWeights & weights, core::ExecutionContext & execution);
|
|
~CohereRuntime();
|
|
std::vector<std::vector<int32_t>> transcribe(const std::vector<std::vector<float>> & samples,
|
|
const std::vector<int32_t> & prompt, int64_t max_tokens);
|
|
|
|
private:
|
|
struct Graphs;
|
|
const CohereAssets & assets_;
|
|
const CohereWeights & weights_;
|
|
core::ExecutionContext & execution_;
|
|
runtime::CacheSlots<std::pair<int64_t, int64_t>, std::unique_ptr<Graphs>> graphs_;
|
|
};
|
|
|
|
std::shared_ptr<runtime::IVoiceModelLoader> make_cohere_asr_loader();
|
|
|
|
} // namespace engine::models::cohere_asr
|