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
46 lines
1.5 KiB
C++
46 lines
1.5 KiB
C++
#pragma once
|
|
|
|
#include "engine/framework/modules/attention/scaled_dot_product_attention.h"
|
|
#include "engine/framework/modules/linear_module.h"
|
|
#include "engine/framework/modules/norm_modules.h"
|
|
#include "engine/framework/modules/streaming_conv_modules.h"
|
|
|
|
#include <vector>
|
|
|
|
namespace engine::models::gigaam_asr {
|
|
|
|
struct ConformerConfig {
|
|
int64_t features = 64;
|
|
int64_t hidden_size = 768;
|
|
int64_t intermediate_size = 3072;
|
|
int64_t heads = 16;
|
|
int64_t conv_kernel = 5;
|
|
int64_t subsampling_kernel = 5;
|
|
float rope_theta = 5000.0f;
|
|
modules::ScaledDotProductAttentionLowering attention =
|
|
modules::ScaledDotProductAttentionLowering::Flash;
|
|
};
|
|
|
|
struct ConformerLayerWeights {
|
|
modules::NormWeights ffn1_norm, attention_norm, conv_norm, ffn2_norm, output_norm;
|
|
modules::LinearWeights ffn1_in, ffn1_out, ffn2_in, ffn2_out;
|
|
modules::LinearWeights query, key, value, attention_out;
|
|
modules::LinearWeights pointwise_in, pointwise_out;
|
|
modules::DepthwiseConv1dWeights depthwise;
|
|
modules::NormWeights depthwise_norm;
|
|
};
|
|
|
|
struct ConformerWeights {
|
|
modules::Conv1dWeights subsampling1, subsampling2;
|
|
std::vector<ConformerLayerWeights> layers;
|
|
};
|
|
|
|
// Features are [1, mel_bins, frames]. Positions index the subsampled frames.
|
|
core::TensorValue build_conformer_encoder(
|
|
core::ModuleBuildContext & ctx,
|
|
const core::TensorValue & features,
|
|
const core::TensorValue & positions,
|
|
const ConformerConfig & config,
|
|
const ConformerWeights & weights);
|
|
|
|
} // namespace engine::models::gigaam_asr
|