Function same_transformer_layer

src/models/stable_audio/same_autoencoder.cpp:297–311

}

core::TensorValue same_transformer_layer(
    core::ModuleBuildContext & ctx,
    const core::TensorValue & input,
    const core::TensorValue & positions,
    const core::TensorValue * attention_mask,
    const StableAudioSameTransformerWeights & weights,
    const StableAudioConfig & config,
    bool sinusoidal) {
    auto attn_in = dynamic_tanh_norm(ctx, input, weights.pre_norm);
    auto attn = same_self_attention(ctx, attn_in, positions, attention_mask, weights.self_attn, config, input.shape.last_dim());
    auto hidden = modules::AddModule{}.build(ctx, input, attn);
    auto ff_in = dynamic_tanh_norm(ctx, hidden, weights.ff_norm);
    auto ff = same_feed_forward(ctx, ff_in, weights.ff, sinusoidal);
    return modules::AddModule{}.build(ctx, hidden, ff);
}

std::vector<float> adapt_interleaved_audio(
    const runtime::AudioBuffer & audio,