Function parse_text_encoder_config

src/models/ace_step/assets.cpp:75–91

        config.rope_theta);
    return config;
}

AceStepDiffusionConfig parse_diffusion_config(const engine::io::json::Value & value) {
    AceStepDiffusionConfig config;
    config.model_type = json::optional_string(value, "model_type", config.model_type);
    config.model_version = json::optional_string(value, "model_version", config.model_version);
    config.hidden_size = json::require_i64(value, "hidden_size");
    config.intermediate_size = json::require_i64(value, "intermediate_size");
    config.num_hidden_layers = json::require_i64(value, "num_hidden_layers");
    config.num_attention_heads = json::require_i64(value, "num_attention_heads");
    config.num_key_value_heads = json::require_i64(value, "num_key_value_heads");
    config.head_dim = json::optional_i64(value, "head_dim", config.hidden_size / config.num_attention_heads);
    config.text_hidden_dim = json::require_i64(value, "text_hidden_dim");
    config.in_channels = json::require_i64(value, "in_channels");
    config.patch_size = json::require_i64(value, "patch_size");
    config.latent_channels = json::require_i64(value, "audio_acoustic_hidden_dim");
    config.pool_window_size = json::require_i64(value, "pool_window_size");
    config.timbre_hidden_dim = json::require_i64(value, "timbre_hidden_dim");
    config.timbre_fix_frame = json::require_i64(value, "timbre_fix_frame");
    config.num_lyric_encoder_hidden_layers = json::require_i64(value, "num_lyric_encoder_hidden_layers");