Function single_token_embedding

src/models/vibevoice/generator.cpp:146–156

}

std::vector<float> single_token_embedding(
    const VibeVoiceDecoderWeightsRuntime & decoder,
    int32_t token,
    int64_t hidden_size) {
    auto embedding = decoder.embed_tokens({token});
    if (embedding.steps != 1 || embedding.hidden_size != hidden_size ||
        static_cast<int64_t>(embedding.values.size()) != hidden_size) {
        throw std::runtime_error("VibeVoice single token embedding shape mismatch");
    }
    return std::move(embedding.values);
}

int64_t max_generation_steps(const VibeVoicePreparedPrompt & prompt, const VibeVoiceGenerationOptions & options, const VibeVoiceDecoderConfig & config) {
    if (prompt.steps <= 0 || config.max_position_embeddings <= prompt.steps) {