LlamaOmniTrainingStage
Enums · AiDotNet.TextToSpeech.MultiModal
Which of LLaMA-Omni’s two training stages (§2.5) a step follows.
Fields
| Field | Summary |
|---|---|
LlamaOmniTrainingStage Joint |
The release’s joint option (tune_speech_generator_only = False): the text loss plus the CTC loss weighted by CtcLossWeight, all of adaptor, language model and decoder training. |
LlamaOmniTrainingStage SpeechDecoder |
Stage 2: only the speech decoder learns the units by CTC (eq. |
LlamaOmniTrainingStage TextResponse |
Stage 1: the speech adaptor and the language model learn the text response (eq. |
int value__ |