GlmVoiceTrainingStage

Enums · AiDotNet.TextToSpeech.MultiModal

Which of GLM-4-Voice’s two language-model stages a training step follows (§4).

Fields

Field Summary
GlmVoiceTrainingStage FineTuning Stage 2, supervised fine-tuning (§4.2.2): the rate decaying from 1e-5 to 1e-6, weight decay 0.1, dropout 0.5 on the hidden states, gradients clipped at 1.0.
GlmVoiceTrainingStage PreTraining Stage 1, joint speech-text pre-training (§4.1.1): AdamW (0.9, 0.95), the rate decaying linearly from 6e-5 to 6e-6, no dropout.
int value__