|
|
from ..llama_configs import LLAMA_CONFIGS |
|
|
|
|
|
|
|
|
class T3Config: |
|
|
start_text_token = 255 |
|
|
stop_text_token = 0 |
|
|
text_tokens_dict_size = 704 |
|
|
max_text_tokens = 2048 |
|
|
|
|
|
start_speech_token = 6561 |
|
|
stop_speech_token = 6562 |
|
|
speech_tokens_dict_size = 8194 |
|
|
max_speech_tokens = 4096 |
|
|
|
|
|
llama_config_name = "Llama_520M" |
|
|
input_pos_emb = "learned" |
|
|
speech_cond_prompt_len = 150 |
|
|
|
|
|
|
|
|
encoder_type = "voice_encoder" |
|
|
speaker_embed_size = 256 |
|
|
use_perceiver_resampler = True |
|
|
emotion_adv = True |
|
|
|
|
|
@property |
|
|
def n_channels(self): |
|
|
return LLAMA_CONFIGS[self.llama_config_name]["hidden_size"] |
|
|
|
