28 lines
647 B
Python
28 lines
647 B
Python
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
|
|
|
|
# For T3CondEnc
|
|
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"]
|