|
|
@@ -355,6 +355,7 @@ class CosyVoice2Model(CosyVoiceModel):
|
|
|
opt_shape = [(2, 80, 200), (2, 1, 200), (2, 80, 200), (2, 80, 200), (1, 4, 2, 100, 512, 2), (12, 4, 2, 100, 512, 2), (1, 4, 2, 100, 512, 2)]
|
|
|
max_shape = [(2, 80, 1500), (2, 1, 1500), (2, 80, 1500), (2, 80, 1500), (1, 4, 2, 200, 512, 2), (12, 4, 2, 200, 512, 2), (1, 4, 2, 200, 512, 2)]
|
|
|
input_names = ["x", "mask", "mu", "cond", 'down_blocks_kv_cache', 'mid_blocks_kv_cache', 'up_blocks_kv_cache']
|
|
|
+ assert self.use_flow_cache is True, "get_trt_kwargs is set for flow cache mode. If you want to use trt with use_flow_cache=False, please set higher max_shape"
|
|
|
return {'min_shape': min_shape, 'opt_shape': opt_shape, 'max_shape': max_shape, 'input_names': input_names}
|
|
|
|
|
|
def token2wav(self, token, prompt_token, prompt_feat, embedding, uuid, finalize=False, speed=1.0):
|
lyuxiang.lx