deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-04 09:29:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

add flow trt wrapper

Browse Source
This commit is contained in:
lyuxiang.lx
2025-04-16 17:57:02 +08:00
parent 7f8bea2669
commit a442317d17
8 changed files with 615 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
cosyvoice/cli/cosyvoice.py
View File

@@ -137,7 +137,7 @@ class CosyVoice:
class CosyVoice2(CosyVoice):
def __init__(self, model_dir, load_jit=False, load_trt=False, fp16=False, use_flow_cache=False):
def __init__(self, model_dir, load_jit=False, load_trt=False, fp16=False, use_flow_cache=False, trt_concurrent=1):
self.instruct = True if '-Instruct' in model_dir else False
self.model_dir = model_dir
self.fp16 = fp16
@@ -159,7 +159,7 @@ class CosyVoice2(CosyVoice):
if torch.cuda.is_available() is False and (load_jit is True or load_trt is True or fp16 is True):
load_jit, load_trt, fp16 = False, False, False
logging.warning('no cuda device, set load_jit/load_trt/fp16 to False')
self.model = CosyVoice2Model(configs['llm'], configs['flow'], configs['hift'], fp16, use_flow_cache)
self.model = CosyVoice2Model(configs['llm'], configs['flow'], configs['hift'], fp16, use_flow_cache, trt_concurrent)
self.model.load('{}/llm.pt'.format(model_dir),
'{}/flow.pt'.format(model_dir) if use_flow_cache is False else '{}/flow.cache.pt'.format(model_dir),
'{}/hift.pt'.format(model_dir))