Spaces:
Running
on
Zero
Running
on
Zero
add trt script TODO
Browse files
cosyvoice/bin/{export.py → export_jit.py}
RENAMED
@@ -44,7 +44,7 @@ def main():
|
|
44 |
torch._C._jit_set_profiling_mode(False)
|
45 |
torch._C._jit_set_profiling_executor(False)
|
46 |
|
47 |
-
cosyvoice = CosyVoice(args.model_dir,
|
48 |
|
49 |
# 1. export llm text_encoder
|
50 |
llm_text_encoder = cosyvoice.model.llm.text_encoder.half()
|
|
|
44 |
torch._C._jit_set_profiling_mode(False)
|
45 |
torch._C._jit_set_profiling_executor(False)
|
46 |
|
47 |
+
cosyvoice = CosyVoice(args.model_dir, load_jit=False, load_trt=False)
|
48 |
|
49 |
# 1. export llm text_encoder
|
50 |
llm_text_encoder = cosyvoice.model.llm.text_encoder.half()
|
cosyvoice/bin/export_trt.py
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# TODO 跟export_jit一样的逻辑,完成flow部分的estimator的onnx导出。
|
2 |
+
# tensorrt的安装方式,再这里写一下步骤提示如下,如果没有安装,那么不要执行这个脚本,提示用户先安装,不给选择
|
3 |
+
try:
|
4 |
+
import tensorrt
|
5 |
+
except ImportError:
|
6 |
+
print('step1, 下载\n step2. 解压,安装whl,')
|
7 |
+
# 安装命令里tensosrt的根目录用环境变量导入,比如os.environ['tensorrt_root_dir']/bin/exetrace,然后python里subprocess里执行导出命令
|
8 |
+
# 后面我会在run.sh里写好执行命令 tensorrt_root_dir=xxxx python cosyvoice/bin/export_trt.py --model_dir xxx
|
cosyvoice/cli/cosyvoice.py
CHANGED
@@ -21,7 +21,7 @@ from cosyvoice.utils.file_utils import logging
|
|
21 |
|
22 |
class CosyVoice:
|
23 |
|
24 |
-
def __init__(self, model_dir,
|
25 |
instruct = True if '-Instruct' in model_dir else False
|
26 |
self.model_dir = model_dir
|
27 |
if not os.path.exists(model_dir):
|
@@ -39,9 +39,12 @@ class CosyVoice:
|
|
39 |
self.model.load('{}/llm.pt'.format(model_dir),
|
40 |
'{}/flow.pt'.format(model_dir),
|
41 |
'{}/hift.pt'.format(model_dir))
|
42 |
-
if
|
43 |
-
self.model.
|
44 |
'{}/llm.llm.fp16.zip'.format(model_dir))
|
|
|
|
|
|
|
45 |
del configs
|
46 |
|
47 |
def list_avaliable_spks(self):
|
|
|
21 |
|
22 |
class CosyVoice:
|
23 |
|
24 |
+
def __init__(self, model_dir, load_jit=True, load_trt=True):
|
25 |
instruct = True if '-Instruct' in model_dir else False
|
26 |
self.model_dir = model_dir
|
27 |
if not os.path.exists(model_dir):
|
|
|
39 |
self.model.load('{}/llm.pt'.format(model_dir),
|
40 |
'{}/flow.pt'.format(model_dir),
|
41 |
'{}/hift.pt'.format(model_dir))
|
42 |
+
if load_jit:
|
43 |
+
self.model.load_jit('{}/llm.text_encoder.fp16.zip'.format(model_dir),
|
44 |
'{}/llm.llm.fp16.zip'.format(model_dir))
|
45 |
+
if load_trt:
|
46 |
+
# TODO
|
47 |
+
self.model.load_trt()
|
48 |
del configs
|
49 |
|
50 |
def list_avaliable_spks(self):
|
cosyvoice/cli/model.py
CHANGED
@@ -53,12 +53,17 @@ class CosyVoiceModel:
|
|
53 |
self.hift.load_state_dict(torch.load(hift_model, map_location=self.device))
|
54 |
self.hift.to(self.device).eval()
|
55 |
|
56 |
-
def
|
57 |
llm_text_encoder = torch.jit.load(llm_text_encoder_model)
|
58 |
self.llm.text_encoder = llm_text_encoder
|
59 |
llm_llm = torch.jit.load(llm_llm_model)
|
60 |
self.llm.llm = llm_llm
|
61 |
|
|
|
|
|
|
|
|
|
|
|
62 |
def llm_job(self, text, text_len, prompt_text, prompt_text_len, llm_prompt_speech_token, llm_prompt_speech_token_len, llm_embedding, this_uuid):
|
63 |
with self.llm_context:
|
64 |
for i in self.llm.inference(text=text.to(self.device),
|
|
|
53 |
self.hift.load_state_dict(torch.load(hift_model, map_location=self.device))
|
54 |
self.hift.to(self.device).eval()
|
55 |
|
56 |
+
def load_jit(self, llm_text_encoder_model, llm_llm_model):
|
57 |
llm_text_encoder = torch.jit.load(llm_text_encoder_model)
|
58 |
self.llm.text_encoder = llm_text_encoder
|
59 |
llm_llm = torch.jit.load(llm_llm_model)
|
60 |
self.llm.llm = llm_llm
|
61 |
|
62 |
+
def load_trt(self):
|
63 |
+
# TODO 你需要的TRT推理的准备
|
64 |
+
self.flow.decoder.estimator = xxx
|
65 |
+
self.flow.decoder.session = xxx
|
66 |
+
|
67 |
def llm_job(self, text, text_len, prompt_text, prompt_text_len, llm_prompt_speech_token, llm_prompt_speech_token_len, llm_embedding, this_uuid):
|
68 |
with self.llm_context:
|
69 |
for i in self.llm.inference(text=text.to(self.device),
|
cosyvoice/flow/flow_matching.py
CHANGED
@@ -77,10 +77,10 @@ class ConditionalCFM(BASECFM):
|
|
77 |
sol = []
|
78 |
|
79 |
for step in range(1, len(t_span)):
|
80 |
-
dphi_dt = self.
|
81 |
# Classifier-Free Guidance inference introduced in VoiceBox
|
82 |
if self.inference_cfg_rate > 0:
|
83 |
-
cfg_dphi_dt = self.
|
84 |
x, mask,
|
85 |
torch.zeros_like(mu), t,
|
86 |
torch.zeros_like(spks) if spks is not None else None,
|
@@ -96,6 +96,14 @@ class ConditionalCFM(BASECFM):
|
|
96 |
|
97 |
return sol[-1]
|
98 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
def compute_loss(self, x1, mask, mu, spks=None, cond=None):
|
100 |
"""Computes diffusion loss
|
101 |
|
|
|
77 |
sol = []
|
78 |
|
79 |
for step in range(1, len(t_span)):
|
80 |
+
dphi_dt = self.forward_estimator(x, mask, mu, t, spks, cond)
|
81 |
# Classifier-Free Guidance inference introduced in VoiceBox
|
82 |
if self.inference_cfg_rate > 0:
|
83 |
+
cfg_dphi_dt = self.forward_estimator(
|
84 |
x, mask,
|
85 |
torch.zeros_like(mu), t,
|
86 |
torch.zeros_like(spks) if spks is not None else None,
|
|
|
96 |
|
97 |
return sol[-1]
|
98 |
|
99 |
+
# TODO
|
100 |
+
def forward_estimator(self):
|
101 |
+
if isinstance(self.estimator, trt):
|
102 |
+
assert self.training is False, 'tensorrt cannot be used in training'
|
103 |
+
return xxx
|
104 |
+
else:
|
105 |
+
return self.estimator.forward
|
106 |
+
|
107 |
def compute_loss(self, x1, mask, mu, spks=None, cond=None):
|
108 |
"""Computes diffusion loss
|
109 |
|