Skip to content

[onnx] add quant export #235

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
robin1001 merged 1 commit into from
Nov 2, 2025
Merged

[onnx] add quant export #235

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 29 additions & 9 deletions wetts/vits/export_onnx.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@

import torch
from model.models import SynthesizerTrn
from onnxruntime.quantization import QuantType, quantize_dynamic
from utils import task


Expand All @@ -30,11 +31,10 @@ def get_args():
required=True,
help="input phone dict")
parser.add_argument("--speaker_table", default=None, help="speaker table")
parser.add_argument(
"--streaming",
action="store_true",
help="export streaming model"
)
parser.add_argument("--streaming",
action="store_true",
help="export streaming model")
parser.add_argument("--quant", action="store_true", help="quant the model")
parser.add_argument(
"--providers",
required=False,
Expand All @@ -45,6 +45,7 @@ def get_args():
args = parser.parse_args()
return args


def add_prefix(filepath, prefix):
filepath = Path(filepath)
return str(filepath.parent / (prefix + filepath.name))
Expand Down Expand Up @@ -92,10 +93,11 @@ def main():
if args.streaming:
net_g.forward = net_g.export_encoder_forward
dummy_input = (seq, seq_len, scales, sid)
encoder_path = add_prefix(args.onnx_model, 'encoder_')
torch.onnx.export(
model=net_g,
args=dummy_input,
f=add_prefix(args.onnx_model, 'encoder_'),
f=encoder_path,
input_names=["input", "input_lengths", "scales", "sid"],
output_names=["z"],
dynamic_axes={
Expand All @@ -112,21 +114,29 @@ def main():
"sid": {
0: "batch"
},
"z": {0: "batch", 1: "L"},
"z": {
0: "batch",
1: "L"
},
},
opset_version=13,
verbose=False,
)

decoder_path = add_prefix(args.onnx_model, 'decoder_')
net_g.forward = net_g.export_decoder_forward
dummy_input = (z, sid)
torch.onnx.export(
model=net_g,
args=dummy_input,
f=add_prefix(args.onnx_model, 'decoder_'),
f=decoder_path,
input_names=["z", "sid"],
output_names=["output"],
dynamic_axes={
"z": {0: "batch", 1: "L"},
"z": {
0: "batch",
1: "L"
},
"output": {
0: "batch",
1: "audio",
Expand All @@ -136,6 +146,16 @@ def main():
opset_version=13,
verbose=False,
)
if args.quant:
encoder_q_path = add_prefix(args.onnx_model, 'encoder_q_')
quantize_dynamic(encoder_path,
encoder_q_path,
weight_type=QuantType.QUInt8)
decoder_q_path = add_prefix(args.onnx_model, 'decoder_q_')
quantize_dynamic(decoder_path,
decoder_q_path,
weight_type=QuantType.QUInt8)

else:
dummy_input = (seq, seq_len, scales, sid)
torch.onnx.export(
Expand Down