tuna2134/sbv2-v2
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Fix: During ONNX inference, the tensor of the return value of the tokenizer is now obtained in Numpy's NDArray format to eliminate the dependency on PyTorch

Browse Source
This commit is contained in:
tsukumi
2024-09-23 07:40:17 +09:00
parent dcca37b9a8
commit b7cd512978
3 changed files with 20 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
style_bert_vits2/nlp/chinese/bert_feature.py
View File

@@ -98,7 +98,7 @@ def extract_bert_feature_onnx(
""" """
tokenizer = onnx_bert_models.load_tokenizer(Languages.ZH) tokenizer = onnx_bert_models.load_tokenizer(Languages.ZH)
inputs = tokenizer(text, return_tensors="pt") inputs = tokenizer(text, return_tensors="np")
session = onnx_bert_models.load_model( session = onnx_bert_models.load_model(
language=Languages.ZH, language=Languages.ZH,
@@ -108,21 +108,21 @@ def extract_bert_feature_onnx(
res = session.run( res = session.run(
[output_name], [output_name],
{ {
"input_ids": inputs["input_ids"].detach().numpy(), # type: ignore "input_ids": inputs["input_ids"],
"token_type_ids": inputs["token_type_ids"].detach().numpy(), # type: ignore "token_type_ids": inputs["token_type_ids"],
"attention_mask": inputs["attention_mask"].detach().numpy(), # type: ignore "attention_mask": inputs["attention_mask"],
}, },
)[0] )[0]
style_res_mean = None style_res_mean = None
if assist_text: if assist_text:
style_inputs = tokenizer(assist_text, return_tensors="pt") style_inputs = tokenizer(assist_text, return_tensors="np")
style_res = session.run( style_res = session.run(
[output_name], [output_name],
{ {
"input_ids": style_inputs["input_ids"].detach().numpy(), # type: ignore "input_ids": style_inputs["input_ids"],
"token_type_ids": style_inputs["token_type_ids"].detach().numpy(), # type: ignore "token_type_ids": style_inputs["token_type_ids"],
"attention_mask": style_inputs["attention_mask"].detach().numpy(), # type: ignore "attention_mask": style_inputs["attention_mask"],
}, },
)[0] )[0]
style_res_mean = np.mean(style_res, axis=0) style_res_mean = np.mean(style_res, axis=0)

12
style_bert_vits2/nlp/english/bert_feature.py
View File

@@ -98,7 +98,7 @@ def extract_bert_feature_onnx(
""" """
tokenizer = onnx_bert_models.load_tokenizer(Languages.EN) tokenizer = onnx_bert_models.load_tokenizer(Languages.EN)
inputs = tokenizer(text, return_tensors="pt") inputs = tokenizer(text, return_tensors="np")
session = onnx_bert_models.load_model( session = onnx_bert_models.load_model(
language=Languages.EN, language=Languages.EN,
@@ -108,19 +108,19 @@ def extract_bert_feature_onnx(
res = session.run( res = session.run(
[output_name], [output_name],
{ {
"input_ids": inputs["input_ids"].detach().numpy(), # type: ignore "input_ids": inputs["input_ids"],
"attention_mask": inputs["attention_mask"].detach().numpy(), # type: ignore "attention_mask": inputs["attention_mask"],
}, },
)[0] )[0]
style_res_mean = None style_res_mean = None
if assist_text: if assist_text:
style_inputs = tokenizer(assist_text, return_tensors="pt") style_inputs = tokenizer(assist_text, return_tensors="np")
style_res = session.run( style_res = session.run(
[output_name], [output_name],
{ {
"input_ids": style_inputs["input_ids"].detach().numpy(), # type: ignore "input_ids": style_inputs["input_ids"],
"attention_mask": style_inputs["attention_mask"].detach().numpy(), # type: ignore "attention_mask": style_inputs["attention_mask"],
}, },
)[0] )[0]
style_res_mean = np.mean(style_res, axis=0) style_res_mean = np.mean(style_res, axis=0)

12
style_bert_vits2/nlp/japanese/bert_feature.py
View File

@@ -111,7 +111,7 @@ def extract_bert_feature_onnx(
assist_text = "".join(text_to_sep_kata(assist_text, raise_yomi_error=False)[0]) assist_text = "".join(text_to_sep_kata(assist_text, raise_yomi_error=False)[0])
tokenizer = onnx_bert_models.load_tokenizer(Languages.JP) tokenizer = onnx_bert_models.load_tokenizer(Languages.JP)
inputs = tokenizer(text, return_tensors="pt") inputs = tokenizer(text, return_tensors="np")
session = onnx_bert_models.load_model( session = onnx_bert_models.load_model(
language=Languages.JP, language=Languages.JP,
@@ -121,19 +121,19 @@ def extract_bert_feature_onnx(
res = session.run( res = session.run(
[output_name], [output_name],
{ {
"input_ids": inputs["input_ids"].detach().numpy(), # type: ignore "input_ids": inputs["input_ids"],
"attention_mask": inputs["attention_mask"].detach().numpy(), # type: ignore "attention_mask": inputs["attention_mask"],
}, },
)[0] )[0]
style_res_mean = None style_res_mean = None
if assist_text: if assist_text:
style_inputs = tokenizer(assist_text, return_tensors="pt") style_inputs = tokenizer(assist_text, return_tensors="np")
style_res = session.run( style_res = session.run(
[output_name], [output_name],
{ {
"input_ids": style_inputs["input_ids"].detach().numpy(), # type: ignore "input_ids": style_inputs["input_ids"],
"attention_mask": style_inputs["attention_mask"].detach().numpy(), # type: ignore "attention_mask": style_inputs["attention_mask"],
}, },
)[0] )[0]
style_res_mean = np.mean(style_res, axis=0) style_res_mean = np.mean(style_res, axis=0)