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

Prevent repeatedly loading the BERT model from the disk. (#37)

Browse Source 
* Prevent repeatedly loading the BERT model from the disk.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
This commit is contained in:
jiangyuxiaoxiao
2023-09-25 12:39:55 +08:00
committed by GitHub
parent c21af23d10
commit 4f2ae91ce2
2 changed files with 14 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
text/chinese_bert.py
View File

@@ -4,6 +4,8 @@ from transformers import AutoTokenizer, AutoModelForMaskedLM
tokenizer = AutoTokenizer.from_pretrained("./bert/chinese-roberta-wwm-ext-large") tokenizer = AutoTokenizer.from_pretrained("./bert/chinese-roberta-wwm-ext-large")
models = dict()
def get_bert_feature(text, word2ph, device=None): def get_bert_feature(text, word2ph, device=None):
if ( if (
@@ -14,14 +16,15 @@ def get_bert_feature(text, word2ph, device=None):
device = "mps" device = "mps"
if not device: if not device:
device = "cuda" device = "cuda"
model = AutoModelForMaskedLM.from_pretrained( if device not in models.keys():
models[device] = AutoModelForMaskedLM.from_pretrained(
"./bert/chinese-roberta-wwm-ext-large" "./bert/chinese-roberta-wwm-ext-large"
).to(device) ).to(device)
with torch.no_grad(): with torch.no_grad():
inputs = tokenizer(text, return_tensors="pt") inputs = tokenizer(text, return_tensors="pt")
for i in inputs: for i in inputs:
inputs[i] = inputs[i].to(device) inputs[i] = inputs[i].to(device)
res = model(**inputs, output_hidden_states=True) res = models[device](**inputs, output_hidden_states=True)
res = torch.cat(res["hidden_states"][-3:-2], -1)[0].cpu() res = torch.cat(res["hidden_states"][-3:-2], -1)[0].cpu()
assert len(word2ph) == len(text) + 2 assert len(word2ph) == len(text) + 2

11
text/japanese_bert.py
View File

@@ -4,6 +4,8 @@ import sys
tokenizer = AutoTokenizer.from_pretrained("./bert/bert-base-japanese-v3") tokenizer = AutoTokenizer.from_pretrained("./bert/bert-base-japanese-v3")
models = dict()
def get_bert_feature(text, word2ph, device=None): def get_bert_feature(text, word2ph, device=None):
if ( if (
@@ -14,14 +16,15 @@ def get_bert_feature(text, word2ph, device=None):
device = "mps" device = "mps"
if not device: if not device:
device = "cuda" device = "cuda"
model = AutoModelForMaskedLM.from_pretrained("./bert/bert-base-japanese-v3").to( if device not in models.keys():
device models[device] = AutoModelForMaskedLM.from_pretrained(
) "./bert/bert-base-japanese-v3"
).to(device)
with torch.no_grad(): with torch.no_grad():
inputs = tokenizer(text, return_tensors="pt") inputs = tokenizer(text, return_tensors="pt")
for i in inputs: for i in inputs:
inputs[i] = inputs[i].to(device) inputs[i] = inputs[i].to(device)
res = model(**inputs, output_hidden_states=True) res = models[device](**inputs, output_hidden_states=True)
res = torch.cat(res["hidden_states"][-3:-2], -1)[0].cpu() res = torch.cat(res["hidden_states"][-3:-2], -1)[0].cpu()
assert inputs["input_ids"].shape[-1] == len(word2ph) assert inputs["input_ids"].shape[-1] == len(word2ph)
word2phone = word2ph word2phone = word2ph