From 4f2ae91ce2f1eeac2b7a71c8bd5e1023dd2a0ab8 Mon Sep 17 00:00:00 2001
From: jiangyuxiaoxiao <654163754@qq.com>
Date: Mon, 25 Sep 2023 12:39:55 +0800
Subject: [PATCH] Prevent repeatedly loading the BERT model from the disk.
 (#37)

* 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>
---
 text/chinese_bert.py  | 11 +++++++----
 text/japanese_bert.py | 11 +++++++----
 2 files changed, 14 insertions(+), 8 deletions(-)

diff --git a/text/chinese_bert.py b/text/chinese_bert.py
index a760719..8159425 100644
--- a/text/chinese_bert.py
+++ b/text/chinese_bert.py
@@ -4,6 +4,8 @@ from transformers import AutoTokenizer, AutoModelForMaskedLM
 
 tokenizer = AutoTokenizer.from_pretrained("./bert/chinese-roberta-wwm-ext-large")
 
+models = dict()
+
 
 def get_bert_feature(text, word2ph, device=None):
     if (
@@ -14,14 +16,15 @@ def get_bert_feature(text, word2ph, device=None):
         device = "mps"
     if not device:
         device = "cuda"
-    model = AutoModelForMaskedLM.from_pretrained(
-        "./bert/chinese-roberta-wwm-ext-large"
-    ).to(device)
+    if device not in models.keys():
+        models[device] = AutoModelForMaskedLM.from_pretrained(
+            "./bert/chinese-roberta-wwm-ext-large"
+        ).to(device)
     with torch.no_grad():
         inputs = tokenizer(text, return_tensors="pt")
         for i in inputs:
             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()
 
     assert len(word2ph) == len(text) + 2
diff --git a/text/japanese_bert.py b/text/japanese_bert.py
index 5cc104d..5dd1964 100644
--- a/text/japanese_bert.py
+++ b/text/japanese_bert.py
@@ -4,6 +4,8 @@ import sys
 
 tokenizer = AutoTokenizer.from_pretrained("./bert/bert-base-japanese-v3")
 
+models = dict()
+
 
 def get_bert_feature(text, word2ph, device=None):
     if (
@@ -14,14 +16,15 @@ def get_bert_feature(text, word2ph, device=None):
         device = "mps"
     if not device:
         device = "cuda"
-    model = AutoModelForMaskedLM.from_pretrained("./bert/bert-base-japanese-v3").to(
-        device
-    )
+    if device not in models.keys():
+        models[device] = AutoModelForMaskedLM.from_pretrained(
+            "./bert/bert-base-japanese-v3"
+        ).to(device)
     with torch.no_grad():
         inputs = tokenizer(text, return_tensors="pt")
         for i in inputs:
             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()
     assert inputs["input_ids"].shape[-1] == len(word2ph)
     word2phone = word2ph