Artrajz
commented
11 months ago 1.bert-base-japanese-v3未能正确加载（但好像不影响中文推理？）。
可能是不小心动了pth之外的文件，我之前复制模型时不小心多复制了一些文件就出现了和你一样的错误
2.模型版本兼容性问题：刚开始推理出的结果是电流杂音，在config里添加"version": "1.1.0-transition"代码后，正确推理出人物音色，但仍然存在咬词不清的问题。
观察了一下这个模型版本应该是1.0.1
3.[ERROR] [utils.load_checkpoint:57] enc_p.emb.weight is not in the checkpoint
我下载这个模型，并加载该模型后没有遇到这个错误，可能模型没下完整。
大佬有空可否考虑做一个纯bert-vits2的api接口？
不是很理解，在这个项目里你可以只加载bert-vits2，也可以只使用bert-vits2的api接口，其他的接口并不影响bert-vits2的使用，开销也并不大。如果只是为了精简代码，可以自行把bert-vits2之外的代码删除。
yuffiesakiya
运行环境
问题描述
加载bert-vits2模型后，可以正确推理出角色的声线，但存在咬词不清、咬文模糊的问题。（模型在原仓库推理没问题） 模型地址：https://huggingface.co/spaces/XzJosh/Wenjing-Bert-VITS2/tree/main/logs/Azuma
问题复现步骤
INFO:root:Loading yaml from C:\Users\LV HENG\Desktop\vits-simple-api1205\config.yml Building prefix dict from the default dictionary ... DEBUG:jieba:Building prefix dict from the default dictionary ... Loading model from cache C:\Users\LVHENG~1\AppData\Local\Temp\jieba.cache DEBUG:jieba:Loading model from cache C:\Users\LVHENG~1\AppData\Local\Temp\jieba.cache Loading model cost 0.430 seconds. DEBUG:jieba:Loading model cost 0.430 seconds. Prefix dict has been built successfully. DEBUG:jieba:Prefix dict has been built successfully. 2023-12-06 22:19:15 [INFO] [bert_handler.load_bert:93] Loading BERT model: C:\Users\LV HENG\Desktop\vits-simple-api1205\bert_vits2/bert/bert-base-japanese-v3 2023-12-06 22:19:15 [ERROR] [bert_handler.load_bert:101] Failed loading C:\Users\LV HENG\Desktop\vits-simple-api1205\bert_vits2/bert/bert-base-japanese-v3. You need to install fugashi to use MecabTokenizer. See https://pypi.org/project/fugashi/ for installation. 2023-12-06 22:19:15 [INFO] [bert_handler.load_bert:102] Trying to download. 2023-12-06 22:19:16 [INFO] [bert_handler._download_model:86] File already exists and verified successfully! 2023-12-06 22:19:16 [INFO] [bert_handler.load_bert:93] Loading BERT model: C:\Users\LV HENG\Desktop\vits-simple-api1205\bert_vits2/bert/bert-base-japanese-v3 2023-12-06 22:19:16 [ERROR] [bert_handler.load_bert:101] Failed loading C:\Users\LV HENG\Desktop\vits-simple-api1205\bert_vits2/bert/bert-base-japanese-v3. You need to install fugashi to use MecabTokenizer. See https://pypi.org/project/fugashi/ for installation. 2023-12-06 22:19:16 [INFO] [bert_handler.load_bert:102] Trying to download. 2023-12-06 22:19:17 [INFO] [bert_handler._download_model:86] File already exists and verified successfully! 2023-12-06 22:19:17 [INFO] [bert_handler.load_bert:93] Loading BERT model: C:\Users\LV HENG\Desktop\vits-simple-api1205\bert_vits2/bert/bert-base-japanese-v3 2023-12-06 22:19:17 [ERROR] [bert_handler.load_bert:101] Failed loading C:\Users\LV HENG\Desktop\vits-simple-api1205\bert_vits2/bert/bert-base-japanese-v3. You need to install fugashi to use MecabTokenizer. See https://pypi.org/project/fugashi/ for installation. 2023-12-06 22:19:17 [INFO] [bert_handler.load_bert:102] Trying to download. 2023-12-06 22:19:17 [INFO] [bert_handler._download_model:86] File already exists and verified successfully! 2023-12-06 22:19:17 [ERROR] [bert_handler.load_bert:109] Failed to load C:\Users\LV HENG\Desktop\vits-simple-api1205\bert_vits2/bert/bert-base-japanese-v3 after 3 retries. 2023-12-06 22:19:17 [INFO] [bert_handler.load_bert:93] Loading BERT model: C:\Users\LV HENG\Desktop\vits-simple-api1205\bert_vits2/bert/chinese-roberta-wwm-ext-large Some weights of the model checkpoint at C:\Users\LV HENG\Desktop\vits-simple-api1205\bert_vits2/bert/chinese-roberta-wwm-ext-large were not used when initializing BertForMaskedLM: ['bert.pooler.dense.bias', 'cls.seq_relationship.bias', 'bert.pooler.dense.weight', 'cls.seq_relationship.weight']
目前主要怀疑错误有： 1.bert-base-japanese-v3未能正确加载（但好像不影响中文推理？）。 2.模型版本兼容性问题：刚开始推理出的结果是电流杂音，在config里添加"version": "1.1.0-transition"代码后，正确推理出人物音色，但仍然存在咬词不清的问题。 3.[ERROR] [utils.load_checkpoint:57] enc_p.emb.weight is not in the checkpoint 怀疑是这条影响了最终的推理结果。
烦请大佬赐教，不胜感激。 另：大佬有空可否考虑做一个纯bert-vits2的api接口？个人感觉bert-vits2到综合表现确实不错，非常感谢！