babysor / MockingBird

🚀AI拟声: 5秒内克隆您的声音并生成任意语音内容 Clone a voice in 5 seconds to generate arbitrary speech in real-time
Other
35.27k stars 5.21k forks source link

保姆级别教程(持续更新各类社区/非官方教程---- #20

Open zhuzaileiting opened 3 years ago

zhuzaileiting commented 3 years ago

(作者借楼编辑ing 社区视频教程: 奶糖 https://www.bilibili.com/video/BV1dq4y137pH

babysor commented 3 years ago

再分享 https://github.com/babysor/MockingBird/issues/122

zhuzaileiting commented 3 years ago

播放都是杂音。。。。。。。。。。。。。 C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main>python synthesizer_train.py mandarin E:\datat\rain_set\train\ Arguments:     run_id:          mandarin     syn_dir:         E:\datat\rain_set\train\     models_dir:      synthesizer/saved_models/     save_every:      1000     backup_every:    25000     force_restart:   False     hparams:

Checkpoint path: synthesizer\saved_models\mandarin\mandarin.pt Loading training data from: E:\datat\rain_set\train\train.txt Using model: Tacotron Using device: cpu

Initialising Tacotron Model...

Trainable Parameters: 30.872M

Loading weights at synthesizer\saved_models\mandarin\mandarin.pt Tacotron weights loaded from step 0 Using inputs from:         E:\datat\rain_set\train\train.txt         E:\datat\rain_set\train\mels         E:\datat\rain_set\train\embeds Traceback (most recent call last):   File "synthesizer_train.py", line 35, in <module>     train(**vars(args))   File "C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main\synthesizer\train.py", line 111, in train     dataset = SynthesizerDataset(metadata_fpath, mel_dir, embed_dir, hparams)   File "C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main\synthesizer\synthesizer_dataset.py", line 12, in init     with metadata_fpath.open("r", encoding="utf-8") as metadata_file:   File "C:\ProgramData\Anaconda3\lib\pathlib.py", line 1221, in open     return io.open(self, mode, buffering, encoding, errors, newline,   File "C:\ProgramData\Anaconda3\lib\pathlib.py", line 1077, in _opener     return self._accessor.open(self, flags, mode) FileNotFoundError: [Errno 2] No such file or directory: 'E:\datat\rain_set\train\train.txt'

C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main>python demo_toolbox.py -d E:\data\train_set\train 2021-08-19 17:56:17.809226: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'cudart64_110.dll'; dlerror: cudart64_110.dll not found 2021-08-19 17:56:17.809396: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine. Arguments:     datasets_root:    E:\data\train_set\train     enc_models_dir:   encoder\saved_models     syn_models_dir:   synthesizer\saved_models     voc_models_dir:   vocoder\saved_models     cpu:              False     seed:             None     no_mp3_support:   False

Warning: you do not have any of the recognized datasets in E:\data\train_set\train. The recognized datasets are:         LibriSpeech/dev-clean         LibriSpeech/dev-other         LibriSpeech/test-clean         LibriSpeech/test-other         LibriSpeech/train-clean-100         LibriSpeech/train-clean-360         LibriSpeech/train-other-500         LibriTTS/dev-clean         LibriTTS/dev-other         LibriTTS/test-clean         LibriTTS/test-other         LibriTTS/train-clean-100         LibriTTS/train-clean-360         LibriTTS/train-other-500         LJSpeech-1.1         VoxCeleb1/wav         VoxCeleb1/test_wav         VoxCeleb2/dev/aac         VoxCeleb2/test/aac         VCTK-Corpus/wav48         aidatatang_200zh/corpus/dev         aidatatang_200zh/corpus/test Feel free to add your own. You can still use the toolbox by recording samples yourself. Loaded encoder "pretrained.pt" trained to step 1564501 Synthesizer using device: cpu Trainable Parameters: 30.872M Traceback (most recent call last):   File "C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main\toolbox__init.py", line 122, in <lambda>     func = lambda: self.synthesize() or self.vocode()   File "C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main\toolbox\init__.py", line 229, in synthesize     specs = self.synthesizer.synthesize_spectrograms(texts, embeds)   File "C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main\synthesizer\inference.py", line 86, in synthesize_spectrograms     self.load()   File "C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main\synthesizer\inference.py", line 64, in load     self._model.load(self.model_fpath)   File "C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main\synthesizer\models\tacotron.py", line 497, in load     self.load_state_dict(checkpoint["model_state"])   File "C:\ProgramData\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 1223, in load_state_dict     raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( RuntimeError: Error(s) in loading state_dict for Tacotron:         size mismatch for encoder.embedding.weight: copying a param with shape torch.Size([75, 512]) from checkpoint, the shape in current model is torch.Size([70, 512]).

| Generating 1/1

Done.

------------------ 原始邮件 ------------------ 发件人: "babysor/Realtime-Voice-Clone-Chinese" @.>; 发送时间: 2021年8月19日(星期四) 下午5:57 @.>; @.>;"State @.>; 主题: Re: [babysor/Realtime-Voice-Clone-Chinese] 来个保姆级别教程@@ (#20)

别害羞,快分享一下卡在哪里啦,我再优化优化

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

zhuzaileiting commented 3 years ago

出一个详细的教程吧,大佬👍

---原始邮件--- 发件人: @.> 发送时间: 2021年8月19日(周四) 下午5:57 收件人: @.>; 抄送: @.>;"State @.>; 主题: Re: [babysor/Realtime-Voice-Clone-Chinese] 来个保姆级别教程@@ (#20)

别害羞,快分享一下卡在哪里啦,我再优化优化

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

babysor commented 3 years ago

这。。看起来你都没train起来synthesizer啊

zhangji261 commented 3 years ago

同求,比如数据集在哪里下载

XiuChen-Liu commented 3 years ago

同求,比如数据集在哪里下载

14 closed裡面有同樣的問題,有放下載連結

zhuzaileiting commented 3 years ago

E:\data\aidatatang_200zh\aidatatang_200zh\corpus\train 数据集解压路径 这一步synthesizer_preprocess_audio.py有问题吗 C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main> python synthesizer_preprocess_audio.py E:\data\aidatatang_200zh\aidatatang_200zh Arguments: datasets_root: E:\data\aidatatang_200zh\aidatatang_200zh out_dir: E:\data\aidatatang_200zh\aidatatang_200zh\SV2TTS\synthesizer n_processes: None skip_existing: False hparams: no_alignments: False dataset: aidatatang_200zh

Using data from: E:\data\aidatatang_200zh\aidatatang_200zh\aidatatang_200zh\corpus\train Traceback (most recent call last): File "synthesizer_preprocess_audio.py", line 63, in preprocess_dataset(**vars(args)) File "C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main\synthesizer\preprocess.py", line 32, in preprocess_dataset assert all(input_dir.exists() for input_dir in input_dirs) AssertionError

babysor commented 3 years ago

E:\data\aidatatang_200zh\aidatatang_200zh\corpus\train 数据集解压路径 这一步synthesizer_preprocess_audio.py有问题吗 C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main> python synthesizer_preprocess_audio.py E:\data\aidatatang_200zh\aidatatang_200zh Arguments: datasets_root: E:\data\aidatatang_200zh\aidatatang_200zh out_dir: E:\data\aidatatang_200zh\aidatatang_200zh\SV2TTS\synthesizer n_processes: None skip_existing: False hparams: no_alignments: False dataset: aidatatang_200zh

Using data from: E:\data\aidatatang_200zh\aidatatang_200zh\aidatatang_200zh\corpus\train Traceback (most recent call last): File "synthesizer_preprocess_audio.py", line 63, in preprocess_dataset(**vars(args)) File "C:\Users\Administrator\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main\synthesizer\preprocess.py", line 32, in preprocess_dataset assert all(input_dir.exists() for input_dir in input_dirs) AssertionError

python synthesizer_preprocess_audio.py E:\data\aidatatang_200zh 不用多一层

zhuzaileiting commented 3 years ago

python synthesizer_preprocess_audio.py E:\data\aidatatang_200zh 不用多一层 解决了

zhuzaileiting commented 3 years ago

python synthesizer_preprocess_audio.py E:\data\aidatatang_200zh 不用多一层 @解决了大佬

zhuzaileiting commented 3 years ago

这步也太慢了。。。 C:\Users\lxd\Desktop\Realtime-Voice-Clone-Chinese-main\Realtime-Voice-Clone-Chinese-main>python synthesizer_train.py mandarin D:\data\aidatatang_200zh\SV2TTS\synthesizer Arguments: run_id: mandarin syn_dir: D:\data\aidatatang_200zh\SV2TTS\synthesizer models_dir: synthesizer/saved_models/ save_every: 1000 backup_every: 25000 force_restart: False hparams:

Checkpoint path: synthesizer\saved_models\mandarin\mandarin.pt Loading training data from: D:\data\aidatatang_200zh\SV2TTS\synthesizer\train.txt Using model: Tacotron Using device: cpu

Initialising Tacotron Model...

Trainable Parameters: 30.872M

Loading weights at synthesizer\saved_models\mandarin\mandarin.pt Tacotron weights loaded from step 0 Using inputs from: D:\data\aidatatang_200zh\SV2TTS\synthesizer\train.txt D:\data\aidatatang_200zh\SV2TTS\synthesizer\mels D:\data\aidatatang_200zh\SV2TTS\synthesizer\embeds Found 122482 samples +----------------+------------+---------------+------------------+ | Steps with r=2 | Batch Size | Learning Rate | Outputs/Step (r) | +----------------+------------+---------------+------------------+ | 20k Steps | 12 | 0.001 | 2 | +----------------+------------+---------------+------------------+

{| Epoch: 1/2 (500/10207) | Loss: 0.9025 | 0.065 steps/s | Step: 0k | }Input at step 500: wo3 yao4 gei3 wang2 ming2 da3 dian4 hua4~__ {| Epoch: 1/2 (1000/10207) | Loss: 0.8266 | 0.071 steps/s | Step: 1k | }Input at step 1000: na4 me wo3 jiu4 chong2 xin1 ren4 shi2 ni3~__ {| Epoch: 1/2 (1500/10207) | Loss: 0.7602 | 0.074 steps/s | Step: 1k | }Input at step 1500: mei3 tian1 dou1 na4 me wan3 shui4 jiao4~_ {| Epoch: 1/2 (2000/10207) | Loss: 0.7415 | 0.075 steps/s | Step: 2k | }Input at step 2000: da3 dian4 hua4 gei3 deng4 han4 ling2~___ {| Epoch: 1/2 (2500/10207) | Loss: 0.6921 | 0.068 steps/s | Step: 2k | }Input at step 2500: zhen1 xiang4 yong3 yuan3 zhi3 you3 yi2 ge4~ {| Epoch: 1/2 (3000/10207) | Loss: 0.6741 | 0.072 steps/s | Step: 3k | }Input at step 3000: xia4 men2 wai4 guo2 yu3 xue2 xiao4 chu1 er4 nian2 ji2 chen2 xiao3 qi2 jia1 de zhu4 zhi3~ {| Epoch: 1/2 (3500/10207) | Loss: 0.6499 | 0.070 steps/s | Step: 3k | }Input at step 3500: ru2 guo3 wo3 he2 ni3 zai4 yi4 qi3~____ {| Epoch: 1/2 (4000/10207) | Loss: 0.6679 | 0.073 steps/s | Step: 4k | }Input at step 4000: fu4 jin4 de ping2 an1 yin2 hang2~ {| Epoch: 1/2 (4500/10207) | Loss: 0.6349 | 0.069 steps/s | Step: 4k | }Input at step 4500: ming2 zi4 shi4 hui3 guo4 cheng2 nuo4 shu1~____ {| Epoch: 1/2 (5000/10207) | Loss: 0.6392 | 0.073 steps/s | Step: 5k | }Input at step 5000: wo3 shen2 me shi2 hou4 cai2 neng2 chong1 man3 dian4~____ {| Epoch: 1/2 (5500/10207) | Loss: 0.6293 | 0.073 steps/s | Step: 5k | }Input at step 5500: wo3 da3 ni3 hao3 bu4 hao3 ma ge2 shi4 chong2 fu4~ {| Epoch: 1/2 (6000/10207) | Loss: 0.6715 | 0.077 steps/s | Step: 6k | }Input at step 6000: ci3 ji4 hao3 wu2 liao2 da3 yi1 dian4 ying3 ming2~ {| Epoch: 1/2 (6500/10207) | Loss: 0.6446 | 0.075 steps/s | Step: 6k | }Input at step 6500: wo3 gei3 ni3 fa1 de ni3 shou1 dao4 le ma~____ {| Epoch: 1/2 (7000/10207) | Loss: 0.6022 | 0.068 steps/s | Step: 7k | }Input at step 7000: ning4 que1 wu2 lan4 zhi3 wei4 yi3 hou4 de du2 yi1 wu2 er4~____ {| Epoch: 1/2 (7500/10207) | Loss: 0.6178 | 0.067 steps/s | Step: 7k | }Input at step 7500: mei2 you3 wang3 luo4 ni3 hai2 hui4 liao2 tian1 ma~__ {| Epoch: 1/2 (8000/10207) | Loss: 0.6041 | 0.068 steps/s | Step: 8k | }Input at step 8000: wo3 bu4 fa1 le wo3 yao4 shui4 jiao4 le~____ {| Epoch: 1/2 (8500/10207) | Loss: 0.6078 | 0.072 steps/s | Step: 8k | }Input at step 8500: ni3 cai1 lai2 cai1 qu4 ye3 cai1 bu4 ming2 bai2~____ {| Epoch: 1/2 (9000/10207) | Loss: 0.6055 | 0.072 steps/s | Step: 9k | }Input at step 9000: ni3 wen4 le wo3 tou2 dou1 da4 le~ {| Epoch: 1/2 (9500/10207) | Loss: 0.5816 | 0.069 steps/s | Step: 9k | }Input at step 9500: xia4 ban1 mei2 you3 mei2 chu1 qu4 guang4~ {| Epoch: 1/2 (10000/10207) | Loss: 0.5664 | 0.068 steps/s | Step: 10k | }Input at step 10000: ni3 jin1 tian1 bu2 shi4 bu4 shang4 ban1 ma~__ {| Epoch: 1/2 (10207/10207) | Loss: 0.5879 | 0.071 steps/s | Step: 10k | } {| Epoch: 2/2 (293/10207) | Loss: 0.5840 | 0.070 steps/s | Step: 10k | }Input at step 10500: ai4 qing2 xiao3 shuo1 ma2 que4 gao3 ding4 hua1 mei3 nan2~_____ {| Epoch: 2/2 (322/10207) | Loss: 0.5892 | 0.070 steps/s | Step: 10k | }

miven commented 3 years ago

@zhuzaileiting 你这是用cpu训练的,GPU速度大概在1.3-2 steps/s

zhuzaileiting commented 3 years ago

...GPU,怎么配置显卡全局配置了呀。。

dukechain2333 commented 3 years ago

Arguments: datasets_root: D:\Data\aidatatang_200zh out_dir: D:\Data\aidatatang_200zh\SV2TTS\synthesizer n_processes: None skip_existing: False hparams: no_alignments: False dataset: aidatatang_200zh

Using data from: D:\Data\aidatatang_200zh\aidatatang_200zh\corpus\train Traceback (most recent call last): File "synthesizer_preprocess_audio.py", line 63, in preprocess_dataset(**vars(args)) File "D:\Realtime-Voice-Clone-Chinese\synthesizer\preprocess.py", line 32, in preprocess_dataset assert all(input_dir.exists() for input_dir in input_dirs) AssertionError

dukechain2333 commented 3 years ago

路径好像没啥问题啊

dukechain2333 commented 3 years ago

image trian里是长这样的嘛

yuzd commented 3 years ago

有群吗 一起交流下怎么跑

babysor commented 3 years ago

WechatIMG54 7天有效

hexiosr commented 3 years ago

WechatIMG54 7天有效

二维码过期了

oceanarium commented 3 years ago

WechatIMG54 7天有效

二维码过期了

5d90f925edfbdc0ba89474faf94a370

josh-zhu commented 3 years ago

群二维码过期了,求更

JackChow6 commented 3 years ago

二维码过期了,求更

babysor commented 3 years ago

adcd56d69910086b43053bbfa804a51

babysor commented 3 years ago

群二维码过期了,求更

见上

JackChow6 commented 3 years ago

群二维码过期了,求更

见上

谢谢你

chloe5685 commented 3 years ago

二维码失效了呜呜呜

babysor commented 3 years ago

f7df5a1addc744a66b1dae512f3db26

@chloe5685

babysor commented 3 years ago

cb0f71d680c3b7d6c2aca6fce1a220d

招募志愿者一名, multilang writer preferred. 包教会和调优,整理后输出教程到社区。

Shaohan777 commented 3 years ago

群200人了 求群成员邀请进群。。我的wx是Sahvyhsu

eliaszon commented 3 years ago

用telegram更好

babysor commented 3 years ago

用telegram更好

辛苦创建一个?

babysor commented 3 years ago

For non-trainer: https://github.com/babysor/MockingBird/wiki For trainer: https://vaj2fgg8yn.feishu.cn/docs/doccn7kAbr3SJz0KM0SIDJ0Xnhd#

cyist commented 3 years ago

D:\Downloads\MockingBird-main>python demo_toolbox.py -d "D:\Downloads\aidatatang_200zh" 2021-10-01 14:32:20.420728: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'cudart64_110.dll'; dlerror: cudart64_110.dll not found 2021-10-01 14:32:20.420998: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine. Arguments: datasets_root: D:\Downloads\aidatatang_200zh enc_models_dir: encoder\saved_models syn_models_dir: synthesizer\saved_models voc_models_dir: vocoder\saved_models cpu: False seed: None no_mp3_support: False

Traceback (most recent call last): File "D:\Downloads\MockingBird-main\demo_toolbox.py", line 43, in Toolbox(**vars(args)) File "D:\Downloads\MockingBird-main\toolbox__init.py", line 74, in init__ self.reset_ui(enc_models_dir, syn_models_dir, voc_models_dir, seed) File "D:\Downloads\MockingBird-main\toolbox__init__.py", line 142, in reset_ui self.ui.populate_models(encoder_models_dir, synthesizer_models_dir, vocoder_models_dir) File "D:\Downloads\MockingBird-main\toolbox\ui.py", line 339, in populate_models raise Exception("No synthesizer models found in %s" % synthesizer_models_dir) Exception: No synthesizer models found in synthesizer\saved_models

Teresa518 commented 3 years ago

求群成员邀请入群,wx是Sinohoney0002

ZhiMasu commented 3 years ago

求群成员邀请进群。。我的wx是myggzhsdd1

ZhiMasu commented 3 years ago

python demo_toolbox.py -d E:\Fun\datasets Arguments: datasets_root: E:\Fun\datasets enc_models_dir: encoder\saved_models syn_models_dir: synthesizer\saved_models voc_models_dir: vocoder\saved_models cpu: False seed: None no_mp3_support: False

Warning: you do not have any of the recognized datasets in E:\Fun\datasets. The recognized datasets are: LibriSpeech/dev-clean LibriSpeech/dev-other LibriSpeech/test-clean LibriSpeech/test-other LibriSpeech/train-clean-100 LibriSpeech/train-clean-360 LibriSpeech/train-other-500 LibriTTS/dev-clean LibriTTS/dev-other LibriTTS/test-clean LibriTTS/test-other LibriTTS/train-clean-100 LibriTTS/train-clean-360 LibriTTS/train-other-500 LJSpeech-1.1 VoxCeleb1/wav VoxCeleb1/test_wav VoxCeleb2/dev/aac VoxCeleb2/test/aac VCTK-Corpus/wav48 aidatatang_200zh/corpus/dev aidatatang_200zh/corpus/test aishell3/test/wav magicdata/train Feel free to add your own. You can still use the toolbox by recording samples yourself. 这里出了什么问题,进去之后数据那两行是灰色的。求教QAQ

ANINwindows commented 3 years ago

@zhuzaileiting 你这是用cpu训练的,GPU速度大概在1.3-2 steps/s

您好,我根据您的文档修改了Batch size调整到36,现在显存已经使用80%,但是显卡使用率还是只有12%,我还需要调整哪个参数呢?

AceLuodan commented 3 years ago

求群成员邀请入群,wx是luo_dan_lwts

tzzhmm commented 3 years ago

求群成员邀请入群,wx是Mr-Sandman___

babysor commented 3 years ago

30611585-CD17-4271-BCA2-4C67A4ACBEE6 建了二群 方便交流

Teresa518 commented 3 years ago

田渣渣 @.***> 于 2021年10月9日周六 下午1:23写道:

求群成员邀请入群,wx是Mr-Sandman___

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/babysor/MockingBird/issues/20#issuecomment-939230399, or unsubscribe https://github.com/notifications/unsubscribe-auth/AS6A2HF5F57YOZ4N2754TLDUF7GUVANCNFSM5CN2A35Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

babysor commented 3 years ago

13BD7788-B0C2-460D-B41F-3BB3F433D936

willis commented 3 years ago

@zhuzaileiting 你这是用cpu训练的,GPU速度大概在1.3-2 steps/s

您好,我根据您的文档修改了Batch size调整到36,现在显存已经使用80%,但是显卡使用率还是只有12%,我还需要调整哪个参数呢?

我3090,设置为48,0.86 setps/s,达不到1.3-2

babysor commented 3 years ago

@zhuzaileiting 你这是用cpu训练的,GPU速度大概在1.3-2 steps/s

您好,我根据您的文档修改了Batch size调整到36,现在显存已经使用80%,但是显卡使用率还是只有12%,我还需要调整哪个参数呢?

我3090,设置为48,0.86 setps/s,达不到1.3-2

不同batch size,step速度没有可比较性。

willis commented 3 years ago

@zhuzaileiting 你这是用cpu训练的,GPU速度大概在1.3-2 steps/s

您好,我根据您的文档修改了Batch size调整到36,现在显存已经使用80%,但是显卡使用率还是只有12%,我还需要调整哪个参数呢?

我3090,设置为48,0.86 setps/s,达不到1.3-2

不同batch size,step速度没有可比较性。

和数据集大小也有关吧

babysor commented 3 years ago

@zhuzaileiting 你这是用cpu训练的,GPU速度大概在1.3-2 steps/s

您好,我根据您的文档修改了Batch size调整到36,现在显存已经使用80%,但是显卡使用率还是只有12%,我还需要调整哪个参数呢?

我3090,设置为48,0.86 setps/s,达不到1.3-2

不同batch size,step速度没有可比较性。

和数据集大小也有关吧

是的

Olivia-Ye commented 3 years ago

能再发一下群吗 过期了耶

tonytonyma commented 3 years ago

二维码失效了

babysor commented 3 years ago

3e121fa4b6044f11a63010c34abf50f

babysor commented 3 years ago

效果提升小技巧合集1 - vega的文章 https://zhuanlan.zhihu.com/p/425692267