首页 Paddle框架 帖子详情
deepspeech tune的性能问题
收藏
快速回复
Paddle框架 问答模型训练 1403 2
deepspeech tune的性能问题
收藏
快速回复
Paddle框架 问答模型训练 1403 2

配置如下。大部分时间,gpu负载为0,cpu也是一核有难多核围观的局面。是我配置有误吗?

----------- Configuration Arguments -----------
alpha_from: 1.0
alpha_to: 3.2
batch_size: 16
beam_size: 300
beta_from: 1.0
beta_to: 7.0
cutoff_prob: 0.99
cutoff_top_n: 40
error_rate_type: cer
lang_model_path: models/lm/zh_giga.no_cna_cmn.prune01244.klm
mean_std_path: data/aishell/mean_std.npz
model_path: checkpoints/aishell/params.latest.tar.gz
num_alphas: 45
num_batches: -1
num_betas: 45
num_conv_layers: 2
num_proc_bsearch: 12
num_proc_data: 12
num_rnn_layers: 3
rnn_layer_size: 1024
share_rnn_weights: 0
specgram_type: linear
trainer_count: 2
tune_manifest: data/aishell/manifest.dev
use_gpu: 1
use_gru: 1
vocab_path: data/aishell/vocab.txt
------------------------------------------------
I0905 16:38:30.185864 12425 Util.cpp:166] commandline: --use_gpu=1 --rnn_use_batch=True --trainer_count=2
[INFO 2018-09-05 16:38:31,620 layers.py:2716] output for __conv_0__: c = 32, h = 81, w = 54, size = 139968
[INFO 2018-09-05 16:38:31,620 layers.py:3361] output for __batch_norm_0__: c = 32, h = 81, w = 54, size = 139968
[INFO 2018-09-05 16:38:31,621 layers.py:7533] output for __scale_sub_region_0__: c = 32, h = 81, w = 54, size = 139968
[INFO 2018-09-05 16:38:31,621 layers.py:2716] output for __conv_1__: c = 32, h = 41, w = 54, size = 70848
[INFO 2018-09-05 16:38:31,621 layers.py:3361] output for __batch_norm_1__: c = 32, h = 41, w = 54, size = 70848
[INFO 2018-09-05 16:38:31,622 layers.py:7533] output for __scale_sub_region_1__: c = 32, h = 41, w = 54, size = 70848
[INFO 2018-09-05 16:38:35,315 model.py:243] begin to initialize the external scorer for decoding
[INFO 2018-09-05 16:38:35,380 model.py:253] language model: is_character_based = 1, max_order = 5, dict_size = 0
[INFO 2018-09-05 16:38:35,380 model.py:254] end initializing scorer
[INFO 2018-09-05 16:38:35,380 tuneaishell.py:122] start tuning ...
I0905 16:38:36.057443 12425 MultiGradientMachine.cpp:99] numLogicalDevices=1 numThreads=2 numDevices=2

0
收藏
回复
全部评论(2)
时间顺序
夜雨飘零1
#2 回复于2018-09

你的batch size大一些,占更多的显存,可能效果会更好一些。

0
回复
A
AIStudio783003
#3 回复于2018-09
你的batch size大一些,占更多的显存,可能效果会更好一些。

已经训练了3天,aishell数据,不知道还要多久。如果中途失败了,我就调整参数重来。

0
回复
需求/bug反馈?一键提issue告诉我们
发现bug?如果您知道修复办法,欢迎提pr直接参与建设飞桨~
在@后输入用户全名并按空格结束,可艾特全站任一用户