显示forward_ algorithm cache is full和backward w algorithm cache is full对训练有影响吗 - Githubissues

Jittor / jittor

Jittor is a high-performance deep learning framework based on JIT compiling and meta-operators.

https://cg.cs.tsinghua.edu.cn/jittor/

Apache License 2.0

3.07k stars 310 forks source link

显示forward_ algorithm cache is full和backward w algorithm cache is full对训练有影响吗 #386

Open wzr0108 opened 2 years ago

wzr0108 commented 2 years ago

[38;5;3m[w 0910 19:24:15.321940 48 cudnn_conv__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFORMAT_abcd_____hash_4d5b3e2d24c769d3_op.cc:213] forward_ algorithm cache is full[m
[38;5;3m[w 0910 19:24:15.391788 84 cudnn_conv__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFORMAT_abcd_____hash_4d5b3e2d24c769d3_op.cc:213] forward_ algorithm cache is full[m
[38;5;3m[w 0910 19:24:15.458474 52 cudnn_conv__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFORMAT_abcd_____hash_4d5b3e2d24c769d3_op.cc:213] forward_ algorithm cache is full[m
[38;5;3m[w 0910 19:24:15.471886 36 cudnn_conv__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFORMAT_abcd_____hash_4d5b3e2d24c769d3_op.cc:213] forward_ algorithm cache is full[m
[38;5;3m[w 0910 19:24:17.551503 84 cudnn_conv_backward_w__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_8e480e8564e59906_op.cc:212] backward w algorithm cache is full[m
[38;5;3m[w 0910 19:24:17.738381 48 cudnn_conv_backward_w__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_8e480e8564e59906_op.cc:212] backward w algorithm cache is full[m
[38;5;3m[w 0910 19:24:17.769248 84 cudnn_conv_backward_x__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_af8994a8aef53c1c_op.cc:208] backward x algorithm cache is full[m
[38;5;3m[w 0910 19:24:17.868283 48 cudnn_conv_backward_x__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_af8994a8aef53c1c_op.cc:208] backward x algorithm cache is full[m
[38;5;3m[w 0910 19:24:17.973627 36 cudnn_conv_backward_w__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_8e480e8564e59906_op.cc:212] backward w algorithm cache is full[m
[38;5;3m[w 0910 19:24:18.226108 36 cudnn_conv_backward_x__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_af8994a8aef53c1c_op.cc:208] backward x algorithm cache is full[m
[38;5;3m[w 0910 19:24:18.388955 52 cudnn_conv_backward_w__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_8e480e8564e59906_op.cc:212] backward w algorithm cache is full[m
[38;5;3m[w 0910 19:24:19.897344 60 cudnn_conv__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFORMAT_abcd_____hash_4d5b3e2d24c769d3_op.cc:213] forward_ algorithm cache is full[m
[38;5;3m[w 0910 19:24:20.219949 68 cudnn_conv__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFORMAT_abcd_____hash_4d5b3e2d24c769d3_op.cc:213] forward_ algorithm cache is full[m
[38;5;3m[w 0910 19:24:21.271607 36 cudnn_conv__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFORMAT_abcd_____hash_4d5b3e2d24c769d3_op.cc:213] forward_ algorithm cache is full[m
[38;5;3m[w 0910 19:24:22.014836 68 cudnn_conv_backward_w__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_8e480e8564e59906_op.cc:212] backward w algorithm cache is full[m
[38;5;3m[w 0910 19:24:22.157015 60 cudnn_conv_backward_w__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_8e480e8564e59906_op.cc:212] backward w algorithm cache is full[m
[38;5;3m[w 0910 19:24:22.420667 60 cudnn_conv_backward_x__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_af8994a8aef53c1c_op.cc:208] backward x algorithm cache is full[m
[38;5;3m[w 0910 19:24:22.451356 68 cudnn_conv_backward_x__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_af8994a8aef53c1c_op.cc:208] backward x algorithm cache is full[m
[38;5;3m[w 0910 19:24:22.647406 36 cudnn_conv_backward_w__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_8e480e8564e59906_op.cc:212] backward w algorithm cache is full[m
[38;5;3m[w 0910 19:24:23.436515 36 cudnn_conv_backward_x__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_af8994a8aef53c1c_op.cc:208] backward x algorithm cache is full[m
[38;5;3m[w 0910 19:24:27.669546 52 cudnn_conv_backward_x__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_af8994a8aef53c1c_op.cc:208] backward x algorithm cache is full[m
[38;5;3m[w 0910 19:24:31.177295 08 cudnn_conv__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFORMAT_abcd_____hash_4d5b3e2d24c769d3_op.cc:213] forward_ algorithm cache is full[m
[38;5;3m[w 0910 19:24:32.823906 08 cudnn_conv_backward_w__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_8e480e8564e59906_op.cc:212] backward w algorithm cache is full[m
[38;5;3m[w 0910 19:24:32.870152 08 cudnn_conv_backward_x__Tx_float32__Ty_float32__Tw_float32__XFORMAT_abcd__WFORMAT_oihw__YFO___hash_af8994a8aef53c1c_op.cc:208] backward x algorithm cache is full[m

lzhengning commented 2 years ago

您好，algorithm cache 是用于存储不同卷积计算方法的缓存。cache full 常出现于输入形状不确定的网络中，对训练结果没有影响。