Replies: 10 comments
-
Hey, this is the MXNet Label Bot. |
Beta Was this translation helpful? Give feedback.
-
FP16 in CPU is not supported now (FYI, OpenMathLib/OpenBLAS#694). But it should be a problem for the inconsistent error message. |
Beta Was this translation helpful? Give feedback.
-
@pengzhao-intel : Thanks for a quick response :) @mxnet-label-bot add [question] |
Beta Was this translation helpful? Give feedback.
-
@marekjg I can reproduce this issue from my side when |
Beta Was this translation helpful? Give feedback.
-
@pengzhao-intel I know it is not supported, problem is from the user point of view with cpp api and cpu device as it seems that everything worked just fine (the output array is created). |
Beta Was this translation helpful? Give feedback.
-
yes, it's a problem. Will fix it. |
Beta Was this translation helpful? Give feedback.
-
Let's see why LOG(FATAL) doesn't work in here :) |
Beta Was this translation helpful? Give feedback.
-
@sandeep-krishnamurthy we had somebody who worked on exception for single- and multi-threaded execution, but I can't recall who. Can you assist here? |
Beta Was this translation helpful? Give feedback.
-
FYI, when commenting out this line in dmlc/logging.h, the error stack trace will be printed normally for |
Beta Was this translation helpful? Give feedback.
-
@marcoabreu I worked on exception handling for the backend. Having said that I am not very familiar with the CPP frontend language binding, and it looks like this issue is specific to CPP binding. The error code is not checked and exception is not being rethrown in the frontend language binding. cc @leleamol who worked on the CPP language binding. |
Beta Was this translation helpful? Give feedback.
-
Description
When running inference in cpp api with naive engine, the computation is not performed but the output array is initialized and no exception is raised. Without naive engine, there's exception thrown (which is expected). Using python api with naive engine will raise exception in both cases so the problem seems to be related to CPP api. It applies to Dense layer but also happend with model with RNN layers so my guess is that whenever GEMM is used, it fails.
Environment info (Required)
Build info (Required if built from source)
gcc
MXNet commit hash:
57927a9
Build config:
cmake -DUSE_CUDA=OFF -DUSE_CPP_PACKAGE=1 -GNinja .. && ninja
Error Message:
With NaiveEngine:
Without NaiveEngine:
Minimum reproducible example and steps to reproduce
Create the model on the machine with GPU (linear-symbol.json and linear-0000.params files) with the following script:
run the inference with the following cpp program:
Steps to reproduce
MXNET_ENGINE_TYPE="NaiveEngine" ./test
and observe no error./test
and observe the error with message that fp16 fully connected layer is not supported in CPUBeta Was this translation helpful? Give feedback.
All reactions