microsoft / ELL

Embedded Learning Library
https://microsoft.github.io/ELL
Other
2.29k stars 295 forks source link

While using .H and .S file in the MXchip the output is not correct but on desktop it is working #225

Closed mr-yamraj closed 5 years ago

mr-yamraj commented 5 years ago

the .ell file is working perfect on the desktop.

but when I implemented it on MXCHIP, It is not working well. When I crosschecked the output in mx chip and the desktop. The initial hidden values which are calculated using the module created by .ell file are correct till the time they are positive. But when a negative value is expected, It is showing incorrect positive value.

Take a look at the screenshot for description. (LEFT window is MXChip prediction using the generated .H and .S file and Right terminal is the output by running the build .ell file on desktop)

Screenshot (3)

I am using the below commands to convert the .ell to .H and .S

/home/yash/ELL/build/bin/compile -imap model.ell -cfn Prediction -cmn completemodel --bitcode -od . --fuseLinearOps True --header --blas false --optimize true --target custom --numBits 32 --cpu cortex-m4 --triple armv6m-gnueabi --features +vfp4,+d16,+soft-float
/usr/lib/llvm-8/bin/opt model.bc -o model.opt.bc -O3
/usr/lib/llvm-8/bin/llc model.opt.bc -o model.S -O3 -filetype=asm -mtriple=armv6m-gnueabi -mcpu=cortex-m4 -relocation-model=pic -float-abi=soft -mattr=+vfp4,+d16

I am attaching my ell It is basically the implementation of audio make_featuriser and fastgrnn prediction. It works well on desktop when I compared the output. But not working well on ELL. Let me know if you need more details. I am attaching my model.ell file model.zip

mr-yamraj commented 5 years ago

I am sorry It was a mistake in my code of MX chip.