Segmentation fault (core dumped)

[purpleyun]

when run this code "w = tfe.define_private_variable(tf.ones(shape=(10,10)))", Segmentation fault (core dumped) occured.

[Cheng (Vince) Hong]

Could you provide your TF version, and the script you ran ?

[Zhicong Huang (Zico)]

when run this code "w = tfe.define_private_variable(tf.ones(shape=(10,10)))", Segmentation fault (core dumped) occured.

@purpleyun it would help to debug the issue if you could provide:

1. Detailed platform information with uname -a if you are on linux or macos
2. How you install the library? Building from source, or using pip install tf-encrypted?
3. TF version, script (showing which mpc protocol you use)

[Khaos Q]

I met the same problems, exactly in the same way. “when run this code "w = tfe.define_private_variable(tf.ones(shape=(10,10)))", Segmentation fault (core dumped) occured.“

all I did is trying to run the tutorial code in the readme.md.

my env:

1. platform is CentOS Linux release 7.8.2003 (Core)
2. i isntalled tfe by the 2nd way, from source, and build completed with no error.
3. python version is 3.6.5.
4. tf version is 1.15.2 as required.

please help. thx

[Cheng (Vince) Hong]

I met the same problems, exactly in the same way. “when run this code "w = tfe.define_private_variable(tf.ones(shape=(10,10)))", Segmentation fault (core dumped) occured.“

all I did is trying to run the tutorial code in the readme.md.

my env:

1. platform is CentOS Linux release 7.8.2003 (Core)
2. i isntalled tfe by the 2nd way, from source, and build completed with no error.
3. python version is 3.6.5.
4. tf version is 1.15.2 as required.

please help. thx

Could you check if the problem exists on tf 1.13.2 and tf 1.14.0 ？

[Khaos Q]

I met the same problems, exactly in the same way. “when run this code "w = tfe.define_private_variable(tf.ones(shape=(10,10)))", Segmentation fault (core dumped) occured.“
all I did is trying to run the tutorial code in the readme.md.
my env:

1. platform is CentOS Linux release 7.8.2003 (Core)
2. i isntalled tfe by the 2nd way, from source, and build completed with no error.
3. python version is 3.6.5.
4. tf version is 1.15.2 as required.

please help. thx

Could you check if the problem exists on tf 1.13.2 and tf 1.14.0 ？

Thank you for quick reply. I will take a try, but why the requirements said 1.15.2?

OK, I have tried both 1.13.2 and 1.14.0, they both work well. problem solved.
Thank you again, and could you do a favor explaining some of the deep reason for the problem in 1.15.2?

[Cheng (Vince) Hong]

I met the same problems, exactly in the same way. “when run this code "w = tfe.define_private_variable(tf.ones(shape=(10,10)))", Segmentation fault (core dumped) occured.“
all I did is trying to run the tutorial code in the readme.md.
my env:

1. platform is CentOS Linux release 7.8.2003 (Core)
2. i isntalled tfe by the 2nd way, from source, and build completed with no error.
3. python version is 3.6.5.
4. tf version is 1.15.2 as required.

please help. thx

Could you check if the problem exists on tf 1.13.2 and tf 1.14.0 ？

Thank you for quick reply. I will take a try, but why the requirements said 1.15.2?

OK, I have tried both 1.13.2 and 1.14.0, they both work well. problem solved. Thank you again, and could you do a favor explaining some of the deep reason for the problem in 1.15.2?

Actually I don't know why :( , I suggested that because I encountered a similar problem last year when using tf1.15. @zicofish @jvmncs got any idea in mind ?

[Khaos Q]

I met the same problems, exactly in the same way. “when run this code "w = tfe.define_private_variable(tf.ones(shape=(10,10)))", Segmentation fault (core dumped) occured.“
all I did is trying to run the tutorial code in the readme.md.
my env:

1. platform is CentOS Linux release 7.8.2003 (Core)
2. i isntalled tfe by the 2nd way, from source, and build completed with no error.
3. python version is 3.6.5.
4. tf version is 1.15.2 as required.

please help. thx

Could you check if the problem exists on tf 1.13.2 and tf 1.14.0 ？

Thank you for quick reply. I will take a try, but why the requirements said 1.15.2?
OK, I have tried both 1.13.2 and 1.14.0, they both work well. problem solved. Thank you again, and could you do a favor explaining some of the deep reason for the problem in 1.15.2?

Actually I don't know why :( , I suggested that because I encountered a similar problem last year when using tf1.15. @zicofish @jvmncs got any idea in mind ?

Anyway, thank you a lot for replying. Looking for further discussion.

[jvmncs]

got any idea in mind ?

Unfortunately nothing solid, I haven't been keeping up with the tf1 compatibility recently. Just looking at the readme code, I don't see anything obvious. One possibility I could imagine is if you built tensorflow with our patch for secure randomness. That secure randomness op uses their Custom Op API in tensorflow_core, but they may have introduced breaking changes to it in the jump from 1.14.0 -> 1.15.2, since that API does not have stability guarantees. I think they were interested in phasing that API out w/ TF2, because they had some other things in the works for adding new Ops to TF (maybe related to their work on TFRT).