New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
C++ compilation of rule '//tensorflow/python:bfloat16_lib' failed #40688
Comments
Possibly duplicate of #40654? I'm also seeing the same issue with |
Thank you :) Will report if this fixes the build |
Confirming that this solves the
Going to check for 2.20 and 10.2 then 1.15.3 and 10.2 |
In our testing, we found that this issue breaks building from source for TF 1.15.x and 2.x. The issue comes from source build being incompatible with numpy Fixing numpy to pre 1.19.0 fixes the issue:
|
Fix numpy to pre-1.19.0 because of breaking ABI change in numpy 1.19.0 (numpy/numpy#15355) See tensorflow/tensorflow#40688.
Thank you, will force Also confirming 2.20 and 10.2 compiles with the |
Confirming successful compilation on 2.20 and 10.2 with Okay to close the issue. I have different problems with 1.15.3 and |
Marking the issue as closed, as it is resolved. Please feel free to re-open the issue if required. Thanks! |
Hi, @mmartial. I also run into the same issue. I've downgraded numpy to 1.18.5 but it did not fix the problem. Here's the error message I received
It seems related to Here's my environment information: Ubuntu: 18.04 And here's the output of certifi 2020.6.20 |
@xlnwel Looking at the above, I wonder: did you use both the PR (or the Perl command) and the To make it work, I had to use either of those. I am putting below the updated
|
Hi @mmartial. I build TF2.2 following the official guide without Dockerfile. Do you mean I should execute |
No, I was referring to #40688 (comment) Note that simply using |
Unfortunately it does not work for me. Maybe I have to open another issue. |
See #40688. PiperOrigin-RevId: 318122157 Change-Id: Ief46c5610f3aaf0cdd7d43ce1a10d6d87e8e8e01
See tensorflow#40688. PiperOrigin-RevId: 318122157 Change-Id: Ief46c5610f3aaf0cdd7d43ce1a10d6d87e8e8e01
The new version of numpy seems to have some API change that doesn't work with TF anymore. See tensorflow/tensorflow#40688 (comment)
See tensorflow issue at tensorflow/tensorflow#40688
@amahendrakar this is still an issue on r2.3; i just tried to build tf branch r2.3 on my ubuntu system and ran into the same issue; the perl rewrite works, we should just fix the code to do a proper static cast. @penpornk who's closest to this code? |
@ebrevdo This is Python glue code so it probably belongs to TF Core folks. But the fixes are simple enough. I can do it. |
It seems @chsigg has already fixed this in 75ea0b3 recently (Jun 26, 2020) by adding an overload function. I tried compiling with the latest code from master and didn't get the error anymore. (It's too late to patch this into releases 2.2.0 and 2.3.0 now, so this issue will be fixed in release 2.4.0.) |
Thank you for the update!
…On Tue, Aug 4, 2020 at 5:40 PM Penporn Koanantakool < ***@***.***> wrote:
It seems @chsigg <https://github.com/chsigg> has already fixed this in
75ea0b3
<75ea0b3>
recently (Jun 26, 2020) by adding an overload function. I tried compiling
with the latest code from master and didn't get the error anymore.
(It's too late to patch this into releases 2.2.0 and 2.3.0 now, so this
issue will be fixed in release 2.4.0.)
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#40688 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AANWFG73UXJSXZ36KTGUOW3R7CTBTANCNFSM4OE7JZYQ>
.
|
I will take this opportunity to update another part of the ARG LATEST_BAZEL=3.4.1 [...] && fgrep _TF_MAX_BAZEL configure.py | grep '=' | perl -ne '$lb="'${LATEST_BAZEL}'";$brv=$1 if (m%\=\s+.([\d\.]+).$+%); sub numit{@g=split(m%\.%,$_[0]);return(1000000*$g[0]+1000*$g[1]+$g[2]);}; if (&numit($brv) > &numit($lb)) { print "$lb" } else {print "$brv"};' > .bazelversion \ && bazel clean \ [...] |
Remember to run bazel clean after downgrading numpy. I downloaded numpy 1.18 and it worked. |
Numpy introduced a breaking API change in version 1.19.x, see [1]. There is a simple fix [2] available in the master branch. [1]: tensorflow/tensorflow#40688 [2]: tensorflow/tensorflow@75ea0b3
Numpy introduced a breaking API change in version 1.19.x, see [1]. There is a simple fix [2] available in the master branch. [1]: tensorflow/tensorflow#40688 [2]: tensorflow/tensorflow@75ea0b3 (cherry picked from commit 8f5bfd6)
Numpy introduced a breaking API change in version 1.19.x, see [1]. There is a simple fix [2] available in the master branch. [1]: tensorflow/tensorflow#40688 [2]: tensorflow/tensorflow@75ea0b3 (cherry picked from commit 8f5bfd6)
Numpy introduced a breaking API change in version 1.19.x, see [1]. There is a simple fix [2] available in the master branch. [1]: tensorflow/tensorflow#40688 [2]: tensorflow/tensorflow@75ea0b3 (cherry picked from commit 8f5bfd6)
…ericFunction. See tensorflow/tensorflow#40688, tensorflow/tensorflow#40654. PiperOrigin-RevId: 318452381 Change-Id: Icc5152f2b020ef19882a49e3c86ac80bbe048d64
Fixed by aafe25d |
Confirmed r2.2 source incompatible with Numpy version 1.18.5 and 1.19.0. Downgrade numpy < 1.18.5 will resolve the issues.
|
System information
Dockerfile
withFROM nvidia/cuda:10.1-cudnn7-devel-ubuntu18.04
Describe the problem
Build fails with
Provide the exact sequence of commands / steps that you executed before running into the problem
Reproducible with the following
Dockerfile
Built using
docker build --tag cto:test .
Note tested with CUDA 10.1, 10.0 and 10.2.
Also occurs with TF 1.15.3
Any other info / logs
I can provide the full build log if requested (91MB)
The text was updated successfully, but these errors were encountered: