We've been waiting for this moment since Tensorflow was open-sourced.
I have been struggling with 5 different versions of Bazel and lots of code patches in Tensorflow to make it work on my humble machines (because Cuda capability 3.0 is not official supported, I will write about the final distilled steps in a future post), but I was terribly disappointed when I discovered that distributed computation was not supported in the initial release.
Fortunatelly, some Spark presentations and github repositories pointed to this feature to be supported soon.
And finally, today, an initial version of the distributed runtime, based on gRPC, has just been pushed into the official TensorFlow repository.
It's very promising...
Actualization (13th April'16): Official distributed version has just been released (release candidate).
I have been struggling with 5 different versions of Bazel and lots of code patches in Tensorflow to make it work on my humble machines (because Cuda capability 3.0 is not official supported, I will write about the final distilled steps in a future post), but I was terribly disappointed when I discovered that distributed computation was not supported in the initial release.
Fortunatelly, some Spark presentations and github repositories pointed to this feature to be supported soon.
And finally, today, an initial version of the distributed runtime, based on gRPC, has just been pushed into the official TensorFlow repository.
It's very promising...
Actualization (13th April'16): Official distributed version has just been released (release candidate).
- 10:45 AM
- 0 Comments
