Dear Sean,
I do agree with you to a certain extent, makes sense. Perhaps I am wrong in
asking for native integrations and not depending on over engineered
external solutions which have their own performance issues, and bottlenecks
in live production environment. But asking and stating ones opinion
Hi Bitfox,
yes distributed training using Pytorch and Tensorflow is really superb and
great and you are spot on. There is actually absolutely no need for
solutions like Ray/ Petastorm etc...
But in case I want to pre process data in SPARK and push the results to
these deep learning libraries, the
On the contrary, distributed deep learning is not data parallel. It's
dominated by the need to share parameters across workers.
Gourav, I don't understand what you're looking for. Have you looked at
Petastorm and Horovod? they _use Spark_, not another platform like Ray. Why
recreate this which has
I have been using tensorflow for a long time, it's not hard to implement a
distributed training job at all, either by model parallelization or data
parallelization. I don't think there is much need to develop spark to
support tensorflow jobs. Just my thoughts...
On Thu, Feb 24, 2022 at 4:36 PM Go
Hi,
I do not think that there is any reason for using over engineered platforms
like Petastorm and Ray, except for certain use cases.
What Ray is doing, except for certain use cases, could have been easily
done by SPARK, I think, had the open source community got that steer. But
maybe I am wrong
Currently we are trying AnalyticsZoo and Ray
Von meinem iPhone gesendet
> Am 23.02.2022 um 04:53 schrieb Bitfox :
>
>
> tensorflow itself can implement the distributed computing via a parameter
> server. Why did you want spark here?
>
> regards.
>
>> On Wed, Feb 23, 2022 at 11:27 AM Vijaya
Petastorm does that https://github.com/uber/petastorm in the sense that
it feeds Spark DFs to those frameworks in distributed training.
I'm not sure what you mean by native integration that is different? these
tools do just what you are talking about and have for a while.
On Wed, Feb 23, 2022 at 7
Hi,
I am sure those who have actually built a data processing pipeline whose
contents have to be then delivered to tensorflow or pytorch (not for POC,
or writing a blog to get clicks, or resolving symptomatic bugs, but in real
life end-to-end application), will perhaps understand some of the issu
Spark does do distributed ML, but not Tensorflow. Barrier execution mode is
an element that things like Horovod uses. Not sure what you are getting at?
Ray is not Spark.
As I say -- Horovod does this already. The upside over TF distributed is
that Spark sets up and manages the daemon processes rath
Hi,
the SPARK community should have been able to build distributed ML
capabilities, and as far as I remember that was the idea initially behind
SPARK 3.x roadmap (barrier execution mode,
https://issues.apache.org/jira/browse/SPARK-24579).
Ray, another Berkeley Labs output like SPARK, is trying to
tensorflow itself can implement the distributed computing via a
parameter server. Why did you want spark here?
regards.
On Wed, Feb 23, 2022 at 11:27 AM Vijayant Kumar
wrote:
> Thanks Sean for your response. !!
>
>
>
> Want to add some more background here.
>
>
>
> I am using Spark3.0+ version
Dependencies? Sure like any python library. What are you asking about
there?
I don't know of a modern alternative on Spark.
Did you read the docs or search? Plenty of examples
On Tue, Feb 22, 2022, 9:27 PM Vijayant Kumar
wrote:
> Thanks Sean for your response. !!
>
>
>
> Want to add some more
Thanks Sean for your response. !!
Want to add some more background here.
I am using Spark3.0+ version with Tensorflow 2.0+.
My use case is not for the image data but for the Time-series data where I am
using LSTM and transformers to forecast.
I evaluated SparkFlow and spark_tensorflow_distribut
13 matches
Mail list logo