News for hackers. Stuff that matters.

A comment on Deep Learning Benchmarks.

Re:Deep Learning Benchmarks by jimfleming on Friday March 04, 2016 @12:11PM (#11225135)
It's nice to see Leaf coming along so well. Part of me would love to be able to build models in rust.

For more benchmarks (including updated TensorFlow performance with cudnn v4) see https://github.com/soumith/convnet-benchmarks

[ Parent | on HN ]
Re:Deep Learning Benchmarks by mjhirn on Friday March 04, 2016 @01:57PM (#11225999)
I just patched the link on the landing page to the repo [1] and updated benchmarks for Leaf 0.2 + cuDNN 4 for Overfeat and VGG. But I couldn't get Torch and Tensorflow running with cuDNN 4, yet.

[1]: https://github.com/autumnai/deep-learning-benchmarks

[ Parent | on HN ]
Re:Deep Learning Benchmarks by vrv on Friday March 04, 2016 @02:19PM (#11226181)
Feel free to ping us (TensorFlow) on github issues to get installation issues resolved -- on cudnn r4 we're doing much better, and we're soon to check in a series of changes to get us roughly on par with Torch on cudnn r4.

We'd love to get a more up-to-date representation of the state of our own progress :)

[ Parent | on HN ]
Re:Deep Learning Benchmarks by mjhirn on Friday March 04, 2016 @04:19PM (#11226942)
Great, I will let you know. Looking forward to the benches.
[ Parent | on HN ]
Re:Deep Learning Benchmarks by narrator on Friday March 04, 2016 @01:15PM (#11225682)
Too bad Tensorflow Cuda only works with the top of the line NVIDIA cards that cost over 1k.
[ Parent | on HN ]
Re:Deep Learning Benchmarks by rough-sea on Friday March 04, 2016 @01:28PM (#11225771)
Not true, works on GTX 750ti out of the box which is currently around $130
[ Parent | on HN ]
Re:Deep Learning Benchmarks by jimfleming on Friday March 04, 2016 @01:20PM (#11225728)
Can you expand on that? As far as I'm aware that's not true (anymore). It runs quite fine on AWS which uses older NVIDIA cards and I know several people use it on older-gen GPU-enabled MBPs.

EDIT: clarification

[ Parent | on HN ]
Re:Deep Learning Benchmarks by narrator on Friday March 04, 2016 @02:20PM (#11226194)
https://www.tensorflow.org/versions/r0.7/get_started/os_setu...

I guess the cards listed there are not an all inclusive list?

[ Parent | on HN ]
Re:Deep Learning Benchmarks by barneso on Friday March 04, 2016 @03:58PM (#11226809)
In my experience, the architecture supports cards with shader model >= 3.0. Occasionally a commit will break the support (eg https://bitbucket.org/eigen/eigen/commits/a19653b8035d8ace53... was required earlier this year) but this is a function of the speed of development and usually straightforward to fix.
[ Parent | on HN ]
Re:Deep Learning Benchmarks by vrv on Friday March 04, 2016 @02:26PM (#11226237)
Yeah, those are just cards we know that work: we don't have all possible GPUs to test our 20+ changes a day on, so we can't formally guarantee it will work on older cards, but in general we try our best to keep it all working for older and even newer cards.
[ Parent | on HN ]
Re:Deep Learning Benchmarks by nightski on Friday March 04, 2016 @03:05PM (#11226516)
Maybe a "reported to work but not guaranteed" section is in order.
[ Parent | on HN ]
Re:Deep Learning Benchmarks by vrv on Friday March 04, 2016 @03:59PM (#11226815)
Keeping that up-to-date and useful is as hard as keeping benchmarks in Deep Learning up to date ;)
[ Parent | on HN ]
Re:Deep Learning Benchmarks by jimfleming on Friday March 04, 2016 @02:27PM (#11226245)
From that page:

> Supported cards include but are not limited to[...]

[ Parent | on HN ]
Re:Deep Learning Benchmarks by reality_hacker on Friday March 04, 2016 @01:47PM (#11225926)
Just curious, does this benchmark include models quality somewhere?
[ Parent | on HN ]
Re:Deep Learning Benchmarks by mjhirn on Friday March 04, 2016 @01:58PM (#11226009)
You mean, comparing the performance after training a model?
[ Parent | on HN ]
Re:Deep Learning Benchmarks by reality_hacker on Friday March 04, 2016 @06:16PM (#11227488)
Performance in terms of model accuracy. How accurate will be those models.
[ Parent | on HN ]
Re:Deep Learning Benchmarks by hobofan on Saturday March 05, 2016 @09:19AM (#11229556)
No accuracy is not included, since that is something that should be constant for each model across frameworks.

These benchmarks aim to highlight the performance differences in terms of speed/memory usage across frameworks and machine configurations.

There is also the practical hurdle that training imagenet models to maximum accuracy takes 1 week+.

[ Parent | on HN ]
Re:Deep Learning Benchmarks by reality_hacker on Saturday March 05, 2016 @05:18PM (#11231345)
So, are you saying that output of all frameworks are exactly and always the same?

Sorry, if I am asking stupid questions.

[ Parent | on HN ]
Re:Deep Learning Benchmarks by hobofan on Monday March 07, 2016 @07:14AM (#11238438)
Not exactly, no. But not even the output of the same framework will always be the same since you are usually randomly initalizing the weights in a network and randomly picking the samples used in SGD[1] (the seed for the RNG could of course be a fixed one to mitigate this somewhat).

But in the end, if you are using the same model, the same solver and the same RNG, yes the output of all frameworks should be the same. In practice this also mostly holds true, since the stochastic processes involved are geared towards finding a good local minimum, which is the same given a model and a dataset.

[1]: https://en.wikipedia.org/wiki/Stochastic_gradient_descent

[ Parent | on HN ]
Re:Deep Learning Benchmarks by reality_hacker on Monday March 07, 2016 @01:00PM (#11240323)
So, there is a strong chance that you are comparing very different things, and frameworks which perform not good in terms of speed can be far superior in terms of models quality.
[ Parent | on HN ]