PlaidML

Today, we hit 2,500 stars on our Github repository! To thank our users for getting us here, we're doing a PlaidML swag giveaway. Enter our giveaway by telling us how you use PlaidML (be sure to include #PlaidMLUserStories in your entry). We'll select a winner on 10/4.

@pillarofautuman 2/2 If you'd like us to look at your specific use case, please file an issue on our Github repository with some code samples and your hardware specs, and we'll take a look! github.com/plaidml/plaidm…

@pillarofautuman 1/2 Hi @pillarofautuman thanks for using PlaidML! Our team loves the cats/dogs dataset (we're even rolling out a demo featuring resnet50 training on it). Regarding performance, PlaidML works best with larger networks/batch sizes since CPU->GPU data transfers are bottlenecked.

@plaidml great work team... However, i am in deep trouble here...i am trying to train the classic 🐈 and 🐕 dataset(10000 total images) and training is taking a lot of time..

ICYMI, the PlaidML team hosted an introductory webinar a couple of weeks ago! If you'd like to learn more about how PlaidML can accelerate your deep learning workflows, catch the recording of our webinar here: intel.ly/2GAsEnl #plaidml #TensorCompiler #DeepLearning

Great news - Tim and Denise from our team will be presenting at our very own webinar on 7/17 at 9AM PDT! Follow the link and sign up for "The PlaidML Tensor Compiler" to stay in the loop. software.seek.intel.com/AIWebinarSeries

@iiacobac We're so happy you found us!

Hey everyone - #plaidml v0.6.0 has been released to PyPI. pypi.org/project/plaidm… We've spent a lot of time working on our polyhedral IR, Stripe. We're so excited for you to try out the optimizations we've made in this release!

Hey @srcgraph github.com/plaidml/plaidml doesn't work on sourcegraph.com, can you help a plaidypus out?

@sergei3000 Yes! More soon

@plaidml Are there any plans to implement multi node distributed training/inference?

Good news! #plaidml v0.5.0 has been released to PyPI. pypi.org/project/plaidm… Release notes and a proper github release inbound. This version supports the latest #keras, has a ton of bug fixes, and contains an early release of our new polyhedral IR: Stripe. More soon!

Read about the polyhederal IR that is the basis for the next version of PlaidML - Stripe IR: arxiv.org/abs/1903.06498

@SShanabrook @denfromufa Cool! Stripe seems like it would be a pretty natural fit. I'll DM you w/ contact info for the PlaidML people at Intel.

@plaidml @denfromufa Sweet! We are working on a high level array IR in Python that could potentially use this as a backend. Example notebook: github.com/Quansight-Labs… current work: github.com/Quansight-Labs…

@SShanabrook @denfromufa Only sort of, the stripe docs aren't in there yet. It's relevant to the current version though.

@plaidml @denfromufa Are these docs still generally relevant? …rtexai-plaidml.readthedocs-hosted.com/en/latest/api/…

@CDerinbogaz Good question! Our new Stripe based backends will make supporting multi-gpu configs quite a bit easier. Not sure exactly where that is on our roadmap – we'll be creating and publishing a roadmap in the next month or so.

@plaidml Great news! One question; when will you support multi gpu setups? Thanks!

@SShanabrook @denfromufa Yes it's designed to have very minimal dependencies and you could use it anywhere the computational model makes sense. Adding a backend right now is fairly easily but a bit ad-hoc. We'll have a GPU / OpenCL example and more docs out soon.

@plaidml @denfromufa Stripe looks very interesting! How hard is it to add and backend? Would it make sense to use it outside plaidml?

Do you love tensor compilers like #plaidml and @TVMProject as much as I do? If so, Intel is hiring people to contribute to #plaidml and develop groundbreaking new HW/SW co-design tools: jobs.intel.com/ShowJob/Id/175…

Our primary contributor Vertex.AI is now part of Intel’s AI group. Exciting developments to follow, stay tuned for more updates.

Glad we can help!

#PlaidML が #Metal サポートしてる! これで #Mac の #Radeon が活用できる! vertex.ai/blog/accelerat… #AMD #GPU #MachineLearning #AI #機械学習

#CVPR18 attendees: Join us tomorrow 6/18 at 9am in ballroom BD at the Autonomous Driving Workshop to hear @DeepScale_ co-founder and @berkeley_ai Prof. Kurt Keutzer cover how automotive challenges computer vision research.