Jiri Simsa

Working on data processing and analysis infrastructure for ML @ Google.

[Ana Klimovic @anaklimovic]

Today at @MLSysConf, @MichaelKuchnik will present Plumber, our tool for diagnosing and removing performance bottlenecks in ML input data pipelines.

Joint work with @jsimsa, @GeorgeAmvrosia2 and Virginia Smith.

Paper: https://t.co/CkXKLcWiwo

[Jiri Simsa @jsimsa]

If you are interested in advancing infrastructure that provides large scale data analysis and processing for ML workloads across Google, my team is hiring:
https://t.co/G6cXp7qRJN

[Ana Klimovic @anaklimovic]

Our VLDB’21 talk about https://t.co/uGweaPNtSE, a ML data processing framework, is now online: https://t.co/t79rzH8c4M

More details in our paper: https://t.co/1M5hJRmnEh

It has been great to collaborate on this work with @mrry @jsimsa & Ihor Indyk!

[Google @Google]

Five years ago, we open sourced @TensorFlow, our machine learning framework that's now the most popular machine learning library in the world. 🌎 To celebrate, we’re sharing few interactive demos and tutorials you can try, no experience required → https://t.co/JxKOSC5R69

[Jiri Simsa @jsimsa]

Awesome to see the success of TensorFlow and JAX, both using https://t.co/u0l4xlIzUJ to ingest data fast enough to train to convergence in under 30 seconds!

[Jeff Dean @JeffDean]

Very excited to see the MLPerf 0.7 results released today, where Google TPUs set records in six of the eight benchmarks!

We need bigger benchmarks, because we can now train the ResNet-50, BERT, Transformer, & SSD benchmarks each in under 30 seconds.

https://t.co/J6vbZ8srs1

[James Bradbury @jekbradbury]

In 2016, when I was working on machine translation, it took me more than a week on a multi-GPU machine to train a competitive system on WMT English-German.

Today, JAX on a TPU v3 supercomputer can train a better model on the same data in 16 seconds! https://t.co/fRK6vwZHDP

[👩‍💻 Paige Bailey @DynamicWebPaige]

👉 https://t.co/Fv77gxVyzu supports *any* machine learning framework (JAX, @TensorFlow, PyTorch, more!), and is a great way to speed up your data input pipelines.

Be sure to try out our new features for https://t.co/Fv77gxVyzu, available in TF 2.3: https://t.co/reGFHiLImI

[TensorFlow @TensorFlow]

🔍Inside TensorFlow: https://t.co/EPERzWIIN0 + tf.distribute

In this presentation, Jiri Simsa showcases best practices. You’ll learn about the input pipeline, parallel extraction, distributed training, and more.

Watch here → https://t.co/3CLRx3yAE6

[Josh Gordon @random_forests]

If your dataset is small, use an in-memory cache:
ds = ds.cache()

If large, create an on-disk cache:
ds = ds.cache("my_file")

Afterwards, you can call ds.batch() and ds.shuffle() as always.

Complete example: https://t.co/gz6aBT3SLQ

[Alluxio @Alluxio]

Speaker spotlight - @jsimsa, tech lead of the https://t.co/q8kr2WIIgg project & software engineer at Google, to present on https://t.co/q8kr2WIIgg the recommended API for creating #TensorFlow input pipelines @ #DataOrchestrationSummit. RSVP: https://t.co/70IJHb9ApG
#opensource

[Jiri Simsa @jsimsa]

Presented https://t.co/u0l4xm0bjj and tf.distribute at #GoogleMLSummit in Tokyo! Stay tuned for a recording.

[Takuo Suzuki @taquo]

Google Developers ML Summit , @JeffDean の基調講演！#GoogleMLSummit

[Jiri Simsa @jsimsa]

Thank you for the kind words!

[Neil Tenenholtz @ntenenz]

Loving the "Inside Tensorflow" series. The latest release on the TF data API highlights just how much effort the @TensorFlow team has invested in making highly performant pipelines accessible to the end user. Major kudos. 👏👏👏 @mrry @jsimsa et al

https://t.co/XPGwMhluLw

[Jeff Dean @JeffDean]

Not only are TPUs fast for doing machine learning, but they are also more energy efficient than alternative platforms, so you can feel great as you train that language model on scientific articles about climate change. https://t.co/eDuEDRYJVz

[Brennan Saeta @bsaeta]

Today in #CloudTPU announcements: (1) @TensorFlow 1.8 now available with a slew of perf improvements (2.7k to 3.2k images/sec on ResNet-50, aka 12.5 hours is now 9 hours to fully train), and (2) we have opened up a new zone (us-central1-b) for HA & load balancing.

[Frank Chen @frankchn]

Our latest DAWNBench results are live: 8h52m for @TensorFlow to train ResNet-50 on ImageNet on a single @GCPcloud TPU (<$60), and just 30 minutes on half a TPU pod!

https://t.co/cgJPXMVjhQ

[Jeff Dean @JeffDean]

We just posted new DAWNBench results for ImageNet classification training time and cost using Google Cloud TPUs+AmoebaNet (architecture learned via evolutionary search). You can train a model to 93% top-5 accuracy in <7.5 hours for <$50.

Results: https://t.co/CS53YcHtKC

[Brennan Saeta @bsaeta]

Cloud TPUs (now in !!open!! beta) are a leap forward in price & performance for Machine Learning. (See https://t.co/mo0XorlWk2 for end-to-end benchmarks.) Spin one up at https://t.co/N9d5o5ZEeV today!

[Derek Murray @mrry]

If you want to find out more about https://t.co/Jc0yFJ2dFt performance after my talk at #TFDevSummit, check out this awesome guide by @jsimsa and @bsaeta! https://t.co/hYwki5a6cL

[Derek Murray @mrry]

I'll be speaking about https://t.co/Jc0yFIKCgT at 10am PDT. Hope you can tune in to the livestream!

https://t.co/hpVemiOlkh https://t.co/CT3R2bPzmH