Samaneh Saadat
@smn_sdt
Followers
598
Following
3K
Media
65
Statuses
480
ML SWE @Google | ML Frameworks | Opinions my own
Seattle
Joined September 2015
Check out the "Learning JAX" video series if you're interested in learning JAX!
11
58
694
Divya, Abheesht, and Samaneh — ML engineers at Google — gave a workshop at #GHC25 about how Google's Agent Development Kit makes it easy to build multi-agent applications. Interested in learning more ➡️ https://t.co/YLQxvqedDR Explore our open roles and opportunities ➡️
0
3
13
The narrative around LLMs is that they got better purely by scaling up pretraining *compute*. In reality, they got better by scaling up pretraining *data*, while compute is only a means to the end of cramming more data into the model. Data is the fundamental bottleneck. You can't
98
188
2K
Francois and Matt are two of the most brilliant people I've worked with. They're excellent at explaining complex ideas in a simple and intuitive way. Don't miss their book.
The 3rd edition of my book Deep Learning with Python is being printed right now, and will be in bookstores within 2 weeks. You can order it now from Amazon or from Manning. This time, we're also releasing the whole thing as a 100% free website. I don't care if it reduces book
0
2
5
My favorite is
colab.research.google.com
it has come to my attention that this is not universal knowledge you can just type https://t.co/prfdqNqr5m or https://t.co/HMVIGA1uSh into your browser and it will immediately open a new google doc or sheet
0
0
4
Try out VaultGemma on KerasHub! https://t.co/BylPNu711o
Introducing VaultGemma 🧠Gemma pre-trained with differential privacy (largest open model trained from scratch like this) 🔒Strong, mathematically-backed privacy guarantees 🤏Just 1B parameters 📈Novel research on scaling laws
1
4
9
Introducing EmbeddingGemma🎉 🔥With only 308M params, this is the top open model under 500M 🌏Trained on 100+ languages 🪆Flexible embeddings (768 to 128 dims) with Matryoshka 🤗Works with your favorite open tools 🤏Runs with as little as 200MB https://t.co/AXPqV4aXr1
28
156
1K
Yay 🎉
We’re heading North – the Pacific Northwest to be exact! Today, we’re returning to Washington State as we lay the groundwork to launch our autonomous ride-hail service in the Seattle metropolitan area. Learn more: https://t.co/3J8gKvbW7Y
0
0
3
You can use Orbax for checkpointing when training your Keras model with the JAX backend. Orbax checkpointing is particularly useful when doing multi-host training using Keras distribution API. We have a new guide showing how to do that.
1
4
11
What I don't like about it is the weather! 😁 Too hot during the day and too windy and cold at night!
2
0
1
Visiting bay area this week and what I love about it is that I get to chat with people who work on very interesting problems.
1
0
6
Incredible to see that the energy used for a median Gemini AI text prompt has dropped 33x in only 12 months. This is a huge achievement – and one that would not have been possible without the work of many Googlers and focused efforts to deliver greater efficiencies across the
AI efficiency is important. Today, Google is sharing a technical paper detailing our comprehensive methodology for measuring the environmental impact of Gemini inference. We estimate that the median Gemini Apps text prompt uses 0.24 watt-hours of energy (equivalent to watching an
3
4
39
