I gave an overview of some of the challenges and advances in our recent Gemini model training runs.
@jeffdean
-
ML Training CO2 Emissions: Debunking Highly Cited Misinformation
By
–
I discussed misinformation in the area of CO2 emissions from ML training, where one highly cited paper has a misleading and incorrect estimate that is off by 88X from the actual CO2 emissions of a one-time neural architecture search done by So et al., and is off by 118,000X from
-
Reinforcement Learning Advances TPU Chip Design Across Generations
By
–
Among many other topics, here are a few things from the talk. I gave an update on our use of reinforcement learning for ASIC chip placement and routing, showing how the adoption of this approach has been increasing from TPU generation to generation (due to further work on the
-
MLSys Talk on ML for Systems and Systems for ML
By
–
I had a great time giving a talk at MLSys this morning on "Advances in ML for Systems and Systems for ML". So many enthusiastic questioners! https://
mlsys.org -
Google Trillium TPU: 4.7X Performance Boost, Doubled Memory
By
–
Very excited about our 6th generation TPU: "Trillium TPUs achieve an impressive 4.7X increase in peak compute performance per chip compared to TPU v5e. We doubled the High Bandwidth Memory (HBM) capacity and bandwidth, and also doubled the Interchip Interconnect (ICI) bandwidth
-
Going Deeper with Inception Model Architecture
By
–
We must go deeper. Hopefully using an Inception model!
-
Gemini Responds to OpenAI Event with Professional Recognition
By
–
Our Gemini model enjoyed watching yesterday's OpenAI event. Nice work, @miramurati, @barret_zoph and @markchen90 (and everyone else behind the scenes)! https://t.co/Dp2jUvSPhf
— Jeff Dean (@JeffDean) 14 mai 2024Our Gemini model enjoyed watching yesterday's OpenAI event. Nice work, @miramurati
, @barret_zoph and @markchen90 (and everyone else behind the scenes)! -
Gemini 1.5 Flash: Multimodal AI with 1M Token Context
By
–
Gemini 1.5 Flash has really great qualities. A really good capable, natively multimodal, 1M token context window (with signup available to get access to a 2M token variant), and super lower latencies and fast response generation. https://t.co/2pfwgJCRzX
— Jeff Dean (@JeffDean) 14 mai 2024Gemini 1.5 Flash has really great qualities. A really good capable, natively multimodal, 1M token context window (with signup available to get access to a 2M token variant), and super lower latencies and fast response generation.
-
Gemini and Astra Join Google IO Keynote Watch Party
By
–
Gemini and the Astra system joined the watch party for the #GoogleIO keynote! The model seemed to understand a lot about what was being presented, and it might have caught a reference or two to Gemini models as well😀 https://t.co/P9p4RlOoyv
— Jeff Dean (@JeffDean) 14 mai 2024Gemini and the Astra system joined the watch party for the #GoogleIO keynote! The model seemed to understand a lot about what was being presented, and it might have caught a reference or two to Gemini models as well
-
Google IO 2026: Major AI and Technology Announcements Expected
By
–
Lots of exciting stuff coming at #GoogleIO! https://t.co/zpA5erfAwo
— Jeff Dean (@JeffDean) 14 mai 2024Lots of exciting stuff coming at #GoogleIO!