Exciting news: we have completed our hero run, pre-training Marin 535B-A23B, a mixture of experts model affectionally known as Mochi. Mochi is Marin’s largest model to date, and we estimate that it is the largest live-streamed pre-training run ever

#JCcrashes A car flipped on 6th west of Grove St yesterday afternoon. These types of crashes are why we need safe street design upgrades on 6th. We welcome the 3 planned improvements, and we urge @jerseycitynj.bsky.social to implement these builds asap. cc @eleanalittle.bsky.social 📷 @rbw.sh

We’re excited to share more about our progress post-training our models. Our latest milestone is the release of Marin 67B-A2B “Snowball” v0.1, a mixture of experts model trained on 10 trillion tokens. In this blog post, Ben Feuer explains what we learned from the process:

Post-Training in the Open

We’re excited to share more about our progress post-training our mixture of experts models. Our latest milestone is the release of Marin 67B-A2B “Snowball” v0.1, a mixture of experts model trained...

openathena.ai

I recently wrote up this piece (benfthomas.substack.com/p/we-need-30...), supported by an interactive dashboard with ctbk.dev data. There are neighborhoods where it's an arms race every morning to get an e-bike. Play with the dashboard here! benfthomas.com/projects/cit... Thank you to @rbw.sh :)

ctbk.dev - Citi Bike Dashboard

NYC Citi Bike trip data: rides per month, station maps, and more.

ctbk.dev

A year ago, Marin had one FTE. Recently, the ten-person team launched a 535 billion parameter MoE model. In the Marin 535B-A23B launch note, David Hall @dlwh.bsky.social notes it's going "almost boringly well." Thank you to our colleagues, partners, and advisors for making this possible. bit.ly/535b

Marin 535B-A23B launch note

A look back at Marin's first year at Open Athena and forward at the 535B-A23B hero run, the largest model the team has ever trained.

openathena.ai

Excited to share MarinDNA, a 1B gLM that rivals Evo 2 40B on variant effect prediction while being 2,330x faster. With @eczech0.bsky.social, we built around a standard Transformer so we could reuse LLM infra and methods while focusing on data curation and scaling. openathena.ai/blog/marin-d... 🧵

Here's the tale of how @jder.bsky.social and I scaled Samudra, a neural ocean emulator capable of predicting 8 years of the ocean on a single GPU, to operate at a full 1/4° resolution (16x the size in bytes). It was quite a humbling process.

Open Athena@openathena.ai · 3mo ago

Simulating ocean climate takes a supercomputer 4,600+ CPU cores to produce 12 simulated years per day (SYPD). Samudra 2 produces 4,800 SYPD on 1 GPU at the same resolution. In a new blog, @al.merose.com reports on Samudra, a neural ocean emulator built in collaboration with NYU & MIT: bit.ly/oa-ss

@runsascoded.com updated his visualization, showing the efficiency of different modes of transit between NYC-NJ. Lincoln tunnel bus lane is the winner - and it only operates for 3 hrs a day! Investment in private vehicle infrastructure, is not a good use of state or city resources. hbt.hccs.dev

Drivers injure 150 people in NYC each day (vs. 3 by cyclists, including ebikes) Drivers kill 2 pedestrians in NYC each week (vs. 2 per year by bikes and ebikes) www.nyc.gov/html/dot/dow...

1 bus lane (Lincoln Tunnel) moves 5x as many people per hour as 5 car lanes combined (Lincoln + Holland Tunnels) github.com/hudcostreets...

Slides from a talk I gave at @njbwc.org summit: bw25.hccs.dev

Data @ Hudson County Complete Streets

New Jersey Bike Walks Summit 2025 slides

bw25.hccs.dev

🚲Bike JC🗽@bikejc.org · 2y ago

Bike JC and @HudCoStreets trustee Ryan Williams @RunsAsCoded explains how we’re using a wide variety of data to support numerous advocacy campaigns for better biking, walking, mass transit and safe streets, at the annual @NJBikeWalk Summit.

Citi Bike ridership data, Oct '24: ctbk.dev NYC: • 5,131,687 rides (all-time high) • 12mo avg: 3,638,786 rides/mo (all-time high) JC+HOB: • 118,005 rides (all-time high) • 12mo avg: 86,942 (all-time high) @citibike.bsky.social @hudcostreets.org

Car crashes kill twice as many people as homicides, in NJ. In 2022, crashes killed 2.4x as many people, the largest disparity on record. crashes.hudcostreets.org#vs-homicides @hudcostreets.org

PATH ridership data, Sept '24: path.hudcostreets.org • Avg weekday: 206,542 rides • Avg weekend: 101,194 rides Sept '24 vs. Sept '19 (pre-COVID): • Weekdays: 69.2% • Weekends: 99.5% Weekend ridership is at pre-COVID levels, despite degraded service. Weekdays continue to set post-COVID highs.