Today we’re proud to be releasing t0-alpha, our first public forecasting model.
t0-alpha is a 102M-parameter transformer built for time-series forecasting, released under Apache 2.0.
https://t.co/WNef9guLpQ
Hey @trq212@bcherny@AnthropicAI - we have a weird thing where we're being rate limited on Claude Tag / CC even though we've activated extra usage on everything - and the limits we see in the console are not reached. Pretty annoying since we don't know how to fix.
t0-alpha is the best of our models so far - and just the beginning of what’s coming next.
Try it on your own data, download it on Hugging Face, forecast through our API, or use it in a few clicks on Retrocast. t0 is now. And it's only the beginning!
Today we’re proud to be releasing t0-alpha, our first public forecasting model.
t0-alpha is a 102M-parameter transformer built for time-series forecasting, released under Apache 2.0.
https://t.co/WNef9guLpQ
Because real forecasting is not just looking at the past and hoping for the best.
It needs to include known context about the future.
This release comes after months of working closely with customers and seeing where forecasting models fail in real-world use cases.
These changes are all about to land on the main branch of `datasets` and `huggingface_hub`, stay tuned !
Also check out @andimarafioti and the team at @forecastingco with @geoffnegiar and co who already have access to all this.
They might share cool stuff soon ;)
@typesfast You should check The Forecasting Company https://t.co/PWMpUzhUgT
Ex-Amazon and Google Brain founder, building a strong forecasting model for retailers and e-comm.
Can integrate via API also, lmk if you want an intro
@typesfast@ycombinator Check out https://t.co/oyvPzEnT15 for the history of Amazon's forecasting methods.
Most e-comm cos should plug in a backtested foundation model (+relevant external data sources) that will continuously improve. That's where we come in.
@typesfast Hi @typesfast!
We're a @ycombinator co (S24) building foundation models for (demand) forecasting.
Best tech at scale is transformer based, pretrained.
Ideal capabilities: probabilistic forecasts, modeling joint distributions over time, SKUs, regions.
we've seen nothing yet! hosted a 9-13 yo vibe-coding event w. @robertkeus this w-e (h/t @antonosika@LovableBuild)
takeaway? AI is unleashing a generation of wildly creative builders beyond anything I'd have imagined
and they grow up *knowing* they can build anything!
I'm at NVIDIA GTC 2025, where you can book me for this @e180labs braindate: Forecasting in the age of AI: leveraging improvements for accurate time series forecasting https://t.co/bDqOAEtwio #braindate#NvidiaGTC2025
Founders who were PhD or post-doc in my lab at Berkeley, **largely funded by NSF / DoD grants**, start-up, market cap (collected by OpenAI Deep Research)
We just released our new website!
Our goal for now is to provide the easiest, fastest benchmarking tools for forecasting models. Upload data, run forecasts at any time-point, compute metrics, zoom-in, zoom-out... and compare models!
Check out our website, and our demo.
🎉A big congratulations to @forecastingco@geoffnegiar and @jodles89 on this incredible milestone! From incorporation to 📍Times Square in just a few months — Proud to support your journey. #NYXPortfolio 💪