I came to the US in 2011 to attend @IllinoisCS. Just found out on the 13th anniversary of first coming here that my EB-1 petition was approved! Excited to continue building Refuel here. The United States is a place where great things are possible - I grew up believing this,

Great customer support starts with caring about your users' problem more than they do

the only viable option here would be to start over, right?

Saturdays

Speculative decoding is one of the best tool in the vLLM's suite of inference optimization tool box, accelerating the inference without accuracy loss. Checkout our blog post for more details about the state of spec decode in vLLM today! 🧵 blog.vllm.ai/2024/10/17/spe…

True conviction draws in missionaries. Hard to fake. Fake conviction draws in mercenaries. Easy to spot.

The Moore's Law Update NOTE: this is a semi-log graph, so a straight line is an exponential; each y-axis tick is 100x. This graph covers a 1,000,000,000,000,000,000,000x improvement in computation/$. Pause to let that sink in. Humanity’s capacity to compute has compounded for

Does training LLMs with reasoning traces (same recipe used in DeepSeek-R1-Distill model family) enhance performance of data labeling, categorization and extraction tasks? Our experiments suggest the improvement is ~5%, but at a substantial increase (>5x) in cost:

We have some big news to share today - Refuel is joining Together AI to help accelerate the future of open source and enterprise AI! together.ai/blog/together-…

Congratulations to our portfolio companies, Together AI and Refuel, on uniting their strengths to power the next generation of AI infrastructure! Together AI’s AI Acceleration Cloud enables developers and enterprises to train and deploy generative AI models with speed,

scarcity ➡️ abundance is a beautiful thing 1850s: Food 1900s: Manufacturing 1950s: Energy 2020s: Intelligence

Excited to launch Agent Bricks, a new way to build auto-optimized agents on your tasks. Agent Bricks uniquely takes a *declarative* approach to agent development: you tell us what you want, and we auto-generate evals and optimize the agent. databricks.com/blog/introduci…

Technical debt is like financial debt in many ways. Borrow from the future to accelerate buildout today. Grow the pie and pay back a much smaller fraction of it in the future

Windows, the web, facebook, iphone, cloud, deep learning, AVs, chatgpt - all in 30 years. Next 30 gonna be 🚀

your data is safe. we don't train on it. because it would make the model worse