Suggestions
Stephen Purpura
Business Owner & Consultant with expertise in data management and AI solutions
Stephen Purpura is a business owner at Autonomous Data Management LLC, a company he founded in June 2021, located in Wilmington, Delaware. His firm specializes in data management and generative AI solutions, utilizing technologies like ChatGPT and Llama. Purpura focuses on optimizing GPU and accelerated computing solutions to enhance cost efficiency, significantly reducing retraining and inference costs for machine learning models.1
Education and Background
Purpura holds a Master's degree in Public Policy from Harvard University, which he completed between 2005 and 2007. He also pursued a PhD in Information Science at Cornell University, where he completed all requirements except for the dissertation, before taking a leave to start his first company, Context Relevant (now Versive) in 2014. He completed his undergraduate studies in Government at the University of Washington.1
Professional Experience
In addition to his role at Autonomous Data Management LLC, Purpura has been a consultant since June 2019. His expertise encompasses behavioral learning, autonomous systems, and machine learning, reflecting a strong background in data science and analytics.12
Honors and Awards
Stephen Purpura has received several accolades throughout his career, including:
- Nominated for the Best Paper Award at the Computer Human Interaction (CHI) Conference in May 2011.
- Recognized as Amazon Web Services Commercial Employee of the Quarter in Q1 2019.
- Finalist for the EY Entrepreneur Of The Year® 2015 in the Pacific Northwest.
- Recipient of the Cornell Department of Computer Science Outstanding Teaching Award in May 2009.1
Highlights
Want to increase the speed of your local LLM by 30% to 60%?
I've been involved in a few discussions this week about why changing the size of the vocabulary in a LLM to align it with a 8 byte, 64 byte, or 256 byte size boundary optimizes the use of NVidia tensor cores. Attached is the paper that helps explain it for ops people.
Apple's MLX platform now has simple & quick support for speech recognition through Whisper.
The MLX Whisper package now works on Macs with Distil-Whisper (and other Transformer compatible Whisper models).
From terminal: pip install -U mlx-whisper
This will install the distil-large-v3 Whisper model that uses MLX and hardware acceleration on a Mac M1, M2, M3, and M4 chip. 10 minutes of audio transcribes in less than 10 seconds. It is also possible to do real-time streaming but not yet with the simple examples available.
