Open-weight models are AI systems with publicly released trained parameters, enabling developers to run, fine-tune, and deploy them independently, rather than solely through a hosted API. Unlike closed-weight models from companies like OpenAI or Anthropic, which are offered as managed services, open-weight models allow organizations to have total control over deployment and usage. These models are increasingly improving in performance, making them credible alternatives for production workloads, especially in terms of customization and data privacy.
Fireworks AI is developing a platform aimed at large-scale service and customization of open-weight models. Features of the platform include optimized inference infrastructure, support for various hardware such as NVIDIA and AMD, and reinforcement fine-tuning capabilities.
Benny Chen, a Co-Founder of Fireworks AI, joins Gregor Vand in this episode to share his journey from Meta’s ML infrastructure teams to co-founding Fireworks AI. They discuss the growing competitiveness of open-weight models, the benefits of custom kernels and speculative decoding for performance enhancement, reinforcement fine-tuning, and more.
Gregor Vand is a security-focused technologist with a background as a CTO in the fields of cybersecurity, cyber insurance, and software engineering, based in Singapore. More information about him can be found via his profile at vand.hk or on LinkedIn.
Please click here to see the transcript of this episode.
Sponsors
Turbopuffer is a serverless vector and full-text search engine built on object storage, offering search capabilities at up to 95% lower cost than traditional databases. For more information, visit turbopuffer.com/sed for a free first month.
Guardsquare provides advanced mobile app security using multi-layered code hardening and runtime application self-protection. Learn more about securing Android and iOS apps at www.Guardsquare.com.
Unblocked offers a context layer for coding agents, synthesizing organizational information to improve agent outputs. Try it for free at getunblocked.com/sedaily.