Uploaded September 2026 | Updated November 2026, 3 days ago
TorchSpec is a torch-native speculative decoding training framework built on Ray, with a disaggregated design that lets each side of drafter training scale independently. Draft models trained with TorchSpec for large models such as Kimi K2.5 have accumulated more than 500K downloads on Hugging Face.
At Ray Summit 2026, Yubo Wang, ML Research Engineer at Together AI / LightSeek, introduces the framework and the disaggregated approach to training speculative decoding draft models.
You'll leave knowing how to train drafters for large models at scale.
This session was part of the first vLLM Conference at Ray Summit, hosted by Inferact.
Liked this video? Check out other Ray Summit breakout session recordings: youtube.com/playlist?list=PLXMguE8Nc9o4
Subscribe to our YouTube channel to stay up-to-date on the future of AI: youtube.com/c/anyscale
đź”— Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com
TorchSpec is a torch-native speculative decoding training framework built on Ray, with a disaggregated design that lets each side of drafter training scale independently. Draft models trained with TorchSpec for large models such as Kimi K2.5 have accumulated more than 500K downloads on Hugging Face.
At Ray Summit 2026, Yubo Wang, ML Research Engineer at Together AI / LightSeek, introduces the framework and the disaggregated approach to training speculative decoding draft models.
You'll leave knowing how to train drafters for large models at scale.
This session was part of the first vLLM Conference at Ray Summit, hosted by Inferact.
Liked this video? Check out other Ray Summit breakout session recordings: youtube.com/playlist?list=PLXMguE8Nc9o4
Subscribe to our YouTube channel to stay up-to-date on the future of AI: youtube.com/c/anyscale
đź”— Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com










