filmov
tv
Faster Model Serving with Ray and Anyscale | Ray Summit 2024
Показать описание
Ray Serve is an industry-leading ML platform for distributed model serving and deployment. In this Ray Summit 2024 breakout session, Edward Oakes and Akshay Malik from Anyscale explore how the Anyscale platform extends Ray Serve to solve key problems for serving large-scale AI models.
They also explore how the challenges of building AI applications have been accentuated by the rise of large-scale generative AI. Larger models are more expensive to initialize and run and even require special techniques such as tensor or pipeline parallelism across multiple GPUs. At the same time, they come with the same production-readiness and developer productivity challenges as hosting all ML models. The solution: Anyscale's Ray Serve
--
Interested in more?
--
🔗 Connect with us:
They also explore how the challenges of building AI applications have been accentuated by the rise of large-scale generative AI. Larger models are more expensive to initialize and run and even require special techniques such as tensor or pipeline parallelism across multiple GPUs. At the same time, they come with the same production-readiness and developer productivity challenges as hosting all ML models. The solution: Anyscale's Ray Serve
--
Interested in more?
--
🔗 Connect with us:
0:30:30
0:25:42
0:30:42
0:09:41
0:30:14
0:41:35
0:30:28
0:28:04
0:31:26
0:32:34
0:28:37
0:00:20
0:41:59
1:10:41
0:30:08
0:23:03
0:21:22
1:10:43
0:00:16
0:31:20
1:49:44
0:00:59
0:00:09
0:30:20