Great work on this and thanks for the detailed write up. In our experience this approach has worked really well for larger-scale multi-node training. We've seen up to 3x improvement in training speed training 32b models.

upvoted an article 5 months ago

Article

No GPU left behind: Unlocking Efficiency with Co-located vLLM in TRL

Jun 3

• 93

liked a model 6 months ago

ibm-granite/granite-3.3-8b-instruct-GGUF

Text Generation • 8B • Updated Apr 16 • 5.95k • 27

liked a model 8 months ago

ibm-granite/granite-3.2-8b-instruct

Text Generation • 8B • Updated Apr 17 • 4.6k • 87

Timothy

AI & ML interests

Recent Activity

Organizations

trbula92's activity

No GPU left behind: Unlocking Efficiency with Co-located vLLM in TRL