Comment by porphyra

I think Atlas might also be slightly faster than vLLM:

https://flowtivity.ai/blog/120-tok-s-1m-context-private-ai-d...