LoRAX: Open-Source Serving for 100s of Fine-Tuned LLMs in Production
predibase.comA great framework for serving many fine-tuned llms in production by quickly swapping adapters for the same base model (eg. Llama-2-70b)
Whoa this looks pretty cool. One question though: is there increased latency when you have multiple adapters on a single base model?
Hey, LoRAX dev here. This was one thing we spent a lot of effort optimizing. The TL;DR is that in most cases latency will be with 80% of the baseline latency with 0 adapters with as many as 128 adapters at once under heavy request load. Check out the section Results in the blog for more details and let me know if you have any questions!