Hacker News new | ask | show | jobs
by brucethemoose2 934 days ago
Quantizing it down to 8 bits seems to be one solution. TensorRT-LLM does this (and I think requires an H100)? exLlama also does this on much lesser hardware.
1 comments

wouldn't that mean trying to fit it on one machine?
Indeed :P

Honestly I'm not sure how context "sharding" works on multiple GPUs atm. Decent, really long context OSS models like Yi 200K and YARN finetunes are very new.