torchchat
torchchat copied to clipboard
[AOTI] Add a --max-seq-length option for export
Summary: This improves best tokens/sec from 73 to 85.