Open darissa opened 5 months ago
Hello, I would like to try out PTV3 with minimum requirement, but may know what is the outcome if disable the flash attention?
Thanks in advanced.
Hi, FlashAttention makes PTv3 faster, and the memory cost and forward latency won't increase even with an enlarged patch size.
Hello, I would like to try out PTV3 with minimum requirement, but may know what is the outcome if disable the flash attention?
Thanks in advanced.