What we talk about when we talk about system designmaheshba.bitbucket.io259 points·by scv119·3 years ago·35 comments
scv119·3 years ago·discussThe tweet is referring to a paper that fine tunes Chinese dataset on english base model. I'm not surprised with LoRA's poor result in this setup.
scv119·3 years ago·discussI believe you can slightly change the flash attention kernel to implement the same kernel of this page attention, since both of them work on the key/value cache at block level.
scv119·3 years ago·discussPretty cool stuff and the results are amazing. Hoping we will see virtual memory get standardized in pytorch or cuda.