Open FeelingFatigued opened 11 months ago
We have a new kernel coming out very soon that obviates that algorithm and scales up to 4M - will post here when we release it!
I noticed the new FlashFFTConv repo was released recently. I've been trying to use that in this H3 model but I am not sure if I am changing the code correctly.
If it is possible, could you update H3 repo with FlashFFTConv? or could you tell me where do I need to change?
Thanks in advnace.
Ack’ing that I’ve seen this, will try to get it in this week or next!
On Fri, Nov 24, 2023 at 3:07 AM FeelingFatigued @.***> wrote:
I noticed the new FlashFFTConv repo was released recently. I've been trying to use that in this H3 model but I am not sure if I am changing the code correctly.
If it is possible, could you update H3 repo with FlashFFTConv? or could you tell me where do I need to change?
Thanks in advnace.
— Reply to this email directly, view it on GitHub https://github.com/HazyResearch/H3/issues/29#issuecomment-1825285275, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABDDIIUOZSRQUODR7NVUOUTYGBIUBAVCNFSM6AAAAAA6VRHSM2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMRVGI4DKMRXGU . You are receiving this because you commented.Message ID: @.***>
Sorry but can I know when will the FlashFFTConv-enabled H3 be available?
Thanks for your help.
Hello,
This repository does not seem to have the complete H3 described in the paper. (e.g., sequence length must be <= 8192 and I couldn't find state passing algorithm in the source)
If it is possible, would you please push the most recent version of the H3 model described in the paper?