lucidrains/megabyte-pytorch issues and pull requests

#20 - Allow usage as single-stage transformer

Issue - State: closed - Opened by eegli 20 days ago - 1 comment

#19 - Regression in padding value and loss calculation

Issue - State: closed - Opened by eegli 20 days ago - 3 comments

#18 - chore: update flash attention config

Pull Request - State: closed - Opened by eegli 29 days ago - 2 comments

#17 - fix: padding and positional embeddings

Pull Request - State: closed - Opened by eegli 29 days ago - 3 comments

#16 - Why does it expect tokens?

Issue - State: closed - Opened by tonydavis629 8 months ago - 1 comment

#15 - GPU used for original paper experiments

Issue - State: closed - Opened by itsnamgyu 10 months ago - 1 comment

#14 - Evaluation metric bits-per-byte

Issue - State: open - Opened by jxiw about 1 year ago - 1 comment

#13 - Why your Attention impl use kv dimention dim_head instead of inner_dim?

Issue - State: closed - Opened by Earthson about 1 year ago - 1 comment

#12 - Training Results and Scaling

Issue - State: open - Opened by MiscellaneousStuff over 1 year ago - 1 comment

#11 - the patch embbeder implementations are different from the original paper

Issue - State: closed - Opened by mikegreen7892003 over 1 year ago - 4 comments

#10 - Minor shape error

Issue - State: closed - Opened by anruigu over 1 year ago - 1 comment

#9 - cleanup

Pull Request - State: closed - Opened by lucidrains over 1 year ago

#8 - Paper variation

Pull Request - State: closed - Opened by lucidrains over 1 year ago

#7 - some implementations are different from the original paper

Issue - State: closed - Opened by ZihaoH over 1 year ago - 2 comments

#6 - No available kernel error

Issue - State: closed - Opened by missflash over 1 year ago - 1 comment

#5 - translation of model sizes from paper to model definition

Issue - State: open - Opened by winglian over 1 year ago

#4 - Some question about the MEGABYTE

Issue - State: closed - Opened by relic-yuexi over 1 year ago - 4 comments

#3 - the string is still divided into pieces

Issue - State: closed - Opened by wac81 over 1 year ago - 1 comment

#2 - What are the implications of this model?

Issue - State: closed - Opened by kyegomez over 1 year ago - 4 comments

#1 - Simplified batch handling in forward_empty()

Pull Request - State: closed - Opened by arman-hk over 1 year ago - 2 comments

Ecosyste.ms: Issues

GitHub / lucidrains/megabyte-pytorch issues and pull requests