Repository navigation
Community project: NextTransformer-V, a multimodal track built on NextTransformer #16
Selen-Suyue
announced in
Announcements
Replies: 2 comments
|
Thanks for the announcement! The raw-pixel replication finishes tonight (about 11:00 UTC); with raw 32 px patches the encoder supplies no position, so it tests the 2D mechanisms where they should matter most. I will post the results in this thread. Questions about the track are welcome in the repository's issues, and the upstream hooks it needs are in #14. |
0 replies
|
Results are in; the README has the details: https://github.com/Enderfga/NextTransformer-V#results
Runs, logs and evaluation files are in results/; the exploratory analyses are described in docs/spatial-addendum.md. |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
NextTransformer-V by @Enderfga takes NextTransformer to interleaved image and text sequences.
NextTransformer-V is a community project, maintained by its author in its own repository. Its results come from its own setup (5.24B tokens at an 8,192-token context), have not been audited by the NextTransformer maintainers, and are not part of the leaderboard. Questions about the project are best asked in its repository.
Building something on NextTransformer? Share it in Show and tell and we will list it under "Built on NextTransformer" in the README.
中文
NextTransformer-V(作者 @Enderfga)把 NextTransformer 扩展到图文交错的序列。
NextTransformer-V 是社区项目,由作者在自己的仓库中维护。它的结果来自自己的实验设置(8,192 长度上下文,52.4 亿 token),未经 NextTransformer 维护者审计,不属于排行榜。关于这个项目的问题,欢迎到它的仓库提出。
在做基于 NextTransformer 的项目?欢迎在 Show and tell 分享,我们会把它收录到 README 的“基于 NextTransformer 的项目”中。
All reactions