Open
Description
https://github.com/automenta/bd3lms_gui
Training and Generation WORKING.
This is inspired by https://github.com/angrysky56/llada_gui_new
Potential for lots of visualization and showing how it works.
It needed some tricky setup, like changing the cache directories from "/share/..." to relative cache/
It defaults to FlashAttention but changing this to 'spds' got it working for me. b3dlms should fallback to this when FlashAttention isn't available.
Would you like to develop this further?
Metadata
Metadata
Assignees
Labels
No labels