[WS1][CUDA][Qwen3-Next] TP4 GDN, convolution and D=256 attention mixers on real weights (RFC #428 stack 6/8) - #507

[WS1][CUDA][Qwen3-Next] TP4 GDN, convolution and D=256 attention mixers on real weights (RFC #428 stack 6/8)#507

fusheng-ji wants to merge 71 commits into

RL-Align:test-qwennextRL-Align/RL-Kernel:test-qwennextfrom

fusheng-ji:feat/ws1-qwen3-next-tp4-mixersfusheng-ji/RL-Kernel:feat/ws1-qwen3-next-tp4-mixersCopy head branch name to clipboard