Skip to content

[Cherry-Pick][RL] Support moe_topk_select using Paddle native operators and Add fused stack-transpose-quant for BlockWiseFP8 MoE weight quantization and swiglu-fp8-quant op for DeepGemmFusedMoE for training alignment (#6850) - #6935

Merged
zoooo0820 merged 16 commits into
PaddlePaddle:release/2.5from
DanielSun11:topk_cp_2.5
Mar 24, 2026
Merged

[Cherry-Pick][RL] Support moe_topk_select using Paddle native operators and Add fused stack-transpose-quant for BlockWiseFP8 MoE weight quantization and swiglu-fp8-quant op for DeepGemmFusedMoE for training alignment (#6850)#6935
zoooo0820 merged 16 commits into
PaddlePaddle:release/2.5from
DanielSun11:topk_cp_2.5

Commits

Commits on Mar 19, 2026

Commits on Mar 23, 2026