Training-free MoE expert pruning by functional replaceability. Six backbones at 25% and 50% expert removal, no recovery training.
-
Nickyang/Qwen3.8-Flash-Next-RAZOR-65B-A6B-E256of512
Image-Text-to-Text • 118B • Updated • 54 • 2 -
Nickyang/Qwen3.8-Flash-Next-RAZOR-96B-A6B-E384of512
Image-Text-to-Text • 149B • Updated • 50 • 2 -
Nickyang/Gemma-4-26B-A4B-it-RAZOR-14B-A4B-E64of128
Image-Text-to-Text • 14B • Updated • 31 • 2 -
Nickyang/Gemma-4-26B-A4B-it-RAZOR-20B-A4B-E96of128
Image-Text-to-Text • 20B • Updated • 129 • 2