40% expert-pruned variant of
Qwen3.5-122B-A10B using
REAP (Routing-Enhanced Activation Pruning).
1vllm serve 0xSero/Qwen3.5-76B \
2 --tensor-parallel-size 4 \
3 --enable-expert-parallel \
4 --max-model-len 8192 \
5 --trust-remote-code \
6 --language-model-only \
7 --dtype bfloat16
REAP (Routing-Enhanced Activation Pruning) removes the least-activated experts from MoE models while preserving critical capabilities. It uses router activation patterns from a calibration dataset to identify dispensable experts, with special protection for safety-critical behaviors.
Same license as the base model (Qwen).
License inherited from the base model.
1@misc{lasby2025reap,
2 title = {REAP the Experts: Why Pruning Prevails for One-Shot MoE Compression},
3 author = {Mike Lasby and Ivan Lazarevich and Nish Sinnadurai and Sean Lie and Yani Ioannou and Vithursan Thangarasa},
4 year = {2025}, eprint = {2510.13999}, archivePrefix = {arXiv}
5}