vllm

Fix for attention layers to remain unquantized during moe_wn16 quant

#12570

Merged