Replaced grouped_gemm with vLLM's Fused MoE Kernel for Inference Optimization

Replaced grouped_gemm with vLLM's Fused MoE Kernel for Inference Optimization #111