adapt both fla-attn and fused_moe

Signed-off-by: Ceng23333 <441651826@qq.com>

Author: Ceng23333

pyproject.toml

@@ -13,6 +13,8 @@ requires-python = ">=3.8"
 [project.optional-dependencies]
 # Same interpreter as InfiniCore built with ``--aten=y`` (see vllm_kernel_reuse_evaluation.md).
 vllm = ["vllm==0.19.0"]
+# MiniCPM5 fused stub + jiuge: vLLM fused MoE + helper deps (flash-attn install separately; see InfiniLM/examples/FLASH_ATTN_AND_VLLM_FUSED_MOE.md).
+vllm-fused-moe = ["vllm==0.19.0", "janus>=1.0.0", "xxhash>=3.0.0"]
 
 classifiers = [
     "Programming Language :: Python :: 3",