From f7dac83d95ae38973b425a8bb2d3a3df9fe9a9c2 Mon Sep 17 00:00:00 2001
From: Cody Yu <hao.yu.cody@gmail.com>
Date: Sat, 29 Jun 2024 06:04:20 -0700
Subject: [PATCH] [Kernel] Raise an exception in MoE kernel if the batch size
 is larger then 65k (#5939)

---
 vllm/model_executor/layers/fused_moe/fused_moe.py | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/vllm/model_executor/layers/fused_moe/fused_moe.py b/vllm/model_executor/layers/fused_moe/fused_moe.py
index b750fc713b43f..ecab77a8b6dfb 100644
--- a/vllm/model_executor/layers/fused_moe/fused_moe.py
+++ b/vllm/model_executor/layers/fused_moe/fused_moe.py
@@ -423,6 +423,11 @@ def fused_experts(hidden_states: torch.Tensor,
     M, _ = hidden_states.shape
     E, N, _ = w1.shape
 
+    if M > 65536:
+        # https://github.com/vllm-project/vllm/issues/5938
+        raise ValueError("MoE kernel does not support more than 65536 tokens, "
+                         f"but got {M}")
+
     if override_config:
         config = override_config
     else: