From f8f9ff57ee365891fe9f54cd46df65cc9d5ccca0 Mon Sep 17 00:00:00 2001
From: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Date: Fri, 12 Jul 2024 15:59:47 -0700
Subject: [PATCH] [Bugfix][TPU] Fix megacore setting for v5e-litepod (#6397)

---
 vllm/attention/backends/pallas.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/vllm/attention/backends/pallas.py b/vllm/attention/backends/pallas.py
index 7a6954ceb6d6a..c45f7b28b2afb 100644
--- a/vllm/attention/backends/pallas.py
+++ b/vllm/attention/backends/pallas.py
@@ -116,7 +116,7 @@ class PallasAttentionBackendImpl(AttentionImpl):
 
         self.megacore_mode = None
         tpu_type = torch_xla.tpu.get_tpu_env()["TYPE"].lower()
-        if not tpu_type.endswith("lite"):
+        if "lite" not in tpu_type:
             if self.num_kv_heads % 2 == 0:
                 self.megacore_mode = "kv_head"
             else: