A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 22:00 UTC
[Bugfix][TPU] Fix megacore setting for v5e-litepod (#6397)
Woosuk Kwon committed
Jul 12, 2024 at 15:59 UTC
f8f9ff57ee365891fe9f54cd46df65cc9d5ccca0
1 file changed
+1
-1
vllm/attention/backends/pallas.py
+1
-1
@@ -116,7 +116,7 @@ class PallasAttentionBackendImpl(AttentionImpl):
116
117
self.megacore_mode = None
118
tpu_type = torch_xla.tpu.get_tpu_env()["TYPE"].lower()
119
- if not tpu_type.endswith("lite"):
119
+ if "lite" not in tpu_type:
120
if self.num_kv_heads % 2 == 0:
121
self.megacore_mode = "kv_head"
122
else: