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 7, 15:15 UTC
[Core]Extract is_last_rank in Ray for tpu to override (#33012)
Signed-off-by: Chenyaaang <chenyangli@google.com>
Chenyaaang committed
Feb 26, 2026 at 19:18 UTC
06be53563ba24b44d788d43ef1afc99a5aef98f4
1 file changed
+4
-1
vllm/v1/executor/ray_utils.py
+4
-1
@@ -108,7 +108,7 @@ try:
108
109
if isinstance(output, AsyncModelRunnerOutput):
110
output = output.get_output()
111
- if not get_pp_group().is_last_rank:
111
+ if not self._is_last_rank():
112
# Case where there are no scheduled requests
113
# but may still be finished requests.
114
assert not output or not output.req_ids
@@ -128,6 +128,9 @@ try:
128
def _is_intermediate_tensors(self, output) -> bool:
129
return isinstance(output, IntermediateTensors)
130
131
+ def _is_last_rank(self) -> bool:
132
+ return get_pp_group().is_last_rank
133
+
134
ray_import_err = None
135
136
except ImportError as e: