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 14, 05:30 UTC
[Misc] log more detailed message for ensure_model_parallel_initialized (#22144)
Signed-off-by: Andy Xie <andy.xning@gmail.com>
Ning Xie committed
Aug 5, 2025 at 10:36 UTC
bd3db7f46965bfc979734a6d4b50cf96184c10d8
1 file changed
+6
-6
vllm/distributed/parallel_state.py
+6
-6
@@ -1125,14 +1125,14 @@ def ensure_model_parallel_initialized(
1125
1126
assert (
1127
get_tensor_model_parallel_world_size() == tensor_model_parallel_size
1128
- ), ("tensor parallel group already initialized, but of unexpected size: "
1129
- f"{get_tensor_model_parallel_world_size()=} vs. "
1130
- f"{tensor_model_parallel_size=}")
1128
+ ), ("tensor parallel group already initialized, but of unexpected size. "
1129
+ f"got: {get_tensor_model_parallel_world_size()=} vs. "
1130
+ f"wanted: {tensor_model_parallel_size=}")
1131
pp_world_size = get_pp_group().world_size
1132
assert (pp_world_size == pipeline_model_parallel_size), (
1133
- "pipeline parallel group already initialized, but of unexpected size: "
1134
- f"{pp_world_size=} vs. "
1135
- f"{pipeline_model_parallel_size=}")
1133
+ "pipeline parallel group already initialized, but of unexpected size. "
1134
+ f"got: {pp_world_size=} vs. "
1135
+ f"wanted: {pipeline_model_parallel_size=}")
1136
1137
1138
def prepare_communication_buffer_for_model(model: torch.nn.Module):