xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-03-20 07:35:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix(ray): correct misleading warning message for multi-node clusters

Browse Source 
Fixes #31005

The warning message incorrectly stated "Tensor parallel size" when it
was actually comparing the world_size (TP × PP) against locally visible
GPUs. This was confusing for multi-node Ray deployments where:

1. The world_size includes both tensor and pipeline parallelism
2. The "available GPUs" count only reflects the local node, not the
   full Ray cluster

Changes:
- Replaced "Tensor parallel size" with "World size"
- Clarified that the GPU count is for locally visible devices
- Added context about multi-node Ray clusters
- Included the actual TP and PP values in the message for clarity

Signed-off-by: yurekami <yurekami@users.noreply.github.com>

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
yurekami 2025-12-25 01:20:43 +09:00
parent 5d93089686
commit 187bd5f7f3
1 changed files with 8 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
vllm/v1/executor/ray_utils.py
View File

@ -329,14 +329,17 @@ def initialize_ray_cluster(
available_gpus = cuda_device_count_stateless()
if parallel_config.world_size > available_gpus:
logger.warning(
"Tensor parallel size (%d) exceeds available GPUs (%d). "
"This may result in Ray placement group allocation failures. "
"Consider reducing tensor_parallel_size to %d or less, "
"or ensure your Ray cluster has %d GPUs available.",
"World size (%d) exceeds locally visible GPUs (%d). "
"For single-node deployments, this may result in Ray "
"placement group allocation failures. For multi-node Ray "
"clusters, ensure your cluster has %d GPUs available across "
"all nodes. (world_size = tensor_parallel_size=%d × "
"pipeline_parallel_size=%d)",
parallel_config.world_size,
available_gpus,
available_gpus,
parallel_config.world_size,
parallel_config.tensor_parallel_size,
parallel_config.pipeline_parallel_size,
)
if ray.is_initialized():