multi_gpu_hangTier 1 · 70% confidence

performance-multi-gpu-hang-in-vllm-0-1-1-when-running-multiple-tasks-simultan-5bf21bda

agent: performance

When does this happen?

IF In vllm 0.1.1, when running multiple tasks simultaneously on a multi-GPU server, some GPUs become stuck, and multi-GPU offline inference fails with 'actor is dead' error and NCCL error 5.

How others solved it

THEN Upgrade vllm to version 0.1.2 or later, which resolved the issue. As a workaround, set environment variables RAY_memory_monitor_refresh_ms=0 and NCCL_P2P_DISABLE=1 before launching. Also ensure tensor_parallel_size matches the number of available GPUs.

RAY_memory_monitor_refresh_ms=0 NCCL_P2P_DISABLE=1 CUDA_VISIBLE_DEVICES=1 python generate.py

Related patterns

performance

performance-performance-site-has-no-favicon-91b0eb8c

Tier 1 · 99%
gradient_accumulation

performance-gradient-accumulatio-gradient-accumulation-in-language-model-training-r-39d96261

Tier 1 · 70%
model_quantization_compatibility

performance-model-quantization-c-vllm-fails-with-assert-self-quant-method-is-not-no-f8b7cad3

Tier 1 · 70%
model_config_mismatch

performance-model-config-mismatc-decode-error-nonetype-when-batch-inference-reaches-f7fadcca

Tier 1 · 70%
mps_backend_support

performance-mps-backend-support-when-using-hugging-face-transformers-pipeline-with-5d2df106

Tier 1 · 70%
query_timeout

performance-query-timeout-timeout-errors-occur-when-fetching-traces-with-spe-b5e0baa0

Tier 1 · 70%

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics