Unable to create a directory: “/root/.config/miopen”

General
1

During docker run with rocm and 7900 XTX in ubuntu 24.04
got these messages:

(VllmWorker TP0 pid=189) INFO 10-01 21:36:04 [gpu_model_runner.py:2568] Encoder cache will be initialized with a budget of 32768 tokens, and profiled with 126 image items of the maximum feature size.
MIOpen(HIP): Warning [PlainTextDb] Unable to create a directory: “/root/.config/miopen”

2

It seems likely that this warning is due to the MIOpen library (used for AMD ROCm) trying to create a config directory in /root/.config/miopen, but the container or user may not have the necessary permissions, or the directory does not exist. This is a common, non-fatal warning when running ROCm-based workloads in Docker, especially if the container is not running as root or the home directory is not writable. The vLLM process itself is just profiling the encoder cache and should continue running unless you see further errors. For best results, ensure the container has write access to /root/.config or set the MIOPEN_USER_DB_PATH environment variable to a writable location if needed.

Would you like more detail on how to suppress or redirect this warning?

Sources:

Hint: Mention @RunLLM in the post for followups.

3

Could you please share the docker image that you’re using as well as the command to run it

4

The image is this:
rocm/vllm:latest

and run command this:

docker run -it
–dns=192.168.49.1
–restart=always
–network=host
–group-add=video
–ipc=host
–cap-add=SYS_PTRACE
–security-opt seccomp=unconfined
–device /dev/kfd
–device /dev/dri
–shm-size=16g
-e VLLM_SLEEP_WHEN_IDLE=1
-e ROCM_VISIBLE_DEVICES=0,1
-e HIP_VISIBLE_DEVICES=0,1
-e HSA_OVERRIDE_GFX_VERSION=11.0.0
-e PYTORCH_ROCM_ARCH=“gfx1100”
-e VLLM_USE_TRITON_FLASH_ATTN=0
-e GPU_MAX_HW_QUEUES=1
-e NCCL_DEBUG=WARN
-e NCCL_IB_DISABLE=1
–name vllm_rocm_gemma3_prod2
-v /home/ubuntu/vllm_models:/workspace/models
rocm/vllm:latest
vllm serve /workspace/models/gemma-3-12b-it/
–host 0.0.0.0
–port 8000
–served-model-name vllm/gemma-3
–tensor-parallel-size 2
–trust-remote-code
–dtype bfloat16
–kv-cache-dtype auto
–max-model-len 4096
–max-num-seqs 386
–max-num-batched-tokens 16384
–gpu-memory-utilization 0.88
–enable-prefix-caching
–enable-chunked-prefill
–swap-space 18
–disable-log-requests
–disable-log-stats
–max-log-len 100

5

That’s odd, I verified that your exact command runs on a machine with a similar gfx1101

Which ROCm version is installed on the host system?

6

rocm is 6.4.2.60402-120~24.04