Peng Jiang comments

Results 24 comments of


                                            Peng Jiang

The gpustack worker got stuck during startup and failed to launch successfully.

@lvturner @guohuanliang1 Please try with v2.0.1 to see if it helps.

Does gpustack 2.0.0 support Nvidia DGX Spark?

It's supported by GPUStack v2.0.0 by default. But you need to add a custom vllm/sglang backend from Nvidia, as it requires CUDA 13.0, which is not yet supported by the...

After upgrading to version 2.0.0 failed, I downgraded to version 0.7.1, but the single-machine multi-GPU deployment of the model failed and reported an error.

The error is because: �[1;36m(EngineCore_0 pid=17619)�[0;0m ERROR 11-27 10:18:36 [core.py:700] Error while creating shared memory segment /dev/shm/nccl-s0On3b (size 33030528), error: No space left on device (28)', please check the stack...

After upgrading to version 2.0.0 failed, I downgraded to version 0.7.1, but the single-machine multi-GPU deployment of the model failed and reported an error.

The other error may also be related to shared memory. How did you deploy GPUStack? For Docker deployment, please ensure that you use the --ipc=host parameter. For K8s deployment, please...