1. Erasure Code Configuration The default erasure parity configuration is as follows: Erasure Set Size Default Parity (EC:M) 1 EC:0 2-3 EC:1 4-5 EC:2 6 - 7 EC:3 8 - 16 EC:4 The maximum parity is ERASURE_SET_SIZE/2, meaning the number of parity blocks cannot exceed the number of data blocks.

VLLM Benchmark is a tool provided by VLLM for testing model performance, supporting a variety of inference backends. This article mainly records some of the process of using VLLM Benchmark to test model performance. 1. Starting the Model Service 1 2 3 4 5 6 7 8 9 10 11 12 13 14 python -m vllm.

1. Building the hccl-test Image Download the dependency packages Python-3.8.18.tgz Ascend-cann-toolkit_8.0.RC2_linux-x86_64.run Ascend-cann-kernels-910b_8.0.RC2_linux.run mpich-3.2.1.tar.gz If downloading is inconvenient, you can also copy them straight out of the image I built. Write the Dockerfile 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 FROM ubuntu:22.

1. Building the nccl-test Image Check the CUDA version 1 2 3 nvidia-smi | grep "CUDA Version" | awk '{print $9}' 12.2 Write the Dockerfile 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 cat > Dockerfile << EOF FROM nvidia/cuda:12.1.0-cudnn8-devel-ubuntu22.04 ENV DEBIAN_FRONTEND=noninteractive ARG CONDA_VERSION WORKDIR /workspace ENV DEBIAN_FRONTEND=noninteractive RUN apt-get update && apt install -y openmpi-bin libopenmpi-dev ssh openssh-server net-tools vim git iputils-ping nfs-common RUN git clone https://github.

In the previous document, we took advantage of the elasticity Kubernetes provides to dynamically create Jenkins Slaves on Kubernetes. This document is mainly a stress test of Jenkins under large-scale builds. 1. Cluster Configuration 1.1 Kubernetes Version The version used here is v1.16.7 1 2 3 4 kubectl version Client Version: version.