update

finetune/Dockerfile

@@ -22,7 +22,7 @@ apt-get update
 # rdma-core: 包含 RDMA 驱动和库
 # ibverbs-utils: 提供用于检查和配置 InfiniBand 设备的工具
 # perftest: 提供用于测试 InfiniBand 性能的工具
-apt-get install -y rdma-core ibverbs-utils perftest
+apt-get install -y rdma-core ibverbs-utils perftest libibverbs-dev infiniband-diags
 apt-get install -y wget curl htop jq vim bash libaio-dev build-essential openssh-server python3 python3-pip bzip2
 apt-get install -y --no-install-recommends software-properties-common build-essential autotools-dev nfs-common pdsh cmake g++ gcc curl wget vim tmux emacs less unzip htop iftop iotop ca-certificates openssh-client openssh-server rsync iputils-ping net-tools sudo llvm-dev re2c
 add-apt-repository ppa:git-core/ppa -y

finetune/docker-compose_nccl.yml

@@ -8,6 +8,8 @@ services:
     environment:
       - PORT=1998
       - PASS=P@88w0rd
+      - NVIDIA_VISIBLE_DEVICES=all
+      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
     volumes:
       - ./id_rsa_finetune:/root/.ssh/id_rsa
       - ./id_rsa.pub:/root/.ssh/id_rsa.pub

finetune/docker-compose_pytorch2.3_device.yml

@@ -51,7 +51,7 @@ services:
       - NVIDIA_DRIVER_CAPABILITIES=compute,utility
       - TMPDIR=/var/tmp
     networks:
-      - network_finetune
+      - my-custom-bridge
     deploy:
       resources:
         reservations:
@@ -63,7 +63,9 @@ services:
       - IPC_LOCK
     devices:
       - /dev/infiniband:/dev/infiniband
-
+# docker swarm init
+# docker swarm join-token manager
+# docker network create -d overlay --subnet=192.168.200.0/24 my-overlay-network
 networks:
-  network_finetune:
-    name: network_finetune
+  my-custom-bridge:
+    external: true

finetune/docker-compose_stack.yml

@@ -0,0 +1,58 @@
+version: '3.8'
+
+services:
+  ubuntu-finetune:
+    build:
+      context: .
+      dockerfile: Dockerfile
+      args:
+        PYTHON_VERSION: "3.10"
+        CUDA_VERSION: "12.1.0"
+        PYTORCH_VERSION: "2.3.0"
+        TORCHVISION_VERSION: "0.18.0"
+        TORCHAUDIO_VERSION: "2.3.0"
+        DS_BUILD_OPS: 1
+        USE_CUDA: 1
+        USE_ROCM: 0
+        USE_XPU: 0
+        CUDA: cu121
+        CUDA_ARCH_LIST: "80;86;89;90"
+        SETUPTOOLS_VERSION: "69.5.1"
+        DCUTLASS_NVCC_ARCHS: "80;86;89;90;90a"
+        DEEPSPEED_VERSION: "master"
+        DEEPSPEED_INSTALL_FLAGS: "--allow_sudo"
+    volumes:
+      - ./src:/bbtft
+      - ./id_rsa_finetune:/root/.ssh/id_rsa
+      - ./id_rsa.pub:/root/.ssh/id_rsa.pub
+    container_name: ubuntu-finetune
+    image: hotwa/deepspeed:pt23
+    shm_size: '32gb'
+    ports:
+      - 3228:22
+    environment:
+      - NVIDIA_VISIBLE_DEVICES=all
+      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
+      - TMPDIR=/var/tmp
+    networks:
+      - my-custom-bridge
+    deploy:
+      replicas: 2
+      resources:
+        reservations:
+          generic_resources:
+            - discrete_resource_spec:
+                kind: nvidia
+                value: 8
+      placement:
+        constraints: [node.platform.os == linux]
+    cap_add:
+      - IPC_LOCK
+    devices:
+      - /dev/infiniband:/dev/infiniband
+
+networks:
+  my-custom-bridge:
+    external: true
+
+# docker stack deploy -c docker-compose_stack.yml rdma_stack