Merge branch 'master' of github.com:lablup/backend.ai-kernels

Author: hephaex

commons/Dockerfile.base.py38-cuda11.6

@@ -1,19 +1,19 @@
 FROM nvidia/cuda:11.6.0-cudnn8-devel-ubuntu20.04
 
 # TensorFlow version is tightly coupled to CUDA and cuDNN so it should be selected carefully
-ARG CUDA=11.5
-ENV CUDA_VERSION=11.5.0
+ARG CUDA=11.6
+ENV CUDA_VERSION=11.6.0
 ARG LIBNVINFER=8.2.0-1
 ARG LIBNVINFER_MAJOR_VERSION=8
 ARG CUDNN=8.2.4.15-1
 ENV CUDNN_VERSION 8.2.4.15
 ENV NCCL=2.11.4
 ENV NCCL_VERSION=2.11.4-1
-ENV NV_LIBNCCL_PACKAGE=libnccl2=2.11.5-1+cuda11.5
-ENV NV_LIBNPP_DEV_PACKAGE=libnpp-dev-11-5=11.5.0.110-1
-ENV NV_LIBNPP_DEV_VERSION=11.5.0.110-1
+ENV NV_LIBNCCL_PACKAGE=libnccl2=2.11.6-1+cuda11.6
+ENV NV_LIBNPP_DEV_PACKAGE=libnpp-dev-11-6=11.6.0.110-1
+ENV NV_LIBNPP_DEV_VERSION=11.6.0.110-1
 ENV NV_LIBCUSPARSE_DEV_VERSION=11.6.0.120-1
-ENV NV_NVML_DEV_VERSION=11.5.120-1
+ENV NV_NVML_DEV_VERSION=11.6.120-1
 ENV NV_CUDA_CUDART_DEV_VERSION=11.5.108-1
 ENV NV_CUDA_LIB_VERSION=11.5.2-1
 ENV NV_LIBCUBLAS_DEV_VERSION=11.6.1.51-1

vendor/ngc-triton/Dockerfile.22.07-py3

@@ -10,9 +10,15 @@ ENV PYTHONUNBUFFERED=1 \
 RUN apt update && \
     apt install -y --no-install-recommends \
       ncurses-term \
-      unzip zlib1g-dev htop && \
+      unzip zlib1g-dev htop curl git && \
     ln -sf /usr/share/terminfo/x/xterm-color /usr/share/terminfo/x/xterm-256color
 
+# Copy example models
+RUN git clone https://github.com/triton-inference-server/server.git /workspace/triton-server && \
+    cd /workspace/triton-server/docs/examples && \
+    ./fetch_models.sh && \
+    chmod -R +r /workspace/triton-server
+
 RUN update-alternatives --install /usr/bin/python python /usr/bin/python3 2
 COPY ./service-defs /etc/backend.ai/service-defs
 LABEL ai.backend.kernelspec="1" \
@@ -22,12 +28,14 @@ LABEL ai.backend.kernelspec="1" \
       ai.backend.accelerators="cuda" \
       ai.backend.resource.min.cpu="1" \
       ai.backend.resource.min.mem="1g" \
-      ai.backend.resource.min.cuda.device=1 \
-      ai.backend.resource.min.cuda.shares=0.1 \
+      ai.backend.resource.min.cuda.device=0 \
+      ai.backend.resource.min.cuda.shares=0 \
       ai.backend.base-distro="ubuntu16.04" \
       ai.backend.runtime-type="python" \
       ai.backend.runtime-path="/usr/bin/python" \
-      ai.backend.service-ports="triton-server:preopen:[8000,8001,8002]"
+      ai.backend.service-ports="triton-server-http:preopen:8000"
+      # ai.backend.service-ports="triton-server-http:preopen:8000,triton-server-grpc:preopen:8001,triton-server-metric:preopen:8002"
+      # ai.backend.service-ports="triton-server:preopen:[8000,8001,8002]"
 
 WORKDIR /home/work
 # vim: ft=dockerfile

vendor/ngc-triton/Dockerfile.22.07-py3-sdk

@@ -7,7 +7,10 @@ ENV PYTHONUNBUFFERED=1 \
     LD_LIBRARY_PATH="/workspace/install/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64" \
     LANG=C.UTF-8
 
-RUN apt update ; apt install ncurses-term -y && \
+RUN apt update && \
+    apt install -y --no-install-recommends \
+      ncurses-term \
+      unzip htop curl git && \
     ln -sf /usr/share/terminfo/x/xterm-color /usr/share/terminfo/x/xterm-256color
 
 RUN update-alternatives --install /usr/bin/python python /usr/bin/python3 2
@@ -19,8 +22,8 @@ LABEL ai.backend.kernelspec="1" \
       ai.backend.accelerators="cuda" \
       ai.backend.resource.min.cpu="1" \
       ai.backend.resource.min.mem="1g" \
-      ai.backend.resource.min.cuda.device=1 \
-      ai.backend.resource.min.cuda.shares=0.1 \
+      ai.backend.resource.min.cuda.device=0 \
+      ai.backend.resource.min.cuda.shares=0 \
       ai.backend.base-distro="ubuntu16.04" \
       ai.backend.runtime-type="python" \
       ai.backend.runtime-path="/usr/bin/python" \

vendor/ngc-triton/service-defs/triton-server-grpc.json

@@ -0,0 +1,7 @@
+{
+    "command": [
+        "/opt/tritonserver/bin/tritonserver",
+        "--model-repository",
+        "/workspace/triton-server/docs/examples/model_repository"
+    ]
+}

vendor/ngc-triton/service-defs/triton-server-http.json

@@ -0,0 +1,7 @@
+{
+    "command": [
+        "/opt/tritonserver/bin/tritonserver",
+        "--model-repository",
+        "/workspace/triton-server/docs/examples/model_repository"
+    ]
+}

vendor/ngc-triton/service-defs/triton-server-metric.json

@@ -0,0 +1,7 @@
+{
+    "command": [
+        "/opt/tritonserver/bin/tritonserver",
+        "--model-repository",
+        "/workspace/triton-server/docs/examples/model_repository"
+    ]
+}