Overview

Full name
docker.io/pytorch/torchserve-kfs
Registry
docker.io
Namespace
pytorch
Repository
torchserve-kfs
Cached tags
40
Last synced
07/06/2026, 06:57 PM

Introduction

PyTorch TorchServe Kubeflow Serving 运行时镜像,加载 MAR 模型包提供 gRPC/REST 推理,面向 KFS 与 GPU 节点,需 NVIDIA 驱动。

Details

torchserve-kfs 是 TorchServe 针对 Kubeflow ModelServer 定制的发行版,配合 InferenceService CR 自动挂载模型存储并暴露预测端口。容器启动需指定 MODEL_STORETS_CONFIG,GPU 推理须 --gpus all 或 K8s nvidia.com/gpu 资源。不适合无 GPU 的普通 Deployment 长期副本;正确路径是在 KServe/KFS 中声明 Predictor 或使用带 GPU 的 Job 批处理。

快速启动

docker run --rm --gpus all \
  -p 8080:8080 -p 8081:8081 \
  -v $(pwd)/model-store:/models \
  docker.io/pytorch/torchserve-kfs:latest \
  torchserve --start --model-store /models --models mymodel=mymodel.mar

推荐实践

docker run -d --name torchserve \
  --gpus all --restart unless-stopped \
  -p 8080:8080 -p 8081:8081 \
  -v /data/model-store:/models \
  -e TS_CONFIG=/models/config.properties \
  -e MODEL_STORE=/models \
  docker.io/pytorch/torchserve-kfs:0.9.0

核心参数说明

  • --gpus all — 启用 CUDA 推理
  • -p 8080/8081 — 推理 REST 与管理 API
  • -v .../model-store:/models — MAR 包与 config.properties
  • -e TS_CONFIG — 工作线程与批处理配置
  • -e MODEL_STORE — 模型目录路径
  • 部署 — 生产请用 KServe InferenceService 而非裸 Pod

Latest tags (20)

View all 40 tags →
TagArchPushed AtSizeDigestSyncWishAction
docker.io/pytorch/torchserve-kfs
:0.12.0-gpu
linux/amd64
2024/10/01 06:273.5 GBsha256:811e77013602
docker.io/pytorch/torchserve-kfs
:0.12.0
linux/arm64linux/amd64
2024/10/01 06:25923.5 MBsha256:17dd3ecdf5c8
docker.io/pytorch/torchserve-kfs
:0.11.1-gpu
linux/amd64
2024/07/19 02:483.5 GBsha256:8682e0808c96
docker.io/pytorch/torchserve-kfs
:0.11.1
linux/amd64
2024/07/19 02:47968.4 MBsha256:eb6ae9863dec
docker.io/pytorch/torchserve-kfs
:0.11.0-gpu
linux/amd64
2024/05/17 04:323.5 GBsha256:6870baa57a9b
docker.io/pytorch/torchserve-kfs
:0.11.0
linux/amd64
2024/05/17 04:31963.9 MBsha256:aa0897405450
docker.io/pytorch/torchserve-kfs
:0.10.0-gpu
linux/amd64
2024/03/15 07:213.5 GBsha256:c0bf621bf222
docker.io/pytorch/torchserve-kfs
:0.10.0
linux/amd64
2024/03/15 07:21965.9 MBsha256:47609d41aa1f
docker.io/pytorch/torchserve-kfs
:0.9.0-gpu
linux/amd64
2023/10/31 02:553.0 GBsha256:69756a69484b
docker.io/pytorch/torchserve-kfs
:0.9.0
linux/amd64
2023/10/31 02:541.0 GBsha256:d6cfdac5d830
docker.io/pytorch/torchserve-kfs
:0.8.2-gpu
linux/amd64
2023/08/29 05:463.1 GBsha256:0cab27524a8f
docker.io/pytorch/torchserve-kfs
:0.8.2
linux/amd64
2023/08/29 05:451001.1 MBsha256:98e0cb8ab48d
docker.io/pytorch/torchserve-kfs
:0.8.1-gpu
linux/amd64
2023/06/15 03:012.6 GBsha256:aaf0896a74ee
docker.io/pytorch/torchserve-kfs
:0.8.1
linux/amd64
2023/06/15 03:01927.7 MBsha256:32382de22a57
docker.io/pytorch/torchserve-kfs
:0.8.0-gpu
linux/amd64
2023/05/20 09:396.1 GBsha256:3c7e2c25399c
docker.io/pytorch/torchserve-kfs
:0.8.0
linux/amd64
2023/05/20 09:382.6 GBsha256:90cfc0d8ce9e
docker.io/pytorch/torchserve-kfs
:0.7.1-gpu
linux/amd64
2023/02/09 06:276.6 GBsha256:dc9f22e396e2
docker.io/pytorch/torchserve-kfs
:0.7.1
linux/amd64
2023/02/09 06:272.0 GBsha256:9cedaf72420e
docker.io/pytorch/torchserve-kfs
:0.7.0-gpu
linux/amd64
2022/12/13 06:006.6 GBsha256:6daae3c54cf6
docker.io/pytorch/torchserve-kfs
:0.7.0
linux/amd64
2022/12/13 06:002.0 GBsha256:05eff20e9277