diff --git a/docs/setup/integration/hugging-face.md b/docs/setup/integration/hugging-face.md index 73007f1d..141aa2a4 100644 --- a/docs/setup/integration/hugging-face.md +++ b/docs/setup/integration/hugging-face.md @@ -27,7 +27,7 @@ thereby accelerating file downloading. For detailed installation documentation based on kubernetes cluster, please refer to [quick-start-kubernetes](../../getting-started/quick-start/kubernetes.md). -## Setup kubernetes cluster {#setup-kubernetes-cluster} +### Setup kubernetes cluster {#setup-kubernetes-cluster} Create kind multi-node cluster configuration file `kind-config.yaml`, configuration content is as follows: @@ -55,7 +55,7 @@ Switch the context of kubectl to kind cluster: kubectl config use-context kind-kind ``` -## Kind loads dragonfly image {#kind-loads-dragonfly-image} +### Kind loads dragonfly image {#kind-loads-dragonfly-image} Pull dragonfly latest images: @@ -73,7 +73,7 @@ kind load docker-image dragonflyoss/manager:latest kind load docker-image dragonflyoss/dfdaemon:latest ``` -## Create dragonfly cluster based on helm charts {#create-dragonfly-cluster-based-on-helm-charts} +### Create dragonfly cluster based on helm charts {#create-dragonfly-cluster-based-on-helm-charts} Create helm charts configuration file `charts-config.yaml` and set `dfdaemon.config.proxy.registryMirror.url` to the address of the Hugging Face Hub's LFS server, configuration content is as follows: diff --git a/docs/setup/integration/stargz.md b/docs/setup/integration/stargz.md index 20c1324d..0ba7bcd3 100644 --- a/docs/setup/integration/stargz.md +++ b/docs/setup/integration/stargz.md @@ -24,7 +24,7 @@ This document will help you experience how to use dragonfly with eStargz. For detailed installation documentation based on kubernetes cluster, please refer to [quick-start-kubernetes](../../getting-started/quick-start/kubernetes.md). -## Setup kubernetes cluster {#setup-kubernetes-cluster} +### Setup kubernetes cluster {#setup-kubernetes-cluster} Create kind multi-node cluster configuration file `kind-config.yaml`, configuration content is as follows: @@ -54,7 +54,7 @@ Switch the context of kubectl to kind cluster: kubectl config use-context kind-kind ``` -## Kind loads dragonfly image {#kind-loads-dragonfly-image} +### Kind loads dragonfly image {#kind-loads-dragonfly-image} Pull dragonfly latest images: @@ -72,7 +72,7 @@ kind load docker-image dragonflyoss/manager:latest kind load docker-image dragonflyoss/dfdaemon:latest ``` -## Create dragonfly cluster based on helm charts {#create-dragonfly-cluster-based-on-helm-charts} +### Create dragonfly cluster based on helm charts {#create-dragonfly-cluster-based-on-helm-charts} Create helm charts configuration file `charts-config.yaml` and enable prefetching, configuration content is as follows: diff --git a/i18n/zh/docusaurus-plugin-content-docs/current/setup/integration/hugging-face.md b/i18n/zh/docusaurus-plugin-content-docs/current/setup/integration/hugging-face.md index 73007f1d..34de8e70 100644 --- a/i18n/zh/docusaurus-plugin-content-docs/current/setup/integration/hugging-face.md +++ b/i18n/zh/docusaurus-plugin-content-docs/current/setup/integration/hugging-face.md @@ -3,33 +3,35 @@ id: hugging-face title: Hugging Face --- -This document will help you experience how to use dragonfly with hugging face. +本文档将帮助您将 Dragonfly 与 Hugging Face 一起使用。 -During the downloading of datasets or models, the file size is large and there are many services -downloading the files at the same time. The bandwidth of the storage will reach the limit and the download will be slow. -Therefore, Dragonfly can be used to eliminate the bandwidth limit of the storage through P2P technology, -thereby accelerating file downloading. +当在 Hugging Face 下载数据集以及模型的时候，文件相对较大且会有并发下载文件的场景。 +这样很容易导致 Hugging Face 的 Git LFS 存储带宽被打满，从而引起下载过慢的情况，影响 +训练以及推理服务的使用。这种方式比较好的解决方案是使用 Dragonfly 的 P2P 技术利用 +每个节点的闲置带宽缓解 Git LFS 存储的带宽压力，从而达到加速效果。在最理想的情况下 +Dragonfly 可以让整个 P2P 集群中只有一个节点回源 Hugging Face 下载数据集或模型，其他 +节点流量均使用集群内 P2P 内网带宽。 -## Prerequisites {#prerequisites} +## 依赖 -| Name | Version | Document | -| ------------------ | ------- | --------------------------------------- | -| Kubernetes cluster | 1.20+ | [kubernetes.io](https://kubernetes.io/) | -| Helm | 3.8.0+ | [helm.sh](https://helm.sh/) | +| 所需软件 | 版本要求 | 文档 | +| ------------------ | -------- | --------------------------------------- | +| Kubernetes cluster | 1.20+ | [kubernetes.io](https://kubernetes.io/) | +| Helm | 3.8.0+ | [helm.sh](https://helm.sh/) | -**Notice:** [Kind](https://kind.sigs.k8s.io/) is recommended if no kubernetes cluster is available for testing. +**注意:** 如果没有可用的 Kubernetes 集群进行测试，推荐使用 [Kind](https://kind.sigs.k8s.io/)。 -## Install dragonfly {#install-dragonfly} +## 安装 Dragonfly -For detailed installation documentation based on kubernetes cluster, please refer to [quick-start-kubernetes](../../getting-started/quick-start/kubernetes.md). +基于 Kubernetes cluster 详细安装文档可以参考 [quick-start-kubernetes](../../getting-started/quick-start/kubernetes.md)。 -## Setup kubernetes cluster {#setup-kubernetes-cluster} +### 使用 Kind 安装 Kubernetes 集群 -Create kind multi-node cluster configuration file `kind-config.yaml`, configuration content is as follows: +创建 Kind 多节点集群配置文件 `kind-config.yaml`, 配置如下: ```yaml kind: Cluster @@ -43,21 +45,21 @@ nodes: - role: worker ``` -Create a kind multi-node cluster using the configuration file: +使用配置文件创建 Kind 集群: ```shell kind create cluster --config kind-config.yaml ``` -Switch the context of kubectl to kind cluster: +切换 Kubectl 的 context 到 Kind 集群: ```shell kubectl config use-context kind-kind ``` -## Kind loads dragonfly image {#kind-loads-dragonfly-image} +### Kind 加载 Dragonfly 镜像 -Pull dragonfly latest images: +下载 Dragonfly latest 镜像: ```shell docker pull dragonflyoss/scheduler:latest @@ -65,7 +67,7 @@ docker pull dragonflyoss/manager:latest docker pull dragonflyoss/dfdaemon:latest ``` -Kind cluster loads dragonfly latest images: +Kind 集群加载 Dragonfly latest 镜像: ```shell kind load docker-image dragonflyoss/scheduler:latest @@ -73,10 +75,10 @@ kind load docker-image dragonflyoss/manager:latest kind load docker-image dragonflyoss/dfdaemon:latest ``` -## Create dragonfly cluster based on helm charts {#create-dragonfly-cluster-based-on-helm-charts} +### 基于 Helm Charts 创建 Dragonfly P2P 集群 -Create helm charts configuration file `charts-config.yaml` and set `dfdaemon.config.proxy.registryMirror.url` to -the address of the Hugging Face Hub's LFS server, configuration content is as follows: +创建 Helm Charts 配置文件 `charts-config.yaml` 并且设置 `dfdaemon.config.proxy.registryMirror.url` 为 +Hugging Face 的 LFS 服务的地址, 配置如下: ```yaml scheduler: @@ -135,7 +137,7 @@ manager: pprofPort: 18066 ``` -Create a dragonfly cluster using the configuration file: +使用配置文件部署 Dragonfly Helm Charts: @@ -166,7 +168,7 @@ NOTES: -Check that dragonfly is deployed successfully: +检查 Dragonfly 是否部署成功: ```shell $ kubectl get po -n dragonfly-system @@ -183,7 +185,7 @@ dragonfly-scheduler-0 1/1 Running 0 3m27s dragonfly-seed-peer-0 1/1 Running 2 (95s ago) 3m27s ``` -Create peer service configuration file `peer-service-config.yaml`, configuration content is as follows: +创建 Peer Service 配置文件 `peer-service-config.yaml` 配置如下: ```yaml apiVersion: v1 @@ -203,27 +205,24 @@ spec: release: dragonfly ``` -Create a peer service using the configuration file: +使用配置文件部署 Peer Service: ```shell kubectl apply -f peer-service-config.yaml ``` -## Use Hub Python Library to download files and distribute traffic through Draognfly {#use-hub-python-library-to-download-files-and-distribute-traffic-through-draognfly} +## 通过 Dragonfly 分发 Hub Python Library 的下载文件流量 -Any API in the [Hub Python Library](https://huggingface.co/docs/huggingface_hub/index) -that uses `Requests` library for downloading files can -distribute the download traffic in the P2P network by -setting `DragonflyAdapter` to the requests `Session`. +任何 [Hub Python Library](https://huggingface.co/docs/huggingface_hub/index) 的 API 使用 +`Requests` 库下载文件，都可以通过设置 `DragonflyAdapter` 将流量使用 Dragonfly 分发。 -### Download a single file with Dragonfly {#download-a-single-file-with-dragonfly} +### 使用 Dragonfly 下载单个文件 -A single file can be downloaded using the [`hf_hub_download`](https://huggingface.co/docs/huggingface_hub/v0.17.1/en/package_reference/file_download#huggingface_hub.hf_hub_download), -distribute traffic through the Dragonfly peer. +下载单个文件可以使用 [`hf_hub_download`](https://huggingface.co/docs/huggingface_hub/v0.17.1/en/package_reference/file_download#huggingface_hub.hf_hub_download)， +并且通过 Dragonfly 分发流量。 -Create `hf_hub_download_dragonfly.py` file. Use `DragonflyAdapter` to forward the file download request of -the LFS protocol to Dragonfly HTTP proxy, so that it can use the P2P network -to distribute file, content is as follows: +创建 `hf_hub_download_dragonfly.py` 文件，使用 `DragonflyAdapter` 将下载流量转发至 Dragonfly HTTP Proxy。 +这样可以通过 P2P 网络分发流量，内容如下： ```python import requests @@ -262,7 +261,7 @@ configure_http_backend(backend_factory=backend_factory) hf_hub_download(repo_id="tiiuae/falcon-rw-1b", filename="pytorch_model.bin") ``` -Download a single file of th LFS protocol with Dragonfly: +通过 Dragonfly 基于 LFS 协议下载单个文件： @@ -273,9 +272,9 @@ $ python3 hf_hub_download_dragonfly.py -#### Verify a single file download with Dragonfly {#verify-a-single-file-download-with-dragonfly} +#### 验证基于 Dragonfly 下载单个文件 -Execute the command: +执行命令： ```shell # find pods @@ -285,7 +284,7 @@ pod_name=dfdaemon-xxxxx kubectl -n dragonfly-system exec -it ${pod_name} -- grep "peer task done" /var/log/dragonfly/daemon/core.log ``` -Example output: +日志输出： @@ -295,15 +294,13 @@ peer task done, cost: 28349ms {"peer": "89.116.64.101-77008-a95a6918-a52b-47f5-9 -### Download a snapshot of the repo with Dragonfly {#download-a-snapshot-of-the-repo-with-dragonfly} +### 使用 Dragonfly 下载仓库快照 -A snapshot of the repo can be downloaded using the [`snapshot_download`](https://huggingface.co/docs/huggingface_hub/v0.17.1/en/package_reference/file_download#huggingface_hub.snapshot_download), -distribute traffic through the Dragonfly peer. +下载仓库快照可以使用 [`snapshot_download`](https://huggingface.co/docs/huggingface_hub/v0.17.1/en/package_reference/file_download#huggingface_hub.snapshot_download)， +并且通过 Dragonfly 分发流量。 -Create `snapshot_download_dragonfly.py` file. Use `DragonflyAdapter` to forward the file download request of -the LFS protocol to Dragonfly HTTP proxy, so that it can use the P2P network -to distribute file. Only the files of the LFS protocol will be distributed -through the Dragonfly P2P network. content is as follows: +创建 `snapshot_download_dragonfly.py` 文件，使用 `DragonflyAdapter` 将下载流量转发至 Dragonfly HTTP Proxy。 +只有 Git LFS 协议的大文件流量会通过 P2P 网络分发，内容如下： ```python import requests @@ -342,7 +339,7 @@ configure_http_backend(backend_factory=backend_factory) snapshot_download(repo_id="tiiuae/falcon-rw-1b") ``` -Download a snapshot of the repo with Dragonfly: +通过 Dragonfly 基于 LFS 协议下载仓库快照： @@ -365,9 +362,9 @@ Fetching 12 files: 100%|██████████████████ -#### Verify a snapshot of the repo download with Dragonfly {#verify-a-snapshot-of-the-repo-download-with-dragonfly} +#### 验证基于 Dragonfly 下载仓库快照 -Execute the command: +执行命令： ```shell # find pods @@ -377,7 +374,7 @@ pod_name=dfdaemon-xxxxx kubectl -n dragonfly-system exec -it ${pod_name} -- grep "peer task done" /var/log/dragonfly/daemon/core.log ``` -Example output: +日志输出： @@ -387,27 +384,26 @@ peer task done, cost: 28349ms {"peer": "89.116.64.101-77008-a95a6918-a52b-47f5-9 -## Performance testing {#performance-testing} +## 性能测试 -Test the performance of single-machine file download by `hf_hub_download` API after the integration of -Hugging Face Python Library and Dragonfly P2P. -Due to the influence of the network environment of the machine itself, the actual download time is not important, -but the ratio of the increase in the download time in different scenarios is very important. +测试 Hugging Face Python Library 的 `hf_hub_download` API 与 Dragonfly 集成后的 +单机模型文件下载的性能。 +由于机器本身网络环境、配置等影响，实际下载时间不具有参考价值， +但是不同场景下载时间所提升的比率是有重要意义的。 ![hugging-face-dragonfly](../../resource/setup/hugging-face-dragonfly.png) -- Hugging Face Python Library: Use `hf_hub_download` API to download models directly. -- Hugging Face Python Library & Dragonfly Cold Boot: Use `hf_hub_download` API to download models via Dragonfly P2P network and no cache hits. -- Hit Dragonfly Remote Peer Cache: Use `hf_hub_download` API to download models via Dragonfly P2P network and hit the remote peer cache. -- Hit Dragonfly Local Peer Cache: Use `hf_hub_download` API to download models via Dragonfly P2P network and hit the local peer cache. -- Hit Hugging Face Cache: Use `hf_hub_download` API to download models via Dragonfly P2P network and hit the Hugging Face local cache. +- Hugging Face Python Library: 使用 `hf_hub_download` API 直接下载模型文件。 +- Hugging Face Python Library & Dragonfly Cold Boot: 使用 `hf_hub_download` API 直接下载模型文件，没有命中任何缓存。 +- Hit Dragonfly Remote Peer Cache: 使用 `hf_hub_download` API 直接下载模型文件，在命中 Dragonfly 的远端 Peer 缓存。 +- Hit Dragonfly Remote Local Cache: 使用 `hf_hub_download` API 直接下载模型文件，在命中 Dragonfly 的本地 Peer 缓存。 +- Hit Hugging Face Cache: 使用 `hf_hub_download` API 直接下载模型文件，在命中 Hugging Face 的缓存。 -Test results show Hugging Face Python Library and Dragonfly P2P integration. -It can effectively reduce the file download time. -Note that this test was a single-machine test, which means that in the case of cache hits, -the performance limitation is on the disk. -If Dragonfly is deployed on multiple machines for P2P download, the models download speed will be faster. +测试结果表明 Hugging Face Python Library 和 Dragonfly 集成。 +能够有效减少模型文件下载时间。测试是在单机情况下基本在缓存命中情况下， +性能瓶颈在于磁盘。如果在多节点并发下载数据集或者模型的情况下， +Dragonfly 效果会更加明显。