Skip to content

Commit 259bb03

Browse files
armstrongttwalker-altarmstrongttwalker-alt
authored
Merge pull request #398 from cheng874/final-update
Updated docker image information and fixed sglang-plugin-fl link errors
2 parents fc23a01 + a681cc9 commit 259bb03

9 files changed

Lines changed: 119 additions & 0 deletions

File tree

‎docs/megatron_lm_fl_en/getting_started/install.md‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,14 @@ You can install Megatron-LM-FL through one of the following methods:
44

55
## Docker (Recommended)
66

7+
### FlagOS Release Image (v0.2.0-rc2)
8+
9+
```bash
10+
docker pull harbor.baai.ac.cn/flagos21-release/megatron-lm-fl:v0.2.0-rc2-nvidia
11+
```
12+
13+
内含：torch 2.4.0a0, triton 3.0.0, trans-engine 2.14.0。适用于千亿参数模型预训练。
14+
715
### CUDA
816

917
```bash

‎docs/megatron_lm_fl_zh/getting_started/install.md‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,14 @@
44

55
## Docker（推荐）
66

7+
### FlagOS 发版镜像（v0.2.0-rc2）
8+
9+
```bash
10+
docker pull harbor.baai.ac.cn/flagos21-release/megatron-lm-fl:v0.2.0-rc2-nvidia
11+
```
12+
13+
内含：torch 2.4.0a0, triton 3.0.0, trans-engine 2.14.0。适用于千亿参数模型预训练。
14+
715
### CUDA
816

917
```bash

‎docs/pytorch_plugin_fl_en/getting_started/install.md‎

Lines changed: 14 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,19 @@
11
# Installation
22

3+
## Docker Images (Recommended)
4+
5+
| Platform | Image | Contents |
6+
|----------|-------|----------|
7+
| NVIDIA GPU | `harbor.baai.ac.cn/flagos21-release/pytorch-plugin-fl:v0.1.0-rc2-nvidia` | torch 2.11.0+cu128, torch_fl 0.1.0, flag_gems 5.0.2, triton 3.6.0 |
8+
| Huawei Ascend | `harbor.baai.ac.cn/flagos21-release/pytorch-plugin-fl:v0.1.0-rc2-ascend` | torch 2.11.0+CPU, torch_npu, torch_fl |
9+
10+
```bash
11+
# NVIDIA
12+
docker pull harbor.baai.ac.cn/flagos21-release/pytorch-plugin-fl:v0.1.0-rc2-nvidia
13+
# Huawei Ascend
14+
docker pull harbor.baai.ac.cn/flagos21-release/pytorch-plugin-fl:v0.1.0-rc2-ascend
15+
```
16+
317
## Build from Source
418

519
### CUDA platform

‎docs/pytorch_plugin_fl_zh/getting_started/install.md‎

Lines changed: 14 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,19 @@
11
# 安装
22

3+
## Docker 镜像（推荐）
4+
5+
| 平台 | 镜像 | 内容 |
6+
|----------|-------|----------|
7+
| NVIDIA GPU | `harbor.baai.ac.cn/flagos21-release/pytorch-plugin-fl:v0.1.0-rc2-nvidia` | torch 2.11.0+cu128, torch_fl 0.1.0, flag_gems 5.0.2, triton 3.6.0 |
8+
| 华为昇腾 | `harbor.baai.ac.cn/flagos21-release/pytorch-plugin-fl:v0.1.0-rc2-ascend` | torch 2.11.0+CPU, torch_npu, torch_fl |
9+
10+
```bash
11+
# NVIDIA
12+
docker pull harbor.baai.ac.cn/flagos21-release/pytorch-plugin-fl:v0.1.0-rc2-nvidia
13+
# 华为昇腾
14+
docker pull harbor.baai.ac.cn/flagos21-release/pytorch-plugin-fl:v0.1.0-rc2-ascend
15+
```
16+
317
## 从源码构建
418

519
### CUDA 平台

‎docs/sglang_plugin_fl_en/getting_started/getting-started.md‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -6,6 +6,7 @@ This section covers the requirements for installing sglang-plugin-FL and guides
66
:maxdepth: 2
77
88
requirements.md
9+
install.md
910
run-inference-task.md
1011
1112
```

‎docs/sglang_plugin_fl_en/getting_started/install.md‎

Lines changed: 28 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,28 @@
1+
# Install sglang-plugin-FL
2+
3+
## Docker Images (Recommended)
4+
5+
Pre-built Docker images for v0.1.0-rc2:
6+
7+
| Platform | Image | Contents |
8+
|----------|-------|----------|
9+
| NVIDIA GPU (dual-node) | `harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-nvidia-dual` | sglang 0.5.11, flag_gems 5.3.0rc2, torch 2.11.0, triton 3.6.0 |
10+
| NVIDIA GPU (single-node) | `harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-nvidia-single` | sglang 0.5.11, flag_gems 5.3.0rc2, torch 2.11.0, triton 3.6.0 |
11+
| Moore Threads MUSA | `harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-musa` | sglang 0.5.12, torch 2.9.0, flag_gems 5.0.2 |
12+
| Moore Threads (SVT) | `harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-mthreads-svt` | sglang 0.5.11, flag_gems 5.3.0rc2, torch 2.9.0, triton 3.1.0 |
13+
| Huawei Ascend | `harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-ascend` | sglang 0.5.12, flag_gems 5.0.2, CANN 8.5.0 |
14+
15+
```bash
16+
# NVIDIA dual-node (cross-node inference)
17+
docker pull harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-nvidia-dual
18+
# NVIDIA single-node
19+
docker pull harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-nvidia-single
20+
# Moore Threads MUSA
21+
docker pull harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-musa
22+
# Moore Threads SVT (full-stack test)
23+
docker pull harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-mthreads-svt
24+
# Huawei Ascend
25+
docker pull harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-ascend
26+
```
27+
28+
Dual-node images support cross-node LLM inference. Single-node images support single-machine inference. SVT images are full-stack test images.

‎docs/sglang_plugin_fl_zh/getting_started/getting-started.md‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -6,6 +6,7 @@
66
:maxdepth: 2
77
88
requirements.md
9+
install.md
910
run-inference-task.md
1011
1112
```

‎docs/sglang_plugin_fl_zh/getting_started/install.md‎

Lines changed: 28 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,28 @@
1+
# 安装 sglang-plugin-FL
2+
3+
## Docker 镜像（推荐）
4+
5+
v0.1.0-rc2 预构建 Docker 镜像：
6+
7+
| 平台 | 镜像 | 内容 |
8+
|----------|-------|----------|
9+
| NVIDIA GPU（双节点） | `harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-nvidia-dual` | sglang 0.5.11, flag_gems 5.3.0rc2, torch 2.11.0, triton 3.6.0 |
10+
| NVIDIA GPU（单节点） | `harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-nvidia-single` | sglang 0.5.11, flag_gems 5.3.0rc2, torch 2.11.0, triton 3.6.0 |
11+
| 摩尔线程 MUSA | `harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-musa` | sglang 0.5.12, torch 2.9.0, flag_gems 5.0.2 |
12+
| 摩尔线程（SVT） | `harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-mthreads-svt` | sglang 0.5.11, flag_gems 5.3.0rc2, torch 2.9.0, triton 3.1.0 |
13+
| 华为昇腾 | `harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-ascend` | sglang 0.5.12, flag_gems 5.0.2, CANN 8.5.0 |
14+
15+
```bash
16+
# NVIDIA 双节点（跨节点推理）
17+
docker pull harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-nvidia-dual
18+
# NVIDIA 单节点
19+
docker pull harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-nvidia-single
20+
# 摩尔线程 MUSA
21+
docker pull harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-musa
22+
# 摩尔线程 SVT（全栈测试）
23+
docker pull harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-mthreads-svt
24+
# 华为昇腾
25+
docker pull harbor.baai.ac.cn/flagos21-release/sglang-plugin-fl:v0.1.0-rc2-ascend
26+
```
27+
28+
双节点镜像支持跨节点大模型推理，单节点镜像支持单机推理。SVT 为全栈测试镜像。

‎docs/vllm_plugin_fl_en/getting_started/install.md‎

Lines changed: 17 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -130,6 +130,23 @@ export USE_FLAGGEMS=0
130130

131131
This section covers running vllm-plugin-FL using pre-built Docker images.
132132

133+
### SVT Full-Stack Test Images (v0.2.0-rc2)
134+
135+
Pre-built SVT images with full FlagOS stack:
136+
137+
| Platform | Image | Contents |
138+
|----------|-------|----------|
139+
| NVIDIA GPU | `harbor.baai.ac.cn/flagos21-release/vllm-plugin-fl:v0.2.0-rc2-nvidia-svt` | vllm 0.20.2, FlagGems 5.3.0-rc2.post1, FlagTree 3.6.0, vllm-plugin-FL 0.2.0-rc2.post1, torch 2.11.0+cu130 |
140+
| Hygon DCU | `harbor.baai.ac.cn/flagos21-release/vllm-plugin-fl:v0.2.0-rc2-hygon-svt` | vllm 0.20.0, FlagGems 5.3.0-rc2.post1, FlagTree 0.5.0-rc2.post1+hcu, vllm-plugin-FL 0.2.0-rc2.post1, torch 2.10.0+das |
141+
142+
```bash
143+
# NVIDIA SVT
144+
docker pull harbor.baai.ac.cn/flagos21-release/vllm-plugin-fl:v0.2.0-rc2-nvidia-svt
145+
146+
# Hygon DCU SVT
147+
docker pull harbor.baai.ac.cn/flagos21-release/vllm-plugin-fl:v0.2.0-rc2-hygon-svt
148+
```
149+
133150
### Hygon DCU
134151

135152
Available for vllm-plugin-FL v0.2.0 (vLLM 0.20.0).

0 commit comments

Comments
 (0)