Skip to content

Commit b7ebd54

Browse files
nullnonenilNULLnullnonenilNULL
authored
Merge pull request baidu-baige#125 from Zachary-wW/docs/remove-enable-lerobot-flag
[docs] docs: remove ENABLE_LEROBOT flag, unify to single Docker image
2 parents c050975 + 7966083 commit b7ebd54

4 files changed

Lines changed: 25 additions & 101 deletions

File tree

‎docs/source/get_started/installation.md‎

Lines changed: 11 additions & 49 deletions
Original file line numberDiff line numberDiff line change
@@ -68,14 +68,18 @@ git clone --recurse-submodules https://github.com/baidu-baige/LoongForge.git
6868
Then build the image:
6969

7070
```bash
71-
docker build --build-arg COMPILE_ENV=hopper --build-arg ENABLE_LEROBOT=false \
71+
docker build --build-arg COMPILE_ENV=hopper \
7272
-t loongforge:latest -f ./LoongForge/docker/Dockerfile .
7373
```
7474

75+
> **Note:** All model families (LLM / VLM / VLA / Diffusion) now share a single
76+
> Docker image. The `ENABLE_LEROBOT` build argument and the separate `_lerobot`
77+
> image variant have been removed — there is no longer a lerobot / non-lerobot
78+
> distinction when building or pulling images.
79+
7580
| Build Arg | Description | Options |
7681
|---|---|---|
7782
| `COMPILE_ENV` | Target GPU architecture | `ampere`, `hopper`|
78-
| `ENABLE_LEROBOT` | Enable LeRobot dependencies for VLA model training (e.g., Pi0.5, GR00T). Disabled by default due to dependency conflicts with the base environment. | `true`, `false` |
7983

8084
After the build finishes, verify:
8185

@@ -91,23 +95,19 @@ LoongForge Docker images are available on Docker Hub:
9195
[https://hub.docker.com/u/loongforge](https://hub.docker.com/u/loongforge).
9296

9397
LoongForge publishes versioned pre-built Docker images. Select the desired tag
94-
from Docker Hub. LeRobot images use the same version tag with a `_lerobot` suffix
95-
when available.
98+
from Docker Hub. A single image covers all model families (LLM / VLM / VLA /
99+
Diffusion) — there is no longer a separate `_lerobot` variant.
96100

97101
| Image | Tag Pattern | Description |
98102
|---|---|---|
99-
| `loongforge/loongforge` | `<version>` | Base image: LLM / VLM / Diffusion training only |
100-
| `loongforge/loongforge` | `<version>_lerobot` | Includes LeRobot dependencies for VLA training (Pi0.5 + GR00T) |
103+
| `loongforge/loongforge` | `<version>` | Unified image: LLM / VLM / VLA / Diffusion training |
101104

102105
```bash
103106
# Set the version tag you want to use, for example: 0.1.1
104107
LOONGFORGE_VERSION=<version>
105108

106-
# Pull the base image
109+
# Pull the image
107110
docker pull loongforge/loongforge:${LOONGFORGE_VERSION}
108-
109-
# Pull the image with LeRobot support
110-
docker pull loongforge/loongforge:${LOONGFORGE_VERSION}_lerobot
111111
```
112112

113113
### Run the container
@@ -116,52 +116,14 @@ docker pull loongforge/loongforge:${LOONGFORGE_VERSION}_lerobot
116116
# Set the version tag you want to use, for example: 0.1.1
117117
LOONGFORGE_VERSION=<version>
118118

119-
# Using the base image (LLM/VLM/Diffusion)
120119
docker run --runtime=nvidia --gpus all -itd --rm \
121120
-v /path/to/your/hf/models:/mnt/cluster/huggingface.co/ \
122121
-v /path/to/data:/mnt/cluster/LoongForge/ \
123122
loongforge/loongforge:${LOONGFORGE_VERSION} /bin/bash
124-
125-
# Using the LeRobot image (VLA: Pi0.5 + GR00T)
126-
docker run --runtime=nvidia --gpus all -itd --rm \
127-
-v /path/to/your/hf/models:/mnt/cluster/huggingface.co/ \
128-
-v /path/to/data:/mnt/cluster/LoongForge/ \
129-
loongforge/loongforge:${LOONGFORGE_VERSION}_lerobot /bin/bash
130123
```
131124

132125
Once inside the container, navigate to `/workspace/LoongForge/examples/` and
133-
launch the desired training script. For GR00T training, remember to activate
134-
the GR00T virtual environment first (see Dual Python Environment below).
135-
136-
### Dual Python Environment in the LeRobot Image
137-
138-
The LeRobot image (`<version>_lerobot`) uses a **dual virtual-environment** setup to resolve
139-
dependency conflicts between Pi0.5 and GR00T:
140-
141-
| Environment | Path | Default? | Use Case |
142-
|---|---|---|---|
143-
| Base (Pi0.5) | System Python | Yes | Pi0.5 VLA training |
144-
| GR00T | `/opt/venvs/gr00t` | No | GR00T-N1.6 VLA training |
145-
146-
**To activate the GR00T environment inside the container:**
147-
148-
```bash
149-
source /opt/venvs/gr00t/bin/activate
150-
# or use the convenience alias:
151-
use-gr00t
152-
```
153-
154-
**To return to the base (Pi0.5) environment:**
155-
156-
```bash
157-
deactivate
158-
```
159-
160-
**For distributed training, use the venv's torchrun:**
161-
162-
```bash
163-
/opt/venvs/gr00t/bin/torchrun ${DISTRIBUTED_ARGS[@]} ...
164-
```
126+
launch the desired training script.
165127

166128
---
167129

‎docs/source/kunlun_tutorial/install_p800.md‎

Lines changed: 5 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -28,22 +28,24 @@ Then build the image:
2828

2929
```bash
3030
BASE_IMAGE=loongforge/loongforge_kunlun:py310_torch25
31-
ENABLE_LEROBOT=false
3231
DEFAULT_XPYTORCH_URL_ARG=https://baidu-kunlun-public.su.bcebos.com/baidu-kunlun-share/20260409/torch25/xpytorch-cp310-torch251-ubuntu2004-x64.run
3332

3433
docker build \
3534
--build-arg BASE_IMAGE=${BASE_IMAGE} \
36-
--build-arg ENABLE_LEROBOT=${ENABLE_LEROBOT} \
3735
--build-arg XPYTORCH_URL_ARG="${DEFAULT_XPYTORCH_URL_ARG}" \
3836
-t LoongForge-kunlun:latest -f LoongForge/docker/Dockerfile.xpu .
3937
# For internal conda image:
4038
#-t LoongForge-kunlun:latest -f LoongForge/docker/Dockerfile.xpu.internal .
4139
```
40+
41+
> **Note:** All model families (LLM / VLM / VLA / Diffusion) now share a single
42+
> Docker image. The `ENABLE_LEROBOT` build argument has been removed — there is
43+
> no longer a lerobot / non-lerobot distinction when building images.
44+
4245
- `BASE_IMAGE` is the base image used for building. Options include:
4346
* `loongforge/loongforge_kunlun:py310_torch25` (default) [available at Docker Hub]
4447
* `iregistry.baidu-int.com/xmlir/xmlir_ubuntu_2004_x86_64:v0.33` [internal use only]
4548
- `XPYTORCH_URL_ARG` is the xpytorch installer url argument.
46-
- `ENABLE_LEROBOT`: enable LeRobot dependencies for VLA model training (e.g., Pi0.5, GR00T). Disabled by default due to dependency conflicts with the base environment. Options: `true`, `false` (default).
4749
After building, you can verify the image:
4850

4951
```bash

‎docs/source_zh/get_started/installation.md‎

Lines changed: 6 additions & 46 deletions
Original file line numberDiff line numberDiff line change
@@ -60,14 +60,15 @@ git clone --recurse-submodules https://github.com/baidu-baige/LoongForge.git
6060
然后构建镜像：
6161

6262
```bash
63-
docker build --build-arg COMPILE_ENV=hopper --build-arg ENABLE_LEROBOT=false \
63+
docker build --build-arg COMPILE_ENV=hopper \
6464
-t loongforge:latest -f ./LoongForge/docker/Dockerfile .
6565
```
6666

67+
> **说明**：所有模型系列（LLM / VLM / VLA / Diffusion）现在共用同一个 Docker 镜像。`ENABLE_LEROBOT` 构建参数以及单独的 `_lerobot` 镜像变体已被移除 —— 构建或拉取镜像时不再区分 lerobot 与非 lerobot。
68+
6769
| 构建参数 | 描述 | 选项 |
6870
|---|---|---|
6971
| `COMPILE_ENV` | 目标 GPU 架构 | `ampere`, `hopper`|
70-
| `ENABLE_LEROBOT` | 启用 VLA 模型训练（如 Pi0.5, GR00T）的 LeRobot 依赖。由于与基础环境存在依赖冲突，默认禁用。 | `true`, `false` |
7172

7273
构建完成后，验证：
7374

@@ -83,22 +84,18 @@ LoongForge Docker 镜像保存在 Docker Hub：
8384
[https://hub.docker.com/u/loongforge](https://hub.docker.com/u/loongforge)。
8485

8586
LoongForge 发布带版本号的预构建 Docker 镜像。请在 Docker Hub 中选择需要的 tag。
86-
LeRobot 镜像在可用时使用相同版本号并追加 `_lerobot` 后缀。
87+
单个镜像覆盖所有模型系列（LLM / VLM / VLA / Diffusion），不再提供单独的 `_lerobot` 变体。
8788

8889
| 镜像 | 标签模式 | 描述 |
8990
|---|---|---|
90-
| `loongforge/loongforge` | `<version>` | 基础镜像：仅支持 LLM / VLM / Diffusion 训练 |
91-
| `loongforge/loongforge` | `<version>_lerobot` | 包含 VLA 训练（Pi0.5 + GR00T）的 LeRobot 依赖 |
91+
| `loongforge/loongforge` | `<version>` | 统一镜像：支持 LLM / VLM / VLA / Diffusion 训练 |
9292

9393
```bash
9494
# 设置需要使用的版本 tag，例如：0.1.1
9595
LOONGFORGE_VERSION=<version>
9696

97-
# 拉取基础镜像
97+
# 拉取镜像
9898
docker pull loongforge/loongforge:${LOONGFORGE_VERSION}
99-
100-
# 拉取带 LeRobot 支持的镜像
101-
docker pull loongforge/loongforge:${LOONGFORGE_VERSION}_lerobot
10299
```
103100

104101
### 运行容器
@@ -107,50 +104,13 @@ docker pull loongforge/loongforge:${LOONGFORGE_VERSION}_lerobot
107104
# 设置需要使用的版本 tag，例如：0.1.1
108105
LOONGFORGE_VERSION=<version>
109106

110-
# 使用基础镜像（LLM/VLM/Diffusion）
111107
docker run --runtime=nvidia --gpus all -itd --rm \
112108
-v /path/to/your/hf/models:/mnt/cluster/huggingface.co/ \
113109
-v /path/to/data:/mnt/cluster/LoongForge/ \
114110
loongforge/loongforge:${LOONGFORGE_VERSION} /bin/bash
115-
116-
# 使用 LeRobot 镜像（VLA: Pi0.5 + GR00T）
117-
docker run --runtime=nvidia --gpus all -itd --rm \
118-
-v /path/to/your/hf/models:/mnt/cluster/huggingface.co/ \
119-
-v /path/to/data:/mnt/cluster/LoongForge/ \
120-
loongforge/loongforge:${LOONGFORGE_VERSION}_lerobot /bin/bash
121111
```
122112

123113
进入容器后，导航到 `/workspace/LoongForge/examples/` 并启动所需的训练脚本。
124-
对于 GR00T 训练，请先激活 GR00T 虚拟环境（参见下方双 Python 环境说明）。
125-
126-
### LeRobot 镜像中的双 Python 环境
127-
128-
LeRobot 镜像（`<version>_lerobot`）使用**双虚拟环境**方案解决 Pi0.5 和 GR00T 之间的依赖冲突：
129-
130-
| 环境 | 路径 | 是否默认 | 用途 |
131-
|---|---|---|---|
132-
| 基础环境（Pi0.5） | 系统 Python | 是 | Pi0.5 VLA 训练 |
133-
| GR00T | `/opt/venvs/gr00t` | 否 | GR00T-N1.6 VLA 训练 |
134-
135-
**在容器中激活 GR00T 环境：**
136-
137-
```bash
138-
source /opt/venvs/gr00t/bin/activate
139-
# 或使用快捷命令：
140-
use-gr00t
141-
```
142-
143-
**返回基础（Pi0.5）环境：**
144-
145-
```bash
146-
deactivate
147-
```
148-
149-
**分布式训练请使用虚拟环境内的 torchrun：**
150-
151-
```bash
152-
/opt/venvs/gr00t/bin/torchrun ${DISTRIBUTED_ARGS[@]} ...
153-
```
154114

155115
---
156116

‎docs/source_zh/kunlun_tutorial/install_p800.md‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -28,22 +28,22 @@ git clone --recurse-submodules https://github.com/baidu-baige/LoongForge.git
2828

2929
```bash
3030
BASE_IMAGE=loongforge/loongforge_kunlun:py310_torch25
31-
ENABLE_LEROBOT=false
3231
DEFAULT_XPYTORCH_URL_ARG=https://baidu-kunlun-public.su.bcebos.com/baidu-kunlun-share/20260409/torch25/xpytorch-cp310-torch251-ubuntu2004-x64.run
3332

3433
docker build \
3534
--build-arg BASE_IMAGE=${BASE_IMAGE} \
36-
--build-arg ENABLE_LEROBOT=${ENABLE_LEROBOT} \
3735
--build-arg XPYTORCH_URL_ARG="${DEFAULT_XPYTORCH_URL_ARG}" \
3836
-t LoongForge-kunlun:latest -f LoongForge/docker/Dockerfile.xpu .
3937
# 内部 conda 镜像：
4038
#-t LoongForge-kunlun:latest -f LoongForge/docker/Dockerfile.xpu.internal .
4139
```
40+
41+
> **说明**：所有模型系列（LLM / VLM / VLA / Diffusion）现在共用同一个 Docker 镜像。`ENABLE_LEROBOT` 构建参数已被移除 —— 构建镜像时不再区分 lerobot 与非 lerobot。
42+
4243
- `BASE_IMAGE` 是用于构建的基础镜像。可选值包括：
4344
* `loongforge/loongforge_kunlun:py310_torch25`（默认）[Docker Hub 提供]
4445
* `iregistry.baidu-int.com/xmlir/xmlir_ubuntu_2004_x86_64:v0.33`[仅限内部使用]
4546
- `XPYTORCH_URL_ARG` 是 xpytorch 安装程序的 URL 参数。
46-
- `ENABLE_LEROBOT`：启用 VLA 模型训练（如 Pi0.5、GR00T）的 LeRobot 依赖。由于与基础环境存在依赖冲突，默认禁用。可选值：`true`、`false`（默认）。
4747

4848
构建完成后，可验证镜像：
4949

0 commit comments

Comments
 (0)