Skip to content

Commit 944845c

Browse files
yaoge123NJU Mirror
andauthored
jetson-pypi: add usage documentation (#62)
* jetson-pypi: add usage documentation * jetson-pypi: replace extra-index-url fallback advice with pinned-version install extra-index-url lets pip pick the best candidate across both indexes, so an unpinned torch could resolve to PyPI's non-Jetson build. Follow the official jetson-ai-lab / pytorch-TensorRT docs instead: install from the Jetson index with a pinned version. --------- Co-authored-by: NJU Mirror <mirror@nju.edu.cn>
1 parent f118080 commit 944845c

2 files changed

Lines changed: 49 additions & 0 deletions

File tree

‎jetson-pypi/jetson-pypi.zh.md‎

Lines changed: 31 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,31 @@
1+
本镜像收录了面向 NVIDIA Jetson 平台的 Python wheel 包(如 torch、torchvision、bitsandbytes、flash-attn 等针对 Jetson 预编译的版本),按 JetPack/CUDA 版本划分通道。
2+
3+
### pip
4+
5+
#### 临时使用
6+
7+
```{ztmpl lang="bash" input="channel"}
8+
pip install --index-url {{endpoint}}/{{channel}}/+simple some-package
9+
```
10+
11+
注意,`+simple` 不能少。
12+
13+
#### 设为默认
14+
15+
```{ztmpl lang="bash" input="channel"}
16+
pip config set global.index-url {{endpoint}}/{{channel}}/+simple
17+
```
18+
19+
镜像中未收录的普通 PyPI 包会跳转到本站 PyPI 镜像,因此单独使用本镜像即可安装全部依赖。
20+
21+
如果您的站点没有提供该跳转,未收录的普通 PyPI 包需要显式从 PyPI(镜像)安装。参考 jetson-pypi 官方教程的做法,此时建议锁定版本号,避免 pip 在两个索引间误选到 PyPI 上的非 Jetson 构建:
22+
23+
```{ztmpl lang="bash" input="channel"}
24+
pip install torch==2.8.0 --index-url {{endpoint}}/{{channel}}/+simple
25+
```
26+
27+
### Astral uv
28+
29+
```{ztmpl lang="bash" input="channel"}
30+
uv pip install --index-url {{endpoint}}/{{channel}}/+simple some-package
31+
```

‎jetson-pypi/zh.yaml‎

Lines changed: 18 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,18 @@
1+
_: NVIDIA Jetson PyPI 软件仓库
2+
block:
3+
- jetson-pypi
4+
filter:
5+
scheme: https
6+
input:
7+
channel:
8+
_: 通道
9+
default: jp6/cu126
10+
option:
11+
jp6/cu126:
12+
_: JetPack 6 / CUDA 12.6
13+
jp6/cu128:
14+
_: JetPack 6 / CUDA 12.8
15+
jp6/cu129:
16+
_: JetPack 6 / CUDA 12.9
17+
sbsa/cu130:
18+
_: SBSA / CUDA 13.0

0 commit comments

Comments
 (0)