Missing jp5/cu114 and jp4/cu102 indexer in pypi.jetson-ai-lab.io

[stylish-llama]

Search before asking

• I have searched the jetson-containers issues and found no similar feature requests.

Bug (?)

With the departure of dusty-nv (many thanks to them) and https://pypi.jetson-ai-lab.dev/jp5/cu114 unavailable, all container builds fail on JP4/5 systems.

Environment

┌───────────────────────┬────────────────────────┐
│ L4T_VERSION   35.6.2  │ JETPACK_VERSION  5.1.5 │
│ CUDA_VERSION  11.4    │ PYTHON_VERSION   3.8   │
│ SYSTEM_ARCH   aarch64 │ LSB_RELEASE      20.04 │
└───────────────────────┴────────────────────────┘

I managed to build a working image of ollama v0.9.6 and v0.10.0-rc0 by hard-coding pip_path = "root/pypi" here:

jetson-containers/packages/cuda/cuda/config.py

Lines 141 to 145 in 696a77f

	pip_path = (
	f"jp{JETPACK_VERSION.major}/{short_version}" if IS_TEGRA
	else f"sbsa/{short_version}" if IS_SBSA
	else f"{DOCKER_ARCH}/{short_version}"
	)

Solution (?)

I belive that, on older platforms that do not support JP6/7, multiple other images could be built without modifying the codebase, by just adding a basic indexers containing only the inherited bases root/pypi.

Additional

A list of previously available wheels is visible at the internet archive, sadly the actual resources are not available

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[Keval-P]

If you need, you can download the packages from here
=> https://drive.google.com/drive/folders/1macyG38SGCiazkiVfvQDahkAHgjhT6Mi