search

dusty-nv / jetson-containers

Machine Learning Containers for NVIDIA Jetson and JetPack-L4T
MIT License
2.18k stars 446 forks source link

Packages need to be smarter about building their builders #448

Open rgobbel opened 6 months ago

rgobbel commented 6 months ago

I've been trying to build a version of local_llm that can load a Mixtral model (a separate issue), and I've noticed that a few of the subpackages fail with a"repository name must be lowercase" error, which really just means that there was a call with BUILD_IMAGE set to None. This happens when there is a Dockerfile.builder, but find_container returns None. For the cases I've tried, the problem disappears if I first build the builder by hand, so it looks like the general fix would be to always build the builder, if a package has one, but the container is not found.

dusty-nv commented 6 months ago

Sorry Randy, yes, what builders are missing?

I plan to refactor the whole builder thing and instead have the builders push the wheels to a custom pip pyindex server that I'm gonna run, which should alleviate this issue wholesale. Plus then ppl will just be able to install my wheels from this custom pyindex without needing to use my containers at all.

From: Randy Gobbel @.> Sent: Friday, March 22, 2024 12:23:28 PM To: dusty-nv/jetson-containers @.> Cc: Subscribed @.***> Subject: [dusty-nv/jetson-containers] Packages need to be smarter about building their builders (Issue #448)

I've been trying to build a version of local_llm that can load a Mixtral model (a separate issue), and I've noticed that a few of the subpackages fail with a"repository name must be lowercase" error, which really just means that there was a call with BUILD_IMAGE set to None. This happens when there is a Dockerfile.builder, but find_container returns None. For the cases I've tried, the problem disappears if I first build the builder by hand, so it looks like the general fix would be to always build the builder, if a package has one, but the container is not found.

— Reply to this email directly, view it on GitHubhttps://github.com/dusty-nv/jetson-containers/issues/448, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ADVEGKYJ2QVPILMXDMTGJY3YZRLIBAVCNFSM6AAAAABFDTX6D2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGIYDEOBWGI4DMMY. You are receiving this because you are subscribed to this thread.Message ID: @.***>

rgobbel commented 6 months ago

I'm not sure of the complete set, but cuda-pytorch and mlc were definitely failing. There are only a few packages in the entire collection that have these: cuda-python, mlc, onnxruntime, pytorch, and faiss. I see that you've made a more recent mlc image that knows about Mixtral models, and that's what I'm really interested in, so I'm not going to go through the whole list, but there it is.

dusty-nv commented 6 months ago

Ok thanks, what is your L4T version?

From: Randy Gobbel @.> Sent: Friday, March 22, 2024 12:31:09 PM To: dusty-nv/jetson-containers @.> Cc: Dustin Franklin @.>; Comment @.> Subject: Re: [dusty-nv/jetson-containers] Packages need to be smarter about building their builders (Issue #448)

I'm not sure of the complete set, but cuda-pytorch and mlc were definitely failing. There are only a few packages in the entire collection that have these: cuda-python, mlc, onnxruntime, pytorch, and faiss. I see that you've made a more recent mlc image that knows about Mixtral models, and that's what I'm really interested in, so I'm not going to go through the whole list, but there it is.

— Reply to this email directly, view it on GitHubhttps://github.com/dusty-nv/jetson-containers/issues/448#issuecomment-2015464244, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ADVEGK55MS4332T57IDY5E3YZRME3AVCNFSM6AAAAABFDTX6D2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMJVGQ3DIMRUGQ. You are receiving this because you commented.Message ID: @.***>

rgobbel commented 6 months ago

I'm running JP 6 (36.2.0)