feat: add support for HF_HUB_USER_AGENT_ORIGIN to add user-agent Origin field in Hub requests.

[Hugoch]

What does this PR do?

Adds support for HF_HUB_USER_AGENT_ORIGIN to add user-agent Origin field in Hub requests.
Useful to track collabs.

[angt]

You can use rust-toolchain.toml:

COPY rust-toolchain.toml rust-toolchain.toml
RUN curl -sSf https://sh.rustup.rs | sh -s -- -y --no-modify-path --default-toolchain none

[Narsil]

We can also not pipe anything into sh while we're at it, but there are unfortunately a lot of these occurrences.

[Hugoch]

I'm not sure of what would be the best way to avoid the pipe. rustup-init checks the architecture to download the right installer which we need to do. We could copy rustup-init in the repo and add a check for rust-init signature for each architecture but needs to be updated for each update. Do you have other ideas in mind?

[Narsil]

That's already alright to checksum the installer (although the installer of the installer should probably have the same treatment.).

There's the simple "we need deterministic builds" aspect. And then there's the "we don't want supply chain attacks" side of things. checksum goes a long way.

Ideal is a key word here.

[dacorvo]

You can use rust-toolchain.toml:

COPY rust-toolchain.toml rust-toolchain.toml
RUN curl -sSf https://sh.rustup.rs | sh -s -- -y --no-modify-path --default-toolchain none

This command does not install any toolchain, which was precisely the whole point of the previous command. Am i missing something ?

[angt]

Yes but the correct toolchain (from rust-toolchain.toml) will be installed with cargo-chef.
At least it works here:

text-generation-inference/Dockerfile_llamacpp

Lines 38 to 41 in d8ff7f2

	COPY rust-toolchain.toml rust-toolchain.toml
	RUN curl -sSf https://sh.rustup.rs | sh -s -- -y --no-modify-path --default-toolchain none
	ENV PATH="/root/.cargo/bin:$PATH"
	RUN cargo install cargo-chef --locked

[dacorvo]

I still don't understand how it could work: the code snippet in llamacpp is different than the one you suggested above for neuron.
This one installs explicitly the 1.85.0 toolchain, instead of specifying none, which doesn't install any toolchain.

[angt]

It was working, the correct toolchain was automatically installed at this step:

RUN cargo install cargo-chef --locked 

But I confirm it is not the case anymore.

[Narsil]

LGTM

[dacorvo]

This pull-request breaks the neuron build: was it tested ?

[Hugoch]

@dacorvo Yes, it builds on https://github.com/huggingface/text-generation-inference/actions/runs/13627605158/job/38160431550
Seems that main now refers to 1.85.0 which fails

[dacorvo]

@dacorvo Yes, it builds on https://github.com/huggingface/text-generation-inference/actions/runs/13627605158/job/38160431550 Seems that main now refers to 1.85.0 which fails

The issue is not related to 1.85.0: the build you are mentioning was before the review.
Here is the failing build with the changes introduced after the review: https://github.com/huggingface/text-generation-inference/actions/runs/13657446044/job/38180115060