[request for feedback] faster downloads with xet
Llama 4 Maverick and Scout are the first major models on Hugging Face uploaded with Xet, to make the upload and download significantly faster (more info here: https://huggingface.co/docs/huggingface_hub/en/guides/download#faster-downloads)
Let us know if you have any feedback!
My download throughput is not that high but I got the max I get get: between 30 MB and 40 MB of download speed.
My network saturated and PC had some freezes due to CPU peaks, even though my PC has decent specs Firefox and some other web-related tools were unusable. It would be great to add a hugginhuggingface-cli download --max-download-speed-mbs 30 option for example to avoid monopolizing the network 🤔
I used hugginhuggingface-cli download meta-llama/Llama-4-Maverick-17B-128E-Instruct, everything went smoothly until I got this error:
{"timestamp":"2025-04-11T08:29:46.422830Z","level":"ERROR","fields":{"message":"get from s3 error code, error: reqwest::Error { kind: Status(403), url: \"https://transfer.xethub.hf.co/xorbs/default/<redacted>?X-Xet-Signed-Range=bytes%<redacted>&Expires=<redacted&Policy=<redacted>&Signature=<redacted>&Key-Pair-Id=<redacted>\" }","caller":"/home/runner/work/xet-core/xet-core/cas_client/src/remote_client.rs:615"},"filename":"/home/runner/work/xet-core/xet-core/error_printer/src/lib.rs","line_number":28}
{"timestamp":"2025-04-11T08:29:46.584669Z","level":"ERROR","fields":{"message":"error fetching 1 term, error: Other(\"single flight error: Real call failed: ReqwestError(reqwest::Error { kind: Status(403), url: \\\"https://transfer.xethub.hf.co/xorbs/default/<redacted>?X-Xet-Signed-Range=bytes%<redacted>&Expires=<redacted>&Policy=<redacted>&Signature=<redacted>&Key-Pair-Id=<redacted>\\\" })\")","caller":"/home/runner/work/xet-core/xet-core/cas_client/src/remote_client.rs:481"},"filename":"/home/runner/work/xet-core/xet-core/error_printer/src/lib.rs","line_number":28}
Which is strange, I am authenticated so I should not encounter a 403 error... then the download resumed successfully.
Then my box crashed, but after using hugginhuggingface-cli download meta-llama/Llama-4-Maverick-17B-128E-Instruct again the download successfully resumed from the snapshot.
Overall, a pretty good experience of retry-on-error given my network issues, I dug a bit in the issue but seems like the fixes are already ongoing.
Edit: I'm sorry to keep editing this comment but it seems like the connection issues keep happening, and my computer freezes more and more, and those freezes often break the connection with the server. I did not have such issues with git LFS, even though it was 4x slower. I will get back to git clone for now...
{"timestamp":"2025-04-11T10:33:33.653598Z","level":"ERROR","fields":{"message":"error fetching 1 term, error: Other(\"single flight error: Real call failed: CasObjectError(InternalIOError(Custom { kind: Other, error: reqwest::Error { kind: Decode, source: hyper::Error(Body, Os { code: 104, kind: ConnectionReset, message: \\\"Connection reset by peer\\\" }) } }))\")","caller":"/home/runner/work/xet-core/xet-core/cas_client/src/remote_client.rs:481"},"filename":"/home/runner/work/xet-core/xet-core/error_printer/src/lib.rs","line_number":28}
cc @rajatarya maybe
@BaptisteRoseau Can you share more about your computer specifications (OS, RAM, type of hard disk, network)? Where are you downloading from? I can go see if I can correlate the errors you are seeing with the CDN logs from the local POP to your location.
@rajatarya
CPU: i5 12700K
RAM: 128GB DDR4 3600MHz
Ubuntu 22.04
2T NVME
Downloading from an Orange 5G Flybox
You can see the timestamps of the errors in my previous post to see when I was downloading 👍