Register hailo library with .hef download counting#2223
Open
evandhoffman wants to merge 3 commits into
Open
Conversation
Adds a library entry for Hailo, mapping its proprietary HEF (Hailo Executable Format) files as the download-count query target. HEFs are the architecture-specific INT8 binaries that run on Hailo-8 and Hailo-15 AI accelerators; they're shipped alongside ONNX or PyTorch source weights but are the actual deployment artifact. Example model repos hosting HEFs: - k10z/birdvision-efficientnet-s - (Hailo's official Hailo Model Zoo publishes them outside HF today, but third-party projects are starting to host on the Hub.) The repoUrl points at the public hailort runtime which is the end-user dependency.
evandhoffman
requested review from
SBrandeis,
Wauplin,
gary149,
julien-c,
ngxson and
pcuenca
as code owners
Author
|
Updated to match the integration guide more precisely after reviewing it:
|
Open
4 tasks
Contributor
|
Hi @evandhoffman thanks for the PR and details. From what I can see in https://huggingface.co/models?other=hailo:
I feel that if hef is a file format we don't have a good way of counting them. If hailo is indeed a library, it's best to split these files to separate repos so we don't double-count or under-count download (i.e. have a .onnx repo and a .hef repo, both with there own counter). Alternative is to add both .hef and .onnx to download counter rule for hailo but that would be relevant only if hailo really support .onnx (which is not the case I think) |
Member
|
cc @danbev for viz btw |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a
hailolibrary entry tomodel-libraries.tsso that HEF (Hailo Executable Format) files hosted on the Hub get counted toward download stats.Why
Hailo makes the Hailo-8 / Hailo-15 PCIe AI accelerators (commonly seen as M.2 add-ons for the Raspberry Pi 5 AI Kit and other edge platforms). Models targeting these chips are compiled to
.hefbinaries via Hailo's Dataflow Compiler; the HEF is the actual deployment artifact, similar in role to a TensorRT engine or a CoreML.mlpackage.Hailo's own model zoo is hosted on GitHub today, but third-party projects are starting to publish HEFs on the Hub alongside their source weights — e.g.
k10z/birdvision-efficientnet-s(a fine-tuned EfficientNet-V2-S for bird species classification compiled to Hailo-8). These repos currently show no download counts because the Hub's default query files (config.json,config.yaml, etc.) aren't matched.What this does
Registers
hailowithcountDownloads: path_extension:"hef", matching the pattern used for other accelerator-binary libraries (e.g. HoloMotion usingpath_extension:"onnx").repoUrlpoints at the publichailortruntime — the end-user dependency that actually loads HEFs at inference time.Snippet is a minimal Hub-aware load pattern (
hf_hub_download+HEF+VDevice) with a pointer to thehailortrepo for the full inference API.Pre-merge checklist
habibi-ttsandhallofilter: falseper the integration guidelibrary_name: hailo:k10z/birdvision-efficientnet-s