Hippius Hub — Drop-in for huggingface_hub

The model registry that swaps in with one import.

A Rust-powered Python client + CLI for an OCI registry at registry.hippius.com. Change one import line and your huggingface_hub code keeps working — pull, push, search, no rewrites.

$ pip install hippius_hub

Get the package View on GitHub

HIPPIUS HUB

Built for ML workflows. Familiar by design.

Keep using huggingface_hub idioms. Push to your own OCI registry. Pull faster with parallel Rust. Find what you need with a real model index.

Drop-in for huggingface_hub

Same import paths, same signatures, same exception types. `from hippius_hub import hf_hub_download` works where the HF call worked — transformers, diffusers, datasets read the cache unchanged.

OCI registry backend

Models live as OCI artifacts at registry.hippius.com — docker / oras / podman / hippius-hub all push and pull the same bytes. Multi-region edge cache (EU + US).

Rust parallel transport

PyO3 extension fans Range requests across parallel TCP. Cache-hit blob pulls peak at 780 MB/s — 4.4× faster than HuggingFace cold on the same hardware.

Searchable AI model index

Every push is parsed server-side (GGUF, safetensors, ONNX, Diffusers). Search by format, architecture, parameter count, quantization — no manual metadata.

DEVELOPER

From zero to a pushed model in 90 seconds.

Pick your interface. Same bytes, same model index, same docker creds across CLI and Python.

One-line drop-in swap

Change the import, keep the rest. The cache layout matches HuggingFace byte-for-byte so transformers/diffusers from_pretrained Just Works.

drop_in.py

01# Before:
02# from huggingface_hub import hf_hub_download, snapshot_download, upload_folder
03 
04from hippius_hub import hf_hub_download, snapshot_download, upload_folder, login
05 
06# Authenticate once — token at https://console.hippius.com/dashboard/settings
07login(token="hf_xxx")
08 
09# Pull one file
10path = hf_hub_download(
11    repo_id="myorg/qwen-7b",
12    filename="model.safetensors",
13    revision="v1",
14)
15 
16# Pull a whole revision in parallel
17local_dir = snapshot_download(
18    repo_id="myorg/qwen-7b",
19    revision="v1",
20    allow_patterns=["*.safetensors", "*.json"],
21    max_workers=8,
22)
23 
24# Push the same way you would to HF
25upload_folder(
26    folder_path="./checkpoints",
27    repo_id="myorg/qwen-7b",
28    revision="v1",
29)
30 
31# transformers/diffusers natively load from the Hippius cache
32import hippius_hub as huggingface_hub  # drop-in swap
33from transformers import AutoModel
34model = AutoModel.from_pretrained("myorg/qwen-7b")

PRICING

Plans for every model.

Prices in USD, billed monthly from your Hippius credits balance. Subscriptions live on-chain through the Marketplace pallet — cancel any time, 30-day grace before anything is deleted.

Free

Forever