umarise-huggingface 0.1.0

Anchor every HuggingFace upload to Bitcoin. Zero-touch provenance for models and datasets.

umarise-huggingface

Anchor every HuggingFace upload to Bitcoin. Automatically.

umarise-huggingface intercepts upload_file() and upload_folder() to automatically compute a local SHA-256 hash and anchor it to Bitcoin via the Umarise Core API. No files are transmitted — only the hash.

Install

pip install umarise-huggingface
export UMARISE_API_KEY=um_your_key_here

Usage

Option 1: Auto-anchor (recommended)

import umarise_huggingface
from huggingface_hub import HfApi

umarise_huggingface.enable()

api = HfApi()
api.upload_file(
    path_or_fileobj="model.safetensors",
    path_in_repo="model.safetensors",
    repo_id="your-org/your-model",
)
# [umarise] ✓ Anchored your-org/your-model:model.safetensors → origin_id: abc123...

Option 2: With transformers push_to_hub

import umarise_huggingface
umarise_huggingface.enable()

# Works automatically with any push_to_hub call
model.push_to_hub("your-org/your-model")
tokenizer.push_to_hub("your-org/your-model")
# Each file is automatically anchored

Option 3: Explicit

from umarise_huggingface import anchor_file, anchor_folder

# Single file
origin_id = anchor_file("model.safetensors")

# Entire directory
results = anchor_folder("./model_output/", pattern="*.safetensors")

What happens

1. You upload via HuggingFace Hub — upload proceeds normally
2. SHA-256 is computed locally — bytes never leave your machine
3. Hash is submitted to Umarise Core API → anchored to Bitcoin
4. origin_id is printed to stderr
5. Within ~2 hours, proof is confirmed on the Bitcoin blockchain

What this proves

• ✓ This exact model/dataset existed no later than time T
• ✓ The artifact was not modified after anchoring
• ✓ Anyone can verify — no Umarise account needed

Verify

# Get the hash of your file
shasum -a 256 model.safetensors

# Verify against Umarise
curl -s https://core.umarise.com/v1-core-verify \
  -H 'Content-Type: application/json' \
  -d '{"hash":"sha256:YOUR_HASH"}' | python3 -m json.tool

Or visit verify-anchoring.org

Design principles

• Never breaks your pipeline — if anchoring fails, uploads continue normally
• Zero storage — only the SHA-256 hash is sent to Umarise
• Vault-independent — HuggingFace stores the data, Umarise stores the proof
• Independently verifiable — no Umarise account needed to verify

Links

• Umarise
• API Reference
• Anchoring Specification
• umarise-wandb
• umarise-mlflow

License

Unlicense — public domain.