HeptagonAI

Proof of learning AI traing that uses Filecoin for dataset storage, Phala TEE for proof verification and Flare Network for data collection and price feeds.

INFO: the project requires local backend

Traning data repository

Filecoin / IPFS

Upload files

Select files and click "Upload"
Estimated cost: ~0 USDC

Download model

Python

Download and run the model on your own machine, the training will generate Proof of Learning.

model.py

Download dataset

Composite dataset

Download the composite dataset made up of users contributions.

cifar10-composite.tar.gz

Train the model

First prepare python3.10 environment with the following libraries:

numpy==1.26.4
torch==1.11.0
scipy==1.10.0

Train the model by running the following command in the shell:

python train.py --save-freq=1000

Upload trained model

Python

Once the model is done training, upload it here to verify the training in TEE.

Upload proof of learning (PoL)

Proof of learning

The proof of learning (PoL) proves that the model was trained correctly on the dataset.

Verify proof of learning

We verify proof of learning inside Phala networks TEE

Tdx quote:

Tdx-quote

Calculate rewards

Flare data connector

We use Flare data connector (FDC) to evaluate data quality of attirbutions and calculate rewards.

Rewards

We use Flare data connector (FDC) to bring the quality of data and number of contributions on chain and Flare time series oracle (FTSO) for price feeds. Model trainer reward is 0.2 CFLR

	Wallet	Data quality	Contributions	Rewards