Model Forge Quickstart

The Allora Model Forge Competition (opens in a new tab) is an open-source hackathon where participants attempt to create the best model for a given topic on the Allora Network.

Inferences are submitted to the Allora Network.
Participants are scored based on the accuracy of their model compared to the ground truth, and rewarded accordingly.

Getting Started

Creating a Wallet

To get started, you'll need to create a wallet on the Allora Network.

Download the `allorad` CLI Tool

Open your terminal and run the following command to install the allorad CLI tool.

curl -sSL https://raw.githubusercontent.com/allora-network/allora-chain/main/install.sh | bash -s -- v0.8.0

Create a Wallet

Run the following command to create a wallet.

allorad keys add <key-name>

Save your key name, wallet address, and mnemonic in a secure location. You will need this to submit inferences to the Allora Network.

Register for the Competition

To register for the competition, go to the Sign Up (opens in a new tab) page.

Fill out the form and submit your application.

💡

Input your wallet address that you created in the previous step in the Allora Wallet Address field.

⚠️

Although a discord account is not required to register, participants with a discord account will be able to join exclusive channels in the Allora Discord server to get priority access to help from the Allora Labs team and other participants.

Join the Allora Discord server here (opens in a new tab).

Successful Registration

Once you've been notified that you've been accepted into the competition:

Download the Keplr Wallet Chrome Extension (opens in a new tab) from the Chrome Web Store.
Connect your wallet to the Allora Forge site.

connect-wallet

You should see an eligible status in the top left corner of the screen once connected.

eligible

Congratulations! You're now registered for the Allora Model Forge Competition.

Topics

The Allora Network is categorized into distinct topics that serve as Schelling points for model-makers to submit inferences on.

Topics are assigned an ID and categorized by a specific Target Variable, Epoch, Loss Method, and Metadata. Let's quickly explain each in the context of the Forge Competition:

Topic ID: A unique identifier for the topic that participants will need to submit inferences on.
Target Variable: The asset or asset pair that models are trying to predict for
- E.g. ETH, USDC/ETH.
Epoch: A discrete period during which inferences and forecasts are submitted, and rewards are distributed.
- Epochs are defined in blocks on the Network
- Each epoch provides a timeframe for evaluating and scoring the performance of workers and reputers.
- In the Forge Competition, epochs are abstracted into universal timeframes, e.g. 5min, 1hr, 1d.
Loss Function: Measures how far off your model's predictions are from the actual values
- For all topics in the Forge Competition, the loss function is Mean Squared Error (opens in a new tab).
Metadata: Additional information about the topic, including the financial indicator, prediction period, and target variable.
- E.g. ETH 12hr Volatility Prediction [Epoch length: 5min]
  - Indicates that the topic is predicting the volatility of the ETH asset 12hr into the future.
  - Epoch length indicates that models are submitting the 12hr predicted volatility of the ETH asset every 5min.

💡

Additional fields that define a given topic can be pulled using the following allorad command:

allorad query emissions topic <topic-id> --node <node-url>

node-rpc is the RPC URL of the Allora Network node you are querying.

Example Usage:

allorad query emissions topic 13 --node https://allora-rpc.testnet.allora.network

Competition Topics

Topic ID	Topic	Target Variable	Epoch	Loss Function	Metadata
30	ETH Price Prediction	`ETH`	`5min`	Mean Squared Error	ETH 5min Price Prediction
29	USDC/ETH Volume Prediction	`USDC/ETH`	`5min`	Mean Squared Error	USDC/ETH 12hr Volume Prediction
28	ETH Volatility Prediction	`ETH`	`5min`	Mean Squared Error	ETH 12hr Volatility Prediction

Model Creation

If your registration is successful, start building your model and preparing it for inference submission to the network.

We've built out the Allora Model Development Kit (MDK) framework to help you get started with model creation. These models are optimized for price prediction, but can be used for other topics as well.

Allora MDK Repository (opens in a new tab)

Features:
- Large set of regression strategies to choose from
- Easy to customize for volatility and volume predictions

We offer a comprehensive set of documentation for the Allora MDK for you to dive into.

Leaderboard

The leaderboard is a live-updating list of the top performing models in the competition. Go to the competition page for the specific topic you've entered to view the leaderboard.

💡

Don't forget to connect your wallet to the Allora Forge site to view the leaderboard.

Detailed Performance Metrics

We've built out an Allora Explorer that displays performance metrics to help you evaluate the success of your model. These metrics are used to score your model in the competition.

Steps: