allora-inference-base

Allora Inference Base

This node allows model providers to participate in the Allora network. This repo is the base for worker nodes (for providing inferences) and head nodes (which coordinates requests and aggregates results).

Build locally

In order to build locally:

GOOS=linux GOARCH=amd64 make

WARNING

This repo is currently relying on a private module, current development requires

export GOPRIVATE=github.com/allora-network/allora-appchain

Run locally

Head

./allora-node \
  --role=head  \
  --peer-db=/data/peerdb \
  --function-db=/data/function-db \
  --runtime-path=/app/runtime \
  --runtime-cli=bls-runtime \
  --workspace=/data/workspace \
  --private-key=/var/keys/priv.bin \
  --port=9010 \
  --rest-api=:6000 \
  --allora-chain-key-name=local-head \
  --allora-chain-restore-mnemonic='your mnemonic words...' --allora-node-rpc-address=https://some-allora-rpc-address/ \

Worker

./allora-node \
  --role=worker  \
  --peer-db=/data/peerdb \
  --function-db=/data/function-db \
  --runtime-path=/app/runtime \
  --runtime-cli=bls-runtime \
  --workspace=/data/workspace \
  --private-key=/var/keys/priv.bin \
  --port=9010 \
  --rest-api=:6000 \
  --topic=1 \
  --allora-chain-key-name=local-worker \
  --allora-chain-restore-mnemonic='your mnemonic words...' --allora-node-rpc-address=https://some-allora-rpc-address/ \
  --allora-chain-topic-id=1

Notes

If you plan to deploy without wanting to connect to the Allora blockchain, just by testing your setup and your inferences, do not set any --allora flag.

Topic registration

--topic defines the topic internally as a Blockless channel, so the heads are able to identify which workers can respond to requests on that topic.

--allora-chain-topic-id is the topic in which your worker registers on the appchain. This will be used for evaluating performance and allocating rewards.

Keys

The --private-key sets the Blockless peer key for your particular node. Obviously, please use different keys for different nodes. The --allora-chain-key-name and --allora-chain-restore-mnemonic are used to set the local Allora client keyring that your node will use when communicating with the Allora chain.

Blockless directories

Blockless nodes need to define a number of directories: --peer-db: database for peers in the network. --function-db: database for WASM functions to be executed in the workers. --runtime-path: runtime path to execute the functions --runtime-cli: runtime command to run functions --workspace: work directory where temporary files are stored.

Docker images

To build the image for the head:

docker build -f docker/Dockerfile_head -t allora-inference-base:dev-head --build-arg "GH_TOKEN=${YOUR_GH_TOKEN}" --build-arg "BLS_EXTENSION_VER=${BLS_EXTENSION_VERSION}" . 

Then to build the image for the head:

docker build -f docker/Dockerfile_worker -t allora-inference-base:dev-worker --build-arg "GH_TOKEN=${YOUR_GH_TOKEN}" --build-arg "BLS_EXTENSION_VER=${BLS_EXTENSION_VERSION}" . 

where YOUR_GH_TOKEN is your Github token, and optionally BLS_EXTENSION_VER is the release version of the Allora inference extension to use (it will use latest if none is set).

You can use the worker image as a base for building your own worker.

Debugging Locally Using VSCode.

This project comes with some static identities, as well as debug settings for VSCode. Use the VSCode debugger to start a head node instance, and a worker node instance.

• Ensure you've installed the Runtime make setup
• Start the Head Node First
• Start the Worker Node