Bittensor Validator Endpoint

Prompt the Bittensor network via an API

Documentation

Allow validators to easily host an API for other services to prompt the Bittensor network.

Features

• Host an API server with API key authentication
• Web Dashboard for statistics, configuration and key management
• btvep CLI for configuration and key management
• API key lifetimes - set expiry dates
• API key credits - limit amount of requests with credits
• Request logs
• Rate limits - Configure default limits or custom per API key
• Drop-in replacement for OpenAI's Chat API
• Easily Filter requests via OpenAI's Moderation Endpoint
• 3 Query strategies
  • top_n - Query top n incentive miners. This allows for querying the whole network with one request, e.g. setting top_n to the number of uids on the subnet.
  • uids - Query specific uids.
  • default - When no strategy is specified in the request, query a default uid. This should normaly be the validator itself to utilize its own miner selection model for prompts.

Planned features

• A frontend for end users to query the backend, similar to chat.openai.com Users can on the same page also specify UIDs, and search through UIDs with descriptions/tags. Login system via email & password, Google and Github
• Adding options for payment for credits. Payment could for example be handled with TAO or an integration with Stripe for card payments.
• Add additional future subnetworks, ex for image and audio.
• LangChain integration

Getting started

# Install with pip
python3 -m pip install https://github.com/fabhed/validator-endpoint/raw/main/dist/btvep-0.1.1-py3-none-any.whl
# Set your hotkey mnemonic (quotes are needed since the mnemonic has spaces in it)
btvep config set hotkey_mnemonic "my_validators_secret_mnemonic_phrase_here"
# Create an API key
btvep key create
# Start the server
btvep start --port 8000

As an alternative to the above approach you can also use docker: Docker Guide

Example request

Make a request with the key you just created (Replace $API_KEY)

curl http://localhost:8000/chat \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_KEY" \
  -H "Endpoint-Version: 2023-05-19" \
  -d '{
     "messages": [{"role": "user", "content": "What is 1+1?"}]
   }'

Tutorials

• Colab Notebook Tutorial: A step-by-step guide on how to use our project in a Colab notebook.

Requirements

• Python 3.10
• If Rate limiting is used Redis is also required

API Usage Examples

Prompt the bittensor network

curl http://localhost:8000/chat \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_KEY" \
  -H "Endpoint-Version: 2023-05-19" \
  -d '{
     "messages": [{"role": "user", "content": "What is 1+1?"}]
   }'

Response:

{
  "choices": [
    {
      "index": 0,
      "responder_hotkey": "5ETyaEdDp2RQDoGazHzdGRmJUSzfrXCrMj5PyoFoskFdtsyH",
      "message": {
        "role": "assistant",
        "content": "this is a test!"
      }
    }
  ]
}

Prompt a specific neurons via uids on the network

curl http://localhost:8000/chat \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_KEY" \
  -H "Endpoint-Version: 2023-05-19" \
  -d '{
     "uids": [0, 1],
     "messages": [{"role": "user", "content": "What is 1+1?"}]
   }'

Response:

{
  "choices": [
    {
      "index": 0,
      "responder_hotkey": "5ETyaEdDp2RQDoGazHzdGRmJUSzfrXCrMj5PyoFoskFdtsyH",
      "message": {
        "role": "assistant",
        "content": "this is a test!"
      }
    }
  ]
}

Prompt top neurons sorted by incentive on the network

curl http://localhost:8000/chat \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_KEY" \
  -H "Endpoint-Version: 2023-05-19" \
  -d '{
     "top_n": 5,
     "messages": [{"role": "user", "content": "What is 1+1?"}]
   }'

This will send the prompt to the top 5 miners, change top_n to 1 to only query the top miner.

Dev requirements

• Poetry: https://python-poetry.org/docs/

Dev setup

# Clone the repo
git clone https://github.com/fabhed/validator-endpoint.git
cd validator-endpoint

# Install depedencies
poetry install

# Launch the shell enviornment
poetry shell

# Run your editor
code .

# You can now run the cli
btvep --help

# Run the server with auto reloading
btvep start --reload

Generate docs

poetry run typer btvep.cli utils docs --output docs/cli.md --name btvep

Build .whl files in /dist

poetry build

grpcio dependency (from bittensor) arm fix

python3 -m pip install --no-binary :all: grpcio --ignore-installed