Step-by-Step Instructions

1. System Update and Package Installation:
   • Update your system to ensure all packages are up to date.

sudo apt update && sudo apt upgrade -y
sudo apt install -y git python3 python3-pip python-is-python3

1. Clone the Repository:

git clone <https://github.com/NicheTensor/NicheImage>
cd NicheImage

1. Create and activate python virtual environment:

python -m venv main_env
source main_env/bin/activate

1. Install requirements:

pip install -e .
pip uninstall uvloop -y
git submodule update --init --recursive
. generation_models/custom_pipelines/scripts/download_antelopev2.sh

NOTE: You need to set --num_gpus 0 and --num_replicas equal to your max_concurrent_requests (defined in point 7 below) to allow the miner to handle multiple requests concurrently.

1. Start the endpoint

source main_env/bin/activate
OPENAI_API_KEY=<your-openai-api-key> \\ # set if you use DallE model
RAY_SERVE_QUEUE_LENGTH_RESPONSE_DEADLINE_S=1.0 \\
pm2 start python --name "miner_endpoint" -- -m services.miner_endpoint.app \\
--model_name <selected-model-name> \\
--num_replicas X --num_gpus 0 \\
--port 10006 # default port

1. Start the miner

pm2 start python --name "miner" \\
-- \\
-m neurons.miner.miner \\
--netuid 23 \\
--wallet.name <wallet_name> --wallet.hotkey <wallet_hotkey> \\
--subtensor.network <network> \\ # default is finney
--axon.port <your_public_port> \\
--generate_endpoint [<http://127.0.0.1:10006/generate>](<http://127.0.0.1:10006/generate>) \\ # change if you use different port or host
--info_endpoint [<http://127.0.0.1:10006/info>](<http://127.0.0.1:10006/info>) \\ # change if you use different port or host
--miner.total_volume <your-generation-volume> # default is 40. Change based on your model timeout value and GPU capacity
--miner.max_concurrent_requests <your-max-concurrent-requests> # default is 4. Change based on your model timeout value and GPU capacity

1. Important notes:
   • miner.total_volume <your-generation-volume> refers to the total number of requests your miner can process
     • This is calculated by the following formula: total_volume = total_available_gpus * 600_seconds / generation_time_per_image_or_text
       • Ex: 2 * 600 / 8 = 150
       • In this case the total volume would be 150 requests
     • Important: Because of the balance between speed_penalty and volume_reward, calculating miner total volume is highly system and situtation dependent.
   • miner.max_concurrent_requests <your-max-concurrent-requests>refers to the number of requests that can be processed in parallel.
     • This is calculated by the following formula: max_concurrent_requests = num_gpus * timeout_limit_for_model / generation_time_per_image_or_text
       • 2 * 12 / 8 = 3
       • In this case the total number of concurrent requests would be 3:
         • 2 GPUs, 12s timeout and 8s per image generation.