impulse-contracts-client 0.3.7

Client for interacting with the Impulse Labs protocol

Impulse Contracts Client

A Python client for interacting with the Impulse protocol, a decentralized computing platform for machine learning jobs.

Overview

The Impulse Contracts Client provides two main executables:

• impulse-node: For node operators to participate in the Impulse network
• impulse-client: For users to submit and monitor jobs on the network

Installation

Requirements

• Python 3.12
• An Ethereum wallet address
• Impulse tokens
• ETH for gas fees
• RPC endpoint for the blockchain
• python command available in your PATH and pointing to Python 3.12 (python --version should return 3.12.x)
• pip command available in your PATH and pointing to Python 3.12 (pip --version should point to python 3.12)

Installing via pip

pip install impulse-contracts-client
export PATH=$HOME/.local/bin:$PATH  # Add Python bin to PATH

Configuration

Both clients create a $HOME/.impulse directory for storing state:

• Node artifacts and configuration
• Client state and job information
• Logs and temporary files

Node Client (impulse-node)

Features

• Interacts with smart contracts for leader election, job assignment, and incentives
• Executes machine learning jobs upon assignment
• Automatically manages node's stake on the node escrow (i.e. top-up)
• Registers node to the Impulse network

Requirements

• A whitelisted Ethereum address
• Impulse tokens transferred to the address
• The impulse artifacts password
• Some ETH for gas costs
• Compatible GPU hardware

Supported GPU Hardware

• 1, 2, 4, or 8x NVIDIA A100 (40GB or 80GB)
• 8x NVIDIA H100 (80GB)

Running the Node

impulse-node

User Client (impulse-client)

Features

• Submits machine learning jobs to the Impulse network
• Monitors job execution and retrieves results
• Manages user's balance on the job escrow contract

Basic Usage

Top up your balance for paying for jobs:

impulse-client topup

Submitting Jobs

Example dummy job:

impulse-client create-job --args '{"shuffle": true, "use_lora": true, "use_qlora": false, "batch_size": 4, "dataset_id": "gs://imp-dev-pipeline-zen-datasets/0ca98b07-9366-4a31-8c83-569961c90294/2024-12-17_21-57-21_text2sql.jsonl", "num_epochs": 1, "job_config_name": "llm_dummy"}' --model llm_dummy --ft_type "LORA" --monitor

Example LLaMA 3.2 1B LoRA job:

impulse-client create-job --args '{"shuffle": true, "use_lora": true, "use_qlora": false, "batch_size": 4, "dataset_id": "gs://imp-dev-pipeline-zen-datasets/0ca98b07-9366-4a31-8c83-569961c90294/2024-12-17_21-57-21_text2sql.jsonl", "num_epochs": 1, "job_config_name": "llm_llama3_2_1b"}' --model llm_llama3_2_1b --ft_type "LORA"

Monitoring Jobs

Monitor all submitted jobs:

impulse-client monitor-all

Development

For development setup and contributing guidelines, see DEVEL.md.