Chisel CLI

Accelerate your Python functions with cloud GPUs using a simple decorator.

Quick Start

1. Install Chisel CLI:

pip install chisel-cli

2. Create your script:

from chisel import capture_trace

@capture_trace(trace_name="matrix_ops")
def matrix_multiply(size=1000):
    import torch
    device = "cuda" if torch.cuda.is_available() else "cpu"
    
    a = torch.randn(size, size, device=device)
    b = torch.randn(size, size, device=device)
    result = torch.mm(a, b)
    
    return result.cpu().numpy()

if __name__ == "__main__":
    result = matrix_multiply(2000)
    print(f"Result shape: {result.shape}")

3. Run on cloud GPU:

# Local execution
python my_script.py

# Cloud GPU execution
chisel python my_script.py

4. Interactive setup:

• Chisel CLI prompts for app name, upload directory, requirements file, and GPU configuration
• First-time authentication opens browser automatically
• Real-time job status and output streaming

Features

• Simple decorator: Just add @capture_trace() to your functions
• Local & cloud: Same code runs locally or on cloud GPUs
• Interactive CLI: Guided setup for job configuration
• Real-time streaming: Live output and job status
• GPU profiling: Automatic performance traces and memory analysis
• Multi-GPU support: Scale from 1 to 8x A100-80GB GPUs

Usage

Basic Example

from chisel import capture_trace

@capture_trace(trace_name="my_function")
def my_gpu_function():
    import torch
    device = "cuda" if torch.cuda.is_available() else "cpu"
    return torch.randn(1000, 1000, device=device)

# Runs locally with: python script.py
# Runs on GPU with: chisel python script.py

Multiple Functions

@capture_trace(trace_name="preprocess")
def preprocess(data): 
    # Data preprocessing
    pass

@capture_trace(trace_name="train")  
def train(data): 
    # Model training
    pass

@capture_trace(trace_name="evaluate")
def evaluate(data): 
    # Model evaluation
    pass

Command Line Arguments

# Pass arguments to your script
chisel python train.py --epochs 100 --batch-size 32

# Or configure the job directly
chisel python train.py --app-name "training-job" --gpu 4

GPU Options

When prompted by the CLI, choose from:

Documentation

• Getting Started - Installation and first steps
• API Reference - Complete function reference
• Examples - Working code examples
• Configuration - Setup and optimization
• Troubleshooting - Common issues and solutions

Examples

See the examples directory for working code:

• Basic usage - Matrix operations
• Command line args - Script with arguments
• Deep learning - PyTorch training
• Multi-GPU - Parallel processing

Authentication

Chisel CLI handles authentication automatically:

# First time: Browser opens for authentication
chisel python my_script.py

# Logout to clear credentials
chisel --logout

Credentials are stored securely in ~/.chisel/credentials.json.

Development

# Clone repository
git clone https://github.com/Herdora/chisel.git
cd chisel

# Install in development mode
pip install -e .

# Run tests
pytest

# Check code style
ruff check src/ examples/

Contributing

We welcome contributions! Please see Development Guide for details.

Support

• 📧 Email: contact@herdora.com
• 🐛 Issues: GitHub Issues
• 💬 Discussions: GitHub Discussions

License

MIT License - see LICENSE for details.