Documentation

chisel lets you profile your inference workloads on cloud GPUs without any infrastructure setup.

Quick Start

Get GPU acceleration in 3 steps:

1. Install

   pip install chisel-cli
   

2. Create your script

   my_script.py

   from chisel import ChiselApp, GPUType
   
   # here we're choosing 1xA100-80GB
   # go to /configuration to learn about GPU types
   app = ChiselApp("my-app", gpu=GPUType.A100_80GB_1)
   
   @app.capture_trace(trace_name="gpu_task")
   def gpu_function():
       import torch
       device = "cuda" if torch.cuda.is_available() else "cpu"
       
       # Your GPU code here
       x = torch.randn(1000, 1000, device=device)
       y = torch.randn(1000, 1000, device=device)
       return torch.mm(x, y).cpu().numpy()
   
   if __name__ == "__main__":
       result = gpu_function()
       print(f"Result shape: {result.shape}")
   

3. Run on GPU

   # Local execution (CPU)
   # executes locally, not using chisel
   python my_script.py
   
   # Cloud GPU execution
   # executes on cloud GPU, using chisel
   chisel python my_script.py
   

   First run opens browser for authentication, then your code runs on A100 GPUs automatically.

GPU Options

Type	GPUs	Memory	Best For
GPUType.A100_80GB_1	1x A100	80GB	Development, inference
GPUType.A100_80GB_2	2x A100	160GB	Medium training
GPUType.A100_80GB_4	4x A100	320GB	Large models
GPUType.A100_80GB_8	8x A100	640GB	Massive models

Documentation

Complete Guide

Detailed setup and examples

API Reference

Complete API documentation

Examples

PyTorch, data processing, multi-GPU

Troubleshooting

Common issues and solutions

Support

• 📧 Email: [e-mail]
• 🐛 Issues: GitHub Issues
• 📖 Source: GitHub Repository