s2r 0.3.2

Convert SLURM submit scripts to Run.ai configurations using AI

s2r - SLURM to Run.ai Converter

Convert SLURM batch scripts to Run.ai configurations using AI.

Installation

pip install s2r

No AWS account or credentials required — the public hosted endpoint is rate-limited to 100 requests per IP per day. If you self-host and want to keep the Lambda behind IAM auth, install with the optional dependency:

pip install 's2r[iam-auth]'

Quick Start

CLI Usage

# Convert file: saves job.yaml, prints runai CLI command to stdout
s2r job.slurm

# Convert file: saves output.yaml only (no CLI output)
s2r job.slurm output.yaml

# Convert from stdin: prints runai CLI command to stdout (no file)
s2r < my_slurm_script.sh

Library Usage

from s2r import convert_slurm_to_runai

slurm_script = """
#!/bin/bash
#SBATCH --job-name=my-job
#SBATCH --gres=gpu:2
#SBATCH --mem=32G

python train.py
"""

runai_config = convert_slurm_to_runai(slurm_script)
print(runai_config)

How It Works

1. Client: The s2r library signs your SLURM script with HMAC-SHA256 (no AWS credentials needed)
2. API: Sends the signed request to an AWS API Gateway HTTP endpoint
3. AI: An AWS Lambda behind the API calls Bedrock (Claude Sonnet 4.6) to perform the conversion
4. Response: Returns the Run.ai YAML configuration or CLI commands

Features

• Free to use: The service is provided at no cost (rate-limited)
• Secure: Signed requests prevent unauthorized API usage
• Rate-limited: 100 requests per IP per day
• Simple: Works with stdin, files, or as a library

Configuration

By default, the tool uses a public API endpoint. If you're deploying your own:

export S2R_API_ENDPOINT=https://your-api-id.execute-api.us-west-2.amazonaws.com/

If you self-host with IAM auth on the endpoint, install the IAM extra and enable signing:

pip install 's2r[iam-auth]'
export S2R_USE_IAM_AUTH=true
export S2R_AWS_REGION=us-west-2

Example

Given this SLURM script:

#!/bin/bash
#SBATCH --job-name=pytorch-training
#SBATCH --nodes=1
#SBATCH --cpus-per-task=8
#SBATCH --mem=32G
#SBATCH --gres=gpu:2
#SBATCH --time=24:00:00

python train.py --epochs 100

The tool will generate an equivalent Run.ai configuration with:

• GPU resource requests (2 GPUs)
• CPU and memory allocations
• Job name and command
• Appropriate container/image specifications

Documentation

• API Reference: Complete API documentation for library and CLI
• Architecture: System design and component details
• Deployment Guide: AWS Lambda deployment instructions
• Troubleshooting: Common issues and solutions
• CLAUDE.md: Quick reference for Claude Code

Current Deployment Status

Deployed in us-west-2:

• API Gateway HTTP API → Lambda s2r-converter → Bedrock
• Model: Claude Sonnet 4.6 (us.anthropic.claude-sonnet-4-6 cross-region inference profile)
• Rate Limit: 100 requests/IP/day (DynamoDB-backed)
• Public endpoint, no AWS credentials required to use the client

Development

See CLAUDE.md for development commands and quick reference.

For detailed architecture and deployment information, see the docs/ directory.

Self-Hosting

To deploy your own instance with API Gateway + Lambda:

git clone https://github.com/dirkpetersen/slurm2runai.git
cd slurm2runai/lambda

# Deploy with SAM (creates API Gateway, Lambda, DynamoDB)
sam deploy --guided

# Point your client at the new endpoint
export S2R_API_ENDPOINT=https://<api-id>.execute-api.<region>.amazonaws.com/

See docs/deployment.md for full instructions.

License

MIT License - see LICENSE file for details.

Contributing

Contributions welcome! Please:

1. Open an issue to discuss changes
2. Follow the code style (ruff)
3. Add tests for new features
4. Update documentation