middleman-ml 0.1.0

Python SDK and CLI for Middleman ML Training Platform

Middleman Python SDK

Python SDK and CLI for the Middleman ML Training Platform.

Installation

pip install middleman-ml

Quick Start

CLI Usage

Configure your API key:

middleman auth
# Enter your API key when prompted

Or set the environment variable:

export MIDDLEMAN_API_KEY=mdlm_your_api_key_here

Create a training job:

middleman jobs create /scripts/train.py /data/dataset \
  --name "My Training Run" \
  --gpu a100 \
  --gpu-count 2 \
  --framework pytorch \
  --max-hours 8

Monitor your job:

middleman jobs watch <job-id>

Check your balance:

middleman billing balance

Python SDK Usage

from middleman import MiddlemanClient

# Initialize client
client = MiddlemanClient(api_key="mdlm_...")

# Create a job
job = client.create_job(
    name="ResNet Training",
    script_path="/scripts/train.py",
    input_data_path="/data/imagenet",
    gpu_type="a100",
    gpu_count=4,
    framework="pytorch",
    max_runtime_hours=24,
)

print(f"Job {job.id} created, queue position: {job.queue_position}")

# Wait for completion
result = client.wait_for_job(job.id)
print(f"Job finished with status: {result.status}")

# Check balance
balance = client.get_balance()
print(f"Available credits: {balance.available}")

CLI Commands

Authentication

middleman auth [API_KEY]     # Configure API key
middleman whoami             # Show account info

Jobs

middleman jobs list          # List jobs
middleman jobs create        # Create a job
middleman jobs status <id>   # Get job details
middleman jobs cancel <id>   # Cancel a job
middleman jobs pause <id>    # Pause a running job
middleman jobs resume <id>   # Resume a paused job
middleman jobs logs <id>     # View job logs
middleman jobs watch <id>    # Watch until completion

Billing

middleman billing balance    # Show credit balance
middleman billing packages   # Show credit packages
middleman billing history    # Transaction history

Data

middleman data upload <file>          # Upload a file
middleman data download-url <job-id>  # Get download URL

API Keys

middleman keys list           # List API keys
middleman keys create <name>  # Create new key
middleman keys revoke <id>    # Revoke a key

SDK Reference

MiddlemanClient

from middleman import MiddlemanClient

client = MiddlemanClient(
    api_key="mdlm_...",           # Or use MIDDLEMAN_API_KEY env var
    base_url="https://...",       # Optional, defaults to production
    timeout=30.0,                 # Request timeout in seconds
)

Jobs API

# List jobs
jobs, total = client.list_jobs(status="running", limit=20)

# Get job details
job = client.get_job("job-uuid")

# Create job
response = client.create_job(
    script_path="/scripts/train.py",
    input_data_path="/data/dataset",
    name="My Job",
    gpu_type="a100",      # t4, v100, a100
    gpu_count=1,          # 1-8
    framework="pytorch",  # pytorch, tensorflow, jax
    requirements_file="/scripts/requirements.txt",
    checkpoint_frequency=10,
    max_runtime_hours=4,
    environment_variables={"WANDB_API_KEY": "..."},
)

# Control jobs
client.cancel_job("job-uuid")
client.pause_job("job-uuid")
client.resume_job("job-uuid")

# Wait for completion
final_job = client.wait_for_job("job-uuid", timeout=3600)

# Stream status updates
for job in client.stream_job_status("job-uuid"):
    print(f"Epoch {job.current_epoch}, Loss: {job.current_loss}")
    if job.is_terminal:
        break

# Get logs
logs = client.get_job_logs("job-uuid")

Billing API

# Get balance
balance = client.get_balance()
print(f"Available: {balance.available} credits")

# List packages
packages = client.get_packages()

# Transaction history
transactions, total = client.get_transactions(limit=50)

Data API

# Upload a file
blob_path = client.upload_file("/local/path/data.zip", job_id="job-uuid")

# Get download URL
download = client.get_download_url("job-uuid", path="outputs/model.pt")
print(download.download_url)

API Keys

# List keys
keys = client.list_api_keys()

# Create key
key_info, full_key = client.create_api_key(
    name="CI/CD Pipeline",
    scopes=["read", "write"],
    expires_in_days=90,
)
print(f"Key: {full_key}")  # Only shown once!

# Revoke key
client.revoke_api_key("key-uuid")

GPU Types and Pricing

GPU Type	VRAM	Credits/Hour
T4	16GB	10
V100	32GB	25
A100	80GB	50

Environment Variables

• MIDDLEMAN_API_KEY - Your API key
• MIDDLEMAN_API_URL - Custom API URL (optional)

License

MIT