InstantMLDocs

Quickstart

Install InstantML, log your first run, and compare runs in the dashboard in five minutes.

Open .md

Log a simulated training run to hosted InstantML — no ML frameworks or datasets required — and see it charted in the dashboard. You need Python 3.11 or newer and an InstantML account.

Install the SDK

Install the package into your environment:

bash
python -m pip install instantml

Log in

Authenticate once from the terminal:

bash
instantml login

The CLI opens a browser device-code page, then stores an SDK credential in ~/.instantml/credentials. The Python SDK picks it up automatically. Confirm with instantml whoami.

Log your first run

Create train.py. It simulates a 50-step training loop with synthetic metrics, so it runs in seconds:

python
import math
import random
import sys

import instantml as im

seed = int(sys.argv[1]) if len(sys.argv) > 1 else 42
random.seed(seed)

run = im.init(
    project="quickstart",
    name=f"baseline-seed-{seed}",
    config={"seed": seed, "model": "tiny-mlp", "learning_rate": 3e-4},
    tags=["quickstart"],
)

for step in range(1, 51):
    loss = math.exp(-step / 18) + random.random() * 0.02
    accuracy = min(0.98, 0.5 + step / 120 + random.random() * 0.01)
    run.log({"train/loss": loss, "eval/accuracy": accuracy}, step=step)

run.finish()
print(f"Finished run {run.run_id}")

Run it:

bash
python train.py

You should see:

text
Finished run <run-id>

Open the dashboard

Open instantml.ai and select the quickstart project. Your run appears in the Runs workspace with live summary values, and the Run Detail view charts train/loss and eval/accuracy over all 50 steps.

Runs workspace showing the quickstart run with its logged metrics
Runs workspace showing the quickstart run with its logged metrics

Compare runs

The real payoff is comparison. Launch three more runs with different seeds:

bash
python train.py 7
python train.py 21
python train.py 99

Back in the Runs workspace, switch the Panels ⇄ Table toggle to Table and select all four runs. A side-by-side comparison renders above the table — metric values and config rows lined up, so the effect of seed is visible at a glance. See Compare runs for goal-aware deltas and reference runs.

Next steps