How to Check MLflow Runs Locally

How to Check MLflow Runs Locally | Generated by AI

Home 2026.06

Question: How can I check MLflow results for a NanoChat training run like the one in the command, similar to how you can view runs in Weights & Biases?

Answer:

MLflow works differently from W&B — it’s not automatically a hosted web dashboard unless you run the MLflow server. Here’s how you can check your run:

1. Check the default MLflow location

By default, MLflow logs runs locally under the mlruns directory in your current working directory (/root/nanochat/mlruns in your case). You can explore it:

cd /root/nanochat/mlruns
ls

You’ll see directories for experiments (0, 1, …), each containing runs with UUID-like names. Inside each run folder, you’ll find:

meta.yaml → basic metadata
params/ → logged parameters
metrics/ → logged metrics
artifacts/ → saved models, outputs

2. Run the MLflow UI locally

MLflow has a built-in web UI. To view your experiments in a browser:

cd /root/nanochat
mlflow ui

By default, it will start at http://127.0.0.1:5000. If you are running this on a remote server, you may need to port-forward:

ssh -L 5000:127.0.0.1:5000 user@remote-server

Then open http://127.0.0.1:5000 in your local browser.

3. Check the run programmatically

You can also inspect metrics and parameters using MLflow’s Python API:

import mlflow

# List all experiments
experiments = mlflow.list_experiments()
for e in experiments:
    print(e.name, e.experiment_id)

# List runs of an experiment
runs = mlflow.list_run_infos(experiment_id="0")  # replace with your experiment id
for r in runs:
    print(r.run_id, r.status)

4. Tips for NanoChat integration

Make sure --tracker=mlflow in your command actually points to the MLflow backend you want (default is local mlruns).
If you want a more W&B-like online dashboard, you can also set up MLflow to log to a remote server or cloud storage.

In short: MLflow is mainly local unless you start its web UI, unlike W&B which is cloud-hosted by default.

You can essentially do mlflow ui in /root/nanochat and open the web page to inspect metrics, parameters, and artifacts.

If you want, I can give you a one-liner to open MLflow UI directly showing just your test-mi300x run, so you don’t have to browse all experiments.

Do you want me to do that?

Back Donate