> ## Documentation Index
> Fetch the complete documentation index at: https://docs.neuronsearchlab.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Model Training

> Follow training runs, compare their results on your data, read their logs and stop runs you no longer need.

**Models → Model Training** lists every training run in your workspace, whether you started it from [Models](https://docs.neuronsearchlab.com/console/models.md), from [Events → Training settings](https://docs.neuronsearchlab.com/console/events.md#training-settings), or a [scheduled retrain](https://docs.neuronsearchlab.com/console/models.md#scheduled-retraining) started it for you.

## Recent jobs

The header counts total, completed, in-progress and failed runs. Each row shows the run, when it started and how long it took, and its status. **Refresh** updates the list.

Select a run to open its details:

| Section | Shows |
| - | - |
| **Training configuration** | The recommender, signal template and settings the run used |
| **Runtime details** | Duration, approval status, the target it was promoted to, and when |
| **Final metrics** | How the model scored on held-out data from your workspace (see below) |
| **Logs** | The run's log, streamed live while it is running. **Restart** replays the stream from the beginning |

**Stop job** cancels a run that is still in progress.

## Metrics

Runs are graded on data held back from training, so the scores reflect how the model would have done on real behaviour it had not seen.

| Metric | Meaning | Higher is |
| - | - | - |
| **NDCG\@10** | Whether the items people went on to engage with appear near the top of the first 10 | Better |
| **Recall\@20** | Share of the items people engaged with that appear in the first 20 | Better |
| **MRR\@10** | How high the first relevant item ranks, within the first 10 | Better |
| **Coverage\@20** | Share of the catalogue that appears in someone's first 20 | Broader |
| **Novelty** | How far recommendations go beyond the most popular items | More adventurous |
| **Diversity** | How different the recommended items are from each other | More varied |
| **Tail share** | Share of recommendations from less popular items | More long tail |
| **Training cost**, **Training time**, **Inference p95** | What the run cost, how long it took, and how fast it serves | Lower is better |

Each metric is marked **Good** or **Needs improvement** against typical values.

## Compare runs

Tick two or more completed runs to compare their metrics side by side. Runs trained in the same batch share a data snapshot and are directly comparable. Runs from different snapshots are flagged: treat their differences with caution, because the data they were graded on is not the same.

## What happens next

A finished run produces a model version in **Your models** on the [Models](https://docs.neuronsearchlab.com/console/models.md) page. From there you deploy it to a context, or let scheduled retraining promote it if it beats the live model.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.