Metadata-Version: 2.4
Name: trailer-sdk
Version: 0.1.12
Requires-Dist: msgpack>=1.0
Requires-Dist: httpx>=0.27
Requires-Dist: scikit-learn>=1.0 ; extra == 'embedding'
Requires-Dist: psycopg2-binary>=2.9 ; extra == 'pg'
Provides-Extra: embedding
Provides-Extra: pg
License-File: LICENSE
Summary: Trailer: Next-gen experiment tracking — TensorBoard + Aim + W&B in Rust
Requires-Python: >=3.10
Description-Content-Type: text/markdown; charset=UTF-8; variant=GFM

[中文](README.zh-CN.md) | **English**

# Trailer

> Next-gen **ML experiment tracking** — a lightning-fast, open-source experiment tracker with a **high-performance Rust core** and a local-first philosophy.

The ultimate trailer for your deep learning experiments — combining TensorBoard's depth, Aim's fluidity, and W&B's ecosystem, powered by a Rust core.

## Features

- ⚡ **High-performance Rust core** — LTTB downsampling, batched ingestion, 100k-entry ring buffer (P99 < 100µs)
- 📊 **Every data type you need** — scalars, text, figures, tables, media, histograms, embeddings, PCA, and model graphs
- 🏔️ **Loss landscape visualization** — filter-normalized 2D loss surfaces as heatmap / contour / interactive 3D, with training-evolution playback, log/linear color scale and a gradient-descent ball animation
- 🚀 **One-command local mode** — `trailer up` starts the full dashboard — zero config, no HTTP overhead, data lands locally
- 🔬 **Cross-project Explore** — line, scatter, and parallel-coordinates charts across runs — saved and shareable
- 🗄️ **Flexible storage** — SQLite (default), PostgreSQL, or TensorBoard-style file mode
- 👥 **Multi-user & permissions** — admin / experimenter roles, anonymous sharing links
- 🔍 **Powerful Compare** — progressive-loading comparison across runs

## Screenshots

<img src="docs/public/ScreenShot_1.png" width="32%"/> <img src="docs/public/ScreenShot_2.png" width="32%"/> <img src="docs/public/ScreenShot_3.png" width="32%"/>

<img src="docs/public/ScreenShot_4.png" width="32%"/> <img src="docs/public/ScreenShot_5.png" width="32%"/> <img src="docs/public/ScreenShot_6.png" width="32%"/>

<img src="docs/public/ScreenShot_7.png" width="32%"/> <img src="docs/public/ScreenShot_8.png" width="32%"/> <img src="docs/public/ScreenShot_9.png" width="32%"/>

## Quick start

```bash
pip install trailer-sdk
```

```python
from trailer import Tracker

t = Tracker(project="my_experiment")
t.log({"train/loss": 0.5, "val/loss": 0.6})
t.finish()
```

Launch the dashboard:

```bash
trailer up          # http://127.0.0.1:5120
```

## Documentation

Full docs (English / 中文): **https://limoncc.github.io/trailer/**

## License

[Elastic License 2.0](LICENSE) — free for internal use; commercial hosting / SaaS requires a separate license. See [LICENSING.md](LICENSING.md).

