erickfm/slippi-public-dataset-v3.7

---
language:
- en
license: cc0-1.0
tags:
- melee
- smash-bros
- slippi
- replay-data
- fighting-games
pretty_name: "Slippi Public Dataset v3.7"
size_categories:
- 100K<n<1M
---

# Slippi Public Dataset v3.7

~95,000 raw Slippi replay files (`.slp`) from Super Smash Bros. Melee tournament sets, organized by character.

Originally compiled and shared by **altf4** on the [Slippi Discord](https://discord.gg/slippi) (October 2020), with contributions from nikki and yashichi. Mirrored here for convenient programmatic access.

> Public SLP Dataset v3 is ready!
> - 95,102 SLP files.
> - Unzips to 200GB.
> - All tournament sets, with varying skill levels.
> - Pruned to remove handwarmers, doubles, <30 second matches.
> - CC0 Licensed, so use it however you want
>
> — altf4, Slippi Discord (10/26/2020)

[Slippi](https://slippi.gg/) is a modified Melee netplay client that records frame-perfect game replays. Each `.slp` file contains the complete game state and controller inputs for every frame of a match at 60 fps.

| Stat | Value |
|------|-------|
| Unique replays | ~95,102 |
| Total files (with character duplication) | ~174,000 |
| Format | `.slp` (Slippi replay binary) |
| Size | ~222 GB |
| Source | Tournament sets (varying skill levels) |
| Filtering | Handwarmers, doubles, and <30s matches removed |
| Game | Super Smash Bros. Melee (NTSC) |
| Frame rate | 60 fps |
| Original source | [Google Drive (altf4)](https://drive.google.com/file/d/1VqRECRNL8Zy4BFQVIHvoVGtfjz4fi9KC/view) |

## Structure

Files are organized into **25 character directories**. Each replay appears under both characters in the match (e.g. a Fox vs Falco game appears in both `FOX/` and `FALCO/`).

```
FOX/
batch_00/ (9,000 files)
batch_01/ (9,000 files)
...
batch_05/ (854 files)
FALCO/
batch_00/ (9,000 files)
...
BOWSER/ (174 files, no batching needed)
PICHU/ (122 files)
...
```

Characters with >9,000 files are split into `batch_XX/` subdirectories. Characters with fewer files have all replays directly in the character folder.

### Characters

| Character | Files | Batched? |
|-----------|-------|----------|
| FOX | 45,854 | 6 batches |
| FALCO | 32,552 | 4 batches |
| MARTH | 26,085 | 3 batches |
| CPTFALCON | 22,794 | 3 batches |
| ZELDA_SHEIK | 13,764 | 2 batches |
| PEACH | 6,679 | No |
| JIGGLYPUFF | 4,998 | No |
| SAMUS | 3,640 | No |
| GANONDORF | 2,784 | No |
| ICE_CLIMBERS | 2,476 | No |
| LUIGI | 2,308 | No |
| YOSHI | 1,781 | No |
| PIKACHU | 1,769 | No |
| DOC | 1,506 | No |
| DK | 1,039 | No |
| LINK | 761 | No |
| MARIO | 719 | No |
| GAMEANDWATCH | 642 | No |
| YLINK | 411 | No |
| ROY | 402 | No |
| KIRBY | 345 | No |
| NESS | 319 | No |
| MEWTWO | 261 | No |
| BOWSER | 174 | No |
| PICHU | 122 | No |

## Usage

### Download all replays for a character

```python
from huggingface_hub import snapshot_download

snapshot_download(
"erickfm/slippi-public-dataset-v3.7",
repo_type="dataset",
local_dir="data/fox",
allow_patterns=["FOX/*"],
)
```

### Parse replays

Raw `.slp` files can be parsed with:
- [py-slippi](https://github.com/hohav/py-slippi) (Python)
- [slippi-js](https://github.com/project-slippi/slippi-js) (JavaScript)
- [peppi](https://github.com/hohav/peppi) (Rust)
- [libmelee](https://github.com/altf4/libmelee) (Python, used for character detection)

## Related

- [erickfm/melee-ranked-replays](https://huggingface.co/datasets/erickfm/melee-ranked-replays) — ~850k ranked ladder replays (platinum+), sharded by character and rank
- [Slippi](https://slippi.gg/) — Melee netplay client

## License

CC0 1.0 — Public domain. Use however you want.