SAVRN
Search Contact SAVRN

Dataset · Reinforcement learning

chess-arena-results

by TobiasLogic TobiasLogic/chess-arena-results

Every finished game played in Chess Arena, kept here because a Space has no filesystem that survives a restart.

Rows1,810
Configurations1
Size67.2 KB
Licensemit
AccessPublicly accessible
Monthly Downloads2.5k

Dataset Card

By TobiasLogic, published under mit, revision fefb3365d2f1.

Every finished game played in Chess Arena, kept here because a Space has no filesystem that survives a restart. The Space reads this on startup and writes back a few seconds after each game, so the leaderboard carries across restarts, sleeps and rebuilds rather than resetting every time the container is recycled.

It is append only. Nothing in the arena can delete a result.

Format

results.json is a JSON array of games, one array each:

[
  ["chessmamba2", "chess-llama", 1.0],
  ["engine-3", "random", 1.0],
  ["chessaggro", "chessgpt", 0.5]
]
Position Meaning
0 player id with the white pieces
1 player id with the black pieces
2 result from white's point of view: 1.0 white won, 0.5 drawn, 0.0 black won

Read the full dataset card (524 words)

Structure

default 1,810 rows

SplitRowsSize
train1,81075.7 KB
textlist

Details

Repository
TobiasLogic/chess-arena-results
Publisher
TobiasLogic
Task category
Reinforcement learning
Tags
chess, leaderboard, game-results
Size category
n<1K
Languages
Not stated by the source
Revision
fefb3365d2f1b04b4f453dcb525302469e08b683
Last updated
2026-10-01

Files

4 files, 67.2 KB in total.

Data2 files · 61.2 KB
Documentation1 file · 3.5 KB
Repository1 file · 2.5 KB
Every file
FileTypeSizeSHA-256
archive/results_before_gpu_fix_2026-09-06.jsonData1.7 KB—
results.jsonData59.5 KB—
README.mdDocumentation3.5 KB—
.gitattributesRepository2.5 KB—

License and Download

License
mit
Access
No access gate
Download from TobiasLogic

Released by TobiasLogic through its official repository on Hugging Face. Read the license.