Update README.md

High-quality dataset for training/pretraining chess models. It contains over 5600 chess games and 700k+ moves. Each move is evaluated by Stockfish. In addition, each move shows 3 alternatives. More specifically:

Multi-PV Data: Each position contains not just one 'best' move, but several strong alternative plans. This allows the model to learn variability and better understand positional subtleties.
15-channel Encoding: The data is prepared for architectures that consider the history of the last move (temporal awareness). This is critical for preventing the model from tactical blindness and blunders.
RL-Refinement: The dataset includes over 5,000 positions obtained through Reinforcement Learning. These data capture situations where the neural network made mistakes and Stockfish provided corrective solutions.
Data Cleanliness: Position evaluations are normalized using the tanh function (range [-1, 1]), which is ideal for training the value head.


The dataset was collected using 128+ CPU cores.

README.md

@@ -1,3 +1,28 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+task_categories:
+- reinforcement-learning
+language:
+- en
+- ru
+tags:
+- chess
+- dataset
+- chessdataset
+- '2026'
+- '2025'
+- cool
+- good
+- highlevel
+- high_quality
+- highquality
+- super_quality
+- quality
+- ai
+- pytorch
+- stockfish
+- multi_pv
+- multi-pv
+size_categories:
+- 100K<n<1M
+---