Leela Chess Zero Engine
More actions
Available in: Builder, Library, Practice, Tools
The Leela Chess Zero Engine card provides deep-learning evaluations of the current position using the t1-256x10 distilled neural network. Unlike classical engines that rely on brute-force search, Lc0 evaluates positions through pattern recognition, long-term planning, and human-like strategic intuition. This makes it a valuable tool for analysing complex middlegames, quiet improvements, positional sacrifices, and other subtle ideas essential for modern opening preparation and training inside the Repertoire Builder.
Network: t1-256x10 Lc0 uses a single fixed neural network in this feature, providing consistent and high-quality deep-learning evaluations.
- ◦ Size: T1-256x10 distilled (swa 2432500)
- ◦ Architecture: 256 filters, 10 residual blocks
- ◦ File Size: 30–40 MB
For the current position, this section displays:
- ◦ Move: Lc0's preferred move based on its neural-network evaluation.
- ◦ Eval: The evaluation score converted into centipawns for compatibility with standard engine displays.
- ◦ Probability: Lc0's predicted likelihood that the chosen move is best.
- ◦ WDL: Win, draw, and loss probabilities based on the network's evaluation of the overall position.
You can click on any move to add it directly to your opened item, or right-click to add the move together with its evaluation, probability, and WDL data as a comment for immediate reference.
This feature helps you:
- + Explore strategic ideas derived from deep-learning evaluation instead of brute-force search
- + Understand quiet improvements, long-term plans, and subtle positional concepts
- + Compare deep-learning suggestions with Stockfish or Maia to gain multiple perspectives on the same position
- + Add Lc0-evaluated moves directly into your repertoire or study material