# Laya Minecraft decision model by Thomas Tesche

> Laya Minecraft selects valid game actions and checks danger. Read its fine-tuning data, Apache licence and disclosed overlap in the evaluation set.

Updated 2026-11-12 (Australia/Adelaide)

## Quick answers

### What is Laya Minecraft by Thomas Tesche?

[Laya Minecraft](<https://decisionmodels.ai/models/laya-minecraft>): An English [Laya](<https://decisionmodels.ai/models/laya>) fine-tune for a Minecraft agent.

[Laya Minecraft facts](<https://decisionmodels.ai/models/laya-minecraft>)

[Source](<https://huggingface.co/thtesche/laya-minecraft-english>)

### Who makes Laya Minecraft?

[Laya Minecraft](<https://decisionmodels.ai/models/laya-minecraft>): Thomas Tesche

[Laya Minecraft facts](<https://decisionmodels.ai/models/laya-minecraft>)

[Source](<https://huggingface.co/thtesche/laya-minecraft-english>)

### Are Laya Minecraft's weights open?

[Laya Minecraft](<https://decisionmodels.ai/models/laya-minecraft>): Public weights, question signatures and train\_report.json; [Laya](<https://decisionmodels.ai/models/laya>) runtime Licence: Apache-2.0 model card

[Laya Minecraft facts](<https://decisionmodels.ai/models/laya-minecraft>)

[Source](<https://huggingface.co/thtesche/laya-minecraft-english>)

By Thomas Tesche · Weights upload: 2026-11-07 · OPEN

An English [Laya](<https://decisionmodels.ai/models/laya>) fine-tune for a Minecraft agent. Code removes impossible actions; the model chooses an intent and answers a danger question, then an executor acts and a verifier checks progress. The encoder is frozen during this training run. Its reported evaluation contains training or calibration overlap.

## Facts

- Base: convaiinnovations/laya, ModernBERT-large
- Inputs: English game-state descriptions and valid action candidates; intent and in\_danger choice questions
- Training: 2,564 verified decision ticks, four epochs; temperatures fitted on 284 calibration items
- Licence: Apache-2.0 model card
- Access: Public weights, question signatures and train\_report.json; [Laya](<https://decisionmodels.ai/models/laya>) runtime

## Claims and sources

- **VENDOR-REPORTED**: The maker reports accuracy 0.9301 before fine-tuning and 0.9325 after, with ECE moving from 0.0625 to 0.0435 on 415 items. Four items overlap training or calibration, so this is not a strictly held-out result. No score or noul items were trained. [source](<https://huggingface.co/thtesche/laya-minecraft-english>)

## Links

- [Source: Model card and weights](<https://huggingface.co/thtesche/laya-minecraft-english>)
- [Source: Training and evaluation report](<https://huggingface.co/thtesche/laya-minecraft-english/raw/main/train_report.json>)
- [Source: Agent source](<https://github.com/thtesche/minecraft-laya-training>)

## Related projects

No related projects are listed in the registry.
