DedeProGames/NanoAndy-350M

TEXT GENERATIONConcurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Jul 15, 2026License:otherArchitecture:Transformer0.0K Featherless Exclusive Cold

DedeProGames/NanoAndy-350M is a compact 354M parameter Minecraft agent model, built on LiquidAI/LFM2.5-350M with a hybrid conv + attention architecture. It is full-fine-tuned for controlling Minecraft characters via Mindcraft-CE, specifically optimized for direct in-game responses by removing chain-of-thought and function-calling data. This model is designed for resource-constrained environments, enabling efficient deployment on low-VRAM GPUs or CPU-only machines.

Loading preview...

Overview

NanoAndy-350M is a compact, 354 million parameter Minecraft agent model developed by DedeProGames. It is a full-fine-tuned version of LiquidAI/LFM2.5-350M, specifically designed to act as the "brain" for bots within the Mindcraft-CE platform, which allows LLMs to control Minecraft characters.

Key Differentiators

This model is optimized for efficiency and direct action, making it suitable for environments with limited resources:

  • "Nano" Design: Achieved by stripping down the training data from Andy-4.1. It removes chain-of-thought reasoning (<think>...</think>) to focus on direct in-game responses.
  • No Function-Calling: Conversations involving external tool-role calls were excluded from training, ensuring the model focuses on Mindcraft's native chat/command format.
  • Resource-Efficient: At 350M parameters, it's ideal for low-VRAM GPUs, CPU-only machines, or running multiple bots simultaneously.

Intended Use

NanoAndy-350M is primarily intended to be integrated into Mindcraft-CE bot profiles (e.g., andy.json) as the chat/coding model, served locally via tools like LM Studio, llama.cpp, or vLLM. It is narrowly tuned for the Mindcraft agent format and is not a general-purpose assistant.

Limitations

  • Not a reasoning model: Lacks chain-of-thought capabilities.
  • No native tool-calling: Expects Mindcraft's native command/chat format.
  • Small model: May struggle with complex planning or multi-step reasoning compared to larger models.
  • English only.