MineDojo: Building Open-Ended Embodied Agents with Internet-Scale Knowledge
Paper • 2206.08853 • Published • 1
Minecraft Bot Code Generation Model
A specialized LLaMA 3.2 model fine-tuned to generate Mineflayer-compatible JavaScript code for Minecraft bot automation.
Model Description
This model generates executable JavaScript code using the Mineflayer API to create intelligent Minecraft bots. It can produce code for various bot behaviors including mining, building, combat, navigation, and server interaction.
Data Creation Process
- Source Dataset: MineDojo - A comprehensive dataset containing Minecraft gameplay data, tutorials, and documentation
- QA Generation: MineDojo data was processed through LLaMA Scout model hosted on Groq to generate question-answer pairs focused on Minecraft bot programming scenarios
- Data Refinement: Generated QA pairs were further processed using ChatComplete by Unstip for quality improvement
- Fine-tuning: The curated dataset was used to fine-tune LLaMA 3.2 3B using Unsloth for 2x faster training
Usage
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("SanthoshToorpu/minecraft-bot-model")
model = AutoModelForCausalLM.from_pretrained("SanthoshToorpu/minecraft-bot-model")
prompt = "Create a bot that automatically farms wheat"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=200)
code = tokenizer.decode(outputs[0], skip_special_tokens=True)
Capabilities
- Movement & Navigation: Pathfinding, exploration, terrain traversal
- Resource Management: Mining, crafting, inventory organization
- Building & Construction: Automated structure creation
- Combat Systems: PvP and PvE bot behaviors
- Server Integration: Chat commands, multiplayer coordination
Training Details
This model was trained 2x faster with Unsloth and Hugging Face's TRL library.
References
Uploaded model
- Developed by: SanthoshToorpu
- License: apache-2.0
- Finetuned from model : unsloth/llama-3.2-3b-instruct-bnb-4bit
This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support