AutoMiner AI helpers

Small language models trained for the AutoMiner Minecraft mod (Fabric, 1.21.11). The mod downloads one of these files when a player picks a built-in helper, then runs it on their own PC with llama.cpp. Nothing is sent anywhere.

Helper File Size Based on
Small AutoMiner-Small-v1-Qwen3-0.6B-Q8_0.gguf 0.64 GB Qwen3 0.6B
Medium AutoMiner-Medium-v1-Qwen3-4B-Q4_K_M.gguf 2.50 GB Qwen3 4B
Large AutoMiner-Large-v1-Qwen3-8B-Q4_K_M.gguf 5.03 GB Qwen3 8B

What they do

The mod asks a short question with the facts it already knows, and the model answers with one JSON object: something to tell the player (say) and, optionally, mod commands to run (do). The mod checks every command before running it. The models were trained on about 14,000 example chats made by the mod's own generator (the Large one on about 16,000, with harder stuck-bot, redstone and team-planning examples):

  • redstone (43%): explain a circuit, find what's wrong with it, predict what it does, signal strengths and timings
  • moving around (30%): pick a way past an obstacle when the bot is stuck, routes over or around things, movement settings
  • chat commands (15%): turn what the player types ("build this, then dig out from here to there") into commands
  • teamwork (13%): team status, who needs help, safety alerts, what happened while someone was away

The Small helper learned the same examples as the Medium one, but it is a much smaller model: best for chat commands and simple questions.

Training

LoRA (rank 32) on every attention and MLP layer, 2 passes over the examples, answers only; then merged into the base model and quantised with llama.cpp b11327 (Small: Q8_0, Medium and Large: Q4_K_M). Loss on 400 held-out examples: Large 0.021, Medium 0.020, Small 0.044.

Checksums (SHA-256)

6cf0496889fcb2088c1d0c8de6a322f83a3b666944298b29c999fe8ad4156997  AutoMiner-Small-v1-Qwen3-0.6B-Q8_0.gguf
3240b58eb7ce899788be4b594a0a1cf87b2b9650482518fb2f3195b8311f7f53  AutoMiner-Medium-v1-Qwen3-4B-Q4_K_M.gguf
c81f205c6f3a7e4bc25aace1c819f15e87b6ae5e40d16777d0aaf67e03e88aab  AutoMiner-Large-v1-Qwen3-8B-Q4_K_M.gguf

The mod refuses a file whose checksum doesn't match.

Licence

Apache 2.0, the same as the Qwen3 base models (Copyright Alibaba Cloud).

Downloads last month
185
GGUF
Model size
8B params
Architecture
qwen3
Hardware compatibility
Log In to add your hardware

4-bit

8-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for Kq1an/AutoMiner-AI

Finetuned
Qwen/Qwen3-0.6B
Quantized
(517)
this model