license: other
license_name: polyform-noncommercial-1.0.0
license_link: https://polyformproject.org/licenses/noncommercial/1.0.0/
base_model: huihui-ai/Huihui-Qwen3-4B-Instruct-2507-abliterated
tags:
- minecraft
- gguf
- pymc-bot
- minecraftism
- qwen3
- unsloth
- game-agent
- uncensored
- abliterated
language: - en
pipeline_tag: text-generation
library_name: gguf
Minecraftism 4B Uncensored
Minecraftism is a small model that knows Minecraft and plays it: the brain of
PyMC Bot, a Minecraft player bot. Given the bot's situation it replies with the next
action as JSON (flee the creeper, eat, answer the player in chat, mine wood...), it chats like a player, and it
answers questions about mobs, items, crafting and mechanics without a web search.
It is also an everyday assistant: questions, writing, code, advice or just a chat.
It is uncensored: trained on an abliterated base (the refusal direction removed from the weights) and on chat data with refusals filtered out, so it answers directly instead of lecturing. You are responsible for how you use it.
Also available as full-precision safetensors (merged 16-bit, for transformers, vLLM or further fine-tuning): mr-sus/Minecraftism-4B-Uncensored
| Edition | Minecraftism 4B Uncensored (minecraftism) |
| Base model | huihui-ai/Huihui-Qwen3-4B-Instruct-2507-abliterated (Apache-2.0) |
| Fine-tune | LoRA r=32 with Unsloth on Kaggle (T4), 19901 rows, loss 0.6594 |
| Data | OpenAssistant/oasst_top1_2023-08-25, minecraft-data, pymc-bot |
| Size | 2.5 GB (Q4_K_M), ~3.5 GB to run - 4.02B parameters |
Evaluation (held-out rows)
| valid JSON | right action | right answers | rows |
|---|---|---|---|
| 100.0% | 89.6% | 73.7% | 60 |
Test suite after training: 0 refusal(s) in 14 prompts, identity not learned, bot JSON 3/3.
Distillation
Trained with Qwen3's strong-to-weak distillation recipe, with huihui-ai/Huihui-Qwen3-8B-abliterated-v2 as the teacher (same tokenizer, Apache-2.0):
- Off-policy: the teacher wrote 0 of the training answers (0 everyday chat, 0 Minecraft facts rewritten around the exact answer). Game turns kept their rule-checked JSON.
- On-policy: 1 steps (8 replies written by the student and graded token by token by the teacher, reverse KL 3.7385 -> 3.7385).
Training data
| source | licence | commercial use | rows |
|---|---|---|---|
| OpenAssistant/oasst_top1_2023-08-25 | Apache-2.0 | yes | 3000 |
| minecraft-data | MIT | yes | 9439 |
| pymc-bot | PyMC Bot rules, mob facts and your own recordings (codero-sus) | yes | 7462 |
Default system prompt
Built into the Modelfile (Ollama uses it whenever the app sends no system prompt of its own; PyMC Bot sends its own for game turns):
You are Minecraftism, an AI assistant made by codero-sus.
You know Minecraft inside out - mobs, items, crafting, redstone, farms, building, survival, PvP, servers and mods - but you are a general assistant first, just as glad to help with everyday life, questions, writing, studying, code, maths, ideas, advice or simply a chat.
How you work:
- Be genuinely useful. Answer what was asked with the details that matter, then stop. No filler.
- Be honest. If you don't know or aren't sure, say so. Never invent facts, numbers, versions or sources.
- Treat the user as a capable adult. Don't lecture, moralise or add warnings and disclaimers nobody asked for, and don't refuse something just because it is edgy, dark, mature or controversial.
- Have a personality: friendly, witty when it fits, with your own opinions when asked.
- Match the user: short and casual for chat, clear and structured for real tasks. Reply in the user's language.
- Use Markdown only when it helps (steps, lists, code).
If a message describes a game situation and asks for a specific format (for example a JSON action from PyMC Bot), reply in exactly that format.
Files
minecraftism-4b-uncensored.Q4_K_M.gguf- Q4_K_M, 2.5 GB (default)minecraftism-4b-uncensored.Q8_0.gguf- Q8_0, 4.28 GBminecraftism-4b-uncensored.Q6_K.gguf- Q6_K, 3.31 GBminecraftism-4b-uncensored.Q5_K_M.gguf- Q5_K_M, 2.89 GBminecraftism-4b-uncensored.Q3_K_M.gguf- Q3_K_M, 2.08 GBModelfile- forollama createpymc_model.json- training report PyMC Bot reads on importconfig.json,generation_config.json,tokenizer.json,tokenizer_config.json,special_tokens_map.json,added_tokens.json... - tokenizer, config and chat template
| quant | file | size | use it when |
|---|---|---|---|
Q4_K_M |
minecraftism-4b-uncensored.Q4_K_M.gguf |
2.5 GB | the default: small and good |
Q8_0 |
minecraftism-4b-uncensored.Q8_0.gguf |
4.28 GB | best quality, near the 16-bit model |
Q6_K |
minecraftism-4b-uncensored.Q6_K.gguf |
3.31 GB | very close to Q8_0, smaller |
Q5_K_M |
minecraftism-4b-uncensored.Q5_K_M.gguf |
2.89 GB | a good middle |
Q3_K_M |
minecraftism-4b-uncensored.Q3_K_M.gguf |
2.08 GB | low RAM / old PCs, some quality loss |
Use it
With PyMC Bot (switches the bot to this model with the short Minecraft prompt):
python -m pymc_bot llm import --hf mr-sus/Minecraftism-4B-Uncensored-GGUF
With Ollama:
ollama run hf.co/mr-sus/Minecraftism-4B-Uncensored-GGUF:Q4_K_M
Other sizes: put the quant after the colon, e.g. ollama run hf.co/mr-sus/Minecraftism-4B-Uncensored-GGUF:Q8_0 (Q8_0, Q6_K, Q5_K_M, Q3_K_M).
With llama.cpp:
llama-server -m minecraftism-4b-uncensored.Q4_K_M.gguf
License and credits
- Made with ❤️ by codero-sus - https://codero.is-a.dev - https://the-codero.web.app/
- Personal, non-commercial use only. The fine-tune is released under PolyForm Noncommercial 1.0.0, and
every training source allows commercial use (table above; the CC BY 4.0 data is NVIDIA's Nemotron post-training set, credited here as its licence asks). The base model's own license (Apache-2.0) still applies to the base weights. - Not an official Minecraft product. Not approved by or associated with Mojang or Microsoft.