Skip to content
Other

Mundus 0.1

Autonomous Neural Game Engine

2 min read

Mundus 0.1 is a two-stage causal transformer world model designed to simulate latent dynamics, interactive environments, and model-based reinforcement learning. It compresses visual frames using VQ-VAE (an image discretization method) and predicts future frames, collisions, and rewards in response to player actions. Available in 88.2M and 232.1M parameter versions, the model runs on modest hardware (requiring from 1.68 GB of VRAM) and is useful for game developers, AI researchers, and interactive neural network enthusiasts.

Key features

  • Two-stage architecture: Combines a VQ-VAE visual tokenizer and a Causal Dynamics Engine built on SwiGLU transformer blocks.
  • Interactive web player: Lets you run simulations in the browser and control the agent in real time using arrow keys or the keyboard.
  • Imagination generation: Generates autoregressive GIF animations of expanded scenarios (model "hallucinations").
  • Efficiency: The smaller 88.2M parameter version is optimized for budget GPUs like the GTX 1050.

What you can do

  • Run a web simulation: Play in the browser via a local server, switching between pure neural generation and physics.
  • Create a GIF animation: Generate a sequence of future frames (rollout) using the `imagine.py` script.
  • Train from scratch: Run transformer dynamics training on your own trajectories and game data.
  • Run on a PC: Use the built-in Pygame interface to test the model on your desktop.

Example

Launching the interactive browser engine with a model checkpoint:

python3 web_player.py --checkpoint checkpoints/mundus_0.1_step_45000.pt --port 8080
Read moreDetails
Parameters: 88.2M (active) and 232.1M (high-capacity) parameters.
Architecture: 12 or 18 SwiGLU transformer blocks, hidden dimension 768 / 1024, RoPE (rotary positional embeddings).
Hardware requirements: From 1.68 GB of VRAM for the 88.2M version up to 3.52 GB of RAM for the 232.1M version.
License: MIT.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.3 GB · ready files0.9 GB
  • FP8 / INT8recommendedweights 0.2 GB0.7 GB
  • 4-bitweights 0.1 GB0.6 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # Tahary/mundus-0.1-world-model
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download Tahary/mundus-0.1-world-model --local-dir ~/models/mundus-0.1-world-model
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to mundus-0.1-world-model
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of mundus-0.1-world-model (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

Tahary/mundus-0.1-world-model
Author
Tahary
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments