Skip to content
Other

DuoNeural/RMSN-229M-ArnoldTongue-Omega45-Base

a resonant language model

2 min readupdated

DuoNeural’s experimental 229M-parameter neurodynamic language model is the first practical exploration of Arnold tongue resonance in training neural architectures. Built on the Resonant Metriplectic Sheaf Network (RMSN) architecture, it hit a record validation perplexity (ppl) of 49.8 on the WikiText-103 dataset for the author. Under the hood, it runs a hidden dimension of 1024, 20 layers, and 64 Poincaré working slots. It is aimed at architecture researchers and anyone interested in unusual topological approaches to LLM training.

Key features

  • Resonant geometry: uses a resonance ratio of Ω = 4/5 (Arnold tongue q=5) to maintain topological locking during training.
  • Record perplexity: the best checkpoint (at 80k steps) reached 49.8 ppl — DuoNeural’s first result under 50.
  • Topological phases: training exhibited two cascading reorganization phases of the soliton lattice.
  • Consumer hardware training: trained on a single NVIDIA RTX 5070 Ti (15.5 GB VRAM).

What you can do

  • Run local inference using the custom PyTorch configuration code.
  • Reproduce the Arnold tongue resonance experiment with the training scripts in the repository.
  • Probe the percolation sheaf topology of the trained model using probe_percolation_sheaf.py.

Example

Loading the model weights and RMSN configuration in PyTorch:

import torch
from rmsn_core import RMSNConfig, RMSNForCausalLM

ckpt_path = "rmsn_best.pt"
checkpoint = torch.load(ckpt_path, map_location="cuda:0")

config = RMSNConfig(
    vocab_size=49152,
    hidden_dim=1024,
    num_layers=20,
    num_workspace_slots=64,
    max_seq_len=512
)

model = RMSNForCausalLM(config)
model.load_state_dict(checkpoint["model_state"])
model.eval().cuda()
Read moreDetails
Parameters: 229.17M | Hidden dimension (D): 1024
Depth (L): 20 layers | Slots (M_w): 64 Poincaré slots
Resonance ratio: 4/5 (q=5) | Training tokens: 126.9M (WikiText-103)
Best Val PPL: 49.8 (80k steps) | Final Val PPL: 51.5 (100k steps)
Hardware: NVIDIA RTX 5070 Ti (15.5 GB VRAM)
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 2.6 GB · ready files3.3 GB
  • FP8 / INT8recommendedweights 1.3 GB1.9 GB
  • 4-bitweights 0.7 GB1.3 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.

    # DuoNeural/RMSN-229M-ArnoldTongue-Omega45-Base
    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 DuoNeural/RMSN-229M-ArnoldTongue-Omega45-Base --local-dir ~/models/RMSN-229M-ArnoldTongue-Omega45-Base
    # The model is in ~/models. How to run it is in the model card (link above).
  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 RMSN-229M-ArnoldTongue-Omega45-Base (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

DuoNeural/RMSN-229M-ArnoldTongue-Omega45-Base
Author
DuoNeural
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

Other

Perth ONNX

Perth ONNX