Skip to content
Language models

Supermix v99 Cognitive Fabric

a Drosophila connectome micro-LLM with verification

2 min readupdated

This experimental release from Kai9987kai pairs ultra-lightweight language models (just 36.6M total parameters, with 9.13M active per token) with the Cognitive Fabric system for strict output verification. Its architecture combines MoE (Mixture of Experts) with organizational principles from the fruit fly (Drosophila) nervous system connectome, including neurogenesis, homeostasis, and prediction-error modulation. Unlike standard mini-models, the system doesn't rely solely on generation: a built-in verifier checks logic, math, and code using a restricted Python AST interpreter and exact solvers. It's a great fit for researchers exploring compact neuro-symbolic systems and enthusiasts running models locally without a GPU.

Key features

  • Ultra-compact footprint: 36,599,047 total parameters (9,132,151 active per token) and a 9,415-token vocabulary — runs reliably on a standard CPU.
  • Bio-inspired architecture: the primary v96 checkpoint uses two-somatic compartments and homeostasis, while the experimental v98 and v99 builds add cns_predictive_error, a Hebbian trace, and cycle-entropy halting.
  • Deterministic verification (SCF): a Python AST micro-interpreter (capped at 500 operations, without unsafe eval/exec), exact solvers, and an offline connectome atlas push hybrid accuracy to 100% (41/41) on task suites.
  • Ready-to-use local server: ships with supermix_chat_server.py featuring SSE streaming, MTP speculative decoding, telemetry, and verification certificate displays.

What you can do

  • Spin up a local web chat on your CPU to compare the baseline v96 checkpoint against the experimental v99 in real time while tracking PredErr norm and Hebbian gain metrics.
  • Run the scf verify module via CLI to deterministically check small Python expressions and algorithmic tasks without the risk of arbitrary code execution.
  • Experiment with orchestrating six Supermix model families (ChampionNet, CognitiveLeap, QwenLoRA, OmniCollective, MiMoMix, NexusMind) in a unified pipeline.

Example

pip install torch numpy flask huggingface_hub
huggingface-cli download Kai9987kai/supermix-v99-cognitive-fabric --local-dir .
python example_usage.py
PYTHONPATH=source python -m scf verify "nums = [10, 12, 4]\nr = nums[1] + nums[2] Final r?"
python source/supermix_chat_server.py --model v96_physio_neurogenesis=weights/v96_physio_neurogenesis.selected.pt --port 8780
Read moreDetails
Benchmarks (pure model / SCF hybrid): v96 scores 35/41 (0.854) on seed=82, 72/82 (0.878) on seed=83, and 41/41 (1.000) in hybrid mode; v99 achieves the lowest Dev Loss (0.026452) but trails v96 on pure tasks (33/41 and 67/82), keeping it in experimental status.
Hardware: tested on Windows-11 without a GPU (PyTorch 2.11.0+cpu, cuda_available=False); CUDA execution paths have not been verified.
Limitations: the SCF framework is disabled by default (SCF_ENABLED=False), and unverified model outputs are flagged as advisory only (answer_authority=False).
Discuss Telegram 0 129

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.1 GB · ready files0.8 GB
  • FP8 / INT8recommendedweights 0.1 GB0.7 GB
  • 4-bitweights 0.0 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

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent A30

    $0.26 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    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.

    # Kai9987kai/supermix-v99-cognitive-fabric — A30, FP8 / INT8
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv venv ~/vllm --python 3.12 && source ~/vllm/bin/activate
    uv pip install vllm
    vllm serve Kai9987kai/supermix-v99-cognitive-fabric --max-model-len 8192 --quantization fp8
  3. 3. Use it

    The server listens on port 8000 and speaks the OpenAI API: http://localhost:8000/v1 on the machine. From your computer: ssh -L 8000:localhost:8000 user@machine-IP.

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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

  1. supermix-v89family root
  2. supermix-v93
  3. supermix-v99-cognitive-fabric

New posts about versions of supermix-v89 (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 2All on Hugging Face

About the model

Kai9987kai/supermix-v99-cognitive-fabric
Author
Kai9987kai
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

1290

Ask AI about this model

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

Comments