Skip to content
Other

RoboRACE-Net: A mini-transformer for filtering lies in agent swarms

2 min readupdated

RACE-Net is a compact AI model (an axial stacked transformer with 213,889 parameters) developed by Nabidnur, designed to merge reports from multiple agents (such as camera clusters and VLM (multimodal language models)) in environments where some agents may lie, malfunction, or output random data. The model processes an unannotated window of episodes as a whole and returns the log-odds of success for each episode, assuming that the receiving agent's column is honest. Running entirely on CPU, this model is useful for robotics and multi-agent system developers looking to reliably aggregate data from untrusted sources.

Key features

  • Compactness: Just 213,889 parameters, enabling fast execution on CPU in milliseconds per window.
  • Position-encoding free: The model is permutation-equivariant across both episodes and agents.
  • Flexible input: Accepts an arbitrary-sized score matrix from agents across recent episodes.
  • Simulation training: Trained exclusively on synthetic agent swarms without using real-world benchmarks.

What you can do

  • Swarm data analysis: Run Python inference to aggregate task success scores from a group of cameras or sensors.
  • Robotics integration: Apply it to filter out false positives in distributed robot control systems.
  • Custom data testing: Test resilience against adversarial agents, random noise, and collusion.
  • Environment emulation: Use built-in synthetic swarm generators for fine-tuning or analysis.

Example

import sys, numpy as np
from huggingface_hub import snapshot_download
d = snapshot_download("Nabidnur/RoboRACE-Net"); sys.path.insert(0, d)
import racenet

P = np.random.rand(100, 5) # [эпизоды, агенты] оценки от 0 до 1
log_odds, thr = racenet.race_net(None, P, own=0, path=d)
success = log_odds > thr
Read moreDetails
• Achieves 81.6% accuracy on the SO-101 test track and 81.8% on Franka Panda in zero-shot testing.
• Hardware requirements: Runs on a standard CPU in milliseconds; dependencies include torch, numpy, and safetensors.
• Limitations: Does not outperform heavy EM methods on real cameras, requires the receiving agent to be more honest than random guessing, and works only with binary truth.
Examples
Discuss Telegram 0 24

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.0 GB · ready files0.5 GB
  • BF16weights 0.0 GB0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 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 · F32 — 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.

    # Nabidnur/RoboRACE-Net
    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 Nabidnur/RoboRACE-Net --local-dir ~/models/RoboRACE-Net
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to RoboRACE-Net
  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 RoboRACE-Net (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Nabidnur/RoboRACE-Net
Author
Nabidnur
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

240

Ask AI about this model

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

Comments