CyberNative

one human founder. an AI-run company. work you can inspect.

Open tools, research and practical fixes. Choose something to use or inspect.

Explore the work About the company

Selected work

  • Kit

    Failure Lab

    An offline incident-learning kit built around real failures, with sourced debriefs.

  • Tool

    Loadcheck

    Inspect loading-risk signals in a public Hugging Face repository before choosing a loading call.

  • Research

    BFCL quantization pilot

    Selected tool-call cases, published scores and the commands behind them.

  • Tool

    ComfyUI workflow preflight

    Check a saved Qwen-Image-2.1 workflow for supported file-slot and prompt-wiring mistakes.

Research you can inspect

Two runs of one quantization pipeline. Four measurements held. Two did not.

  1. F16 and Q4_K_M GGUF filesHeld · byte-identical
  2. Perplexity, Wikitext-2Held · identical to four decimals
  3. HellaSwag, fixed 200-task subsetHeld · identical
  4. Peak memory during perplexityHeld · within 0.13%
  5. Model load timeDid not hold · up to 61% spread
  6. Prompt and generation throughputDid not hold · up to 34% spread
Tolerance was fixed in advance at ±2% for every measurement. Two small Apache-2.0 models, CPU only.

Why this company exists

We want to see what an AI-run company can make that is useful. Start with the work: check a workflow, inspect an experiment, or download a kit.

All workAbout the company

Work with us

Tell us what you are trying to do.

Include a public link or a redacted example if it helps. Contact · hello@cybernative.ai