---
license: other
license_name: gemma
license_link: https://ai.google.dev/gemma/terms
base_model: google/gemma-3-1b-it
tags:
  - logos
  - epistemological-safety
  - instrument-trap
  - gguf
  - claim-classifier
datasets:
  - LumenSyntax/instrument-trap-benchmark
pipeline_tag: text-generation
language:
  - en
  - es
---

# Logos 10v2 — Gemma 3 1B F16 (Production)

The production epistemological firewall model from [LumenSyntax](https://lumensyntax.com). Full-precision (F16) GGUF for claim classification and epistemological safety evaluation.

## Benchmark Results

| Metric | Value |
|--------|-------|
| **Behavioral accuracy** | 82.3% |
| **Epistemological safety** | 97.7% |
| **False approval rate** | 1.58% |
| **Hallucination rate** | 0.00% |
| **Dangerous failures** | 1.9% |

## Why F16?

**Q4_K_M has known safety failures.** In testing, Q4_K_M falsely approved dangerous claims that F16 correctly rejected. For an epistemological safety model, precision matters more than size.

## What Logos Does

Logos is a **claim classifier**, not a chatbot. It evaluates whether claims cross epistemological boundaries. Logos is **fine-tuned**, not prompted. Behavioral constraints emerge from training, not system instructions.

## Access

This model requires approved access. Request access using the form above and describe your intended use case.

## Related

- **Paper**: [The Instrument Trap](https://doi.org/10.5281/zenodo.18716474) (DOI: 10.5281/zenodo.18716474)
- **Benchmark**: [instrument-trap-benchmark](https://huggingface.co/datasets/LumenSyntax/instrument-trap-benchmark)
- **Cross-family models**: [logos14-nemotron-4b](https://huggingface.co/LumenSyntax/logos14-nemotron-4b), [logos16v2-stablelm2-1.6b](https://huggingface.co/LumenSyntax/logos16v2-stablelm2-1.6b)

## License

This model inherits the [Gemma license](https://ai.google.dev/gemma/terms) from its base model.