---
tags:
- classify-and-augment
- llm-annotated
---

# davanstrien/aud-llama-3.2-3b-instruct-20260428

LLM-annotated dataset produced by [classify-and-augment](https://github.com/davanstrien/classify-and-augment).

## Configuration

- **Model**: `meta-llama/Llama-3.2-3B-Instruct`
- **Labels**: `positive`, `negative`
- **Input rows**: 180
- **Output rows**: 221

## Label distribution

| Label | Real | Synthetic | Total |
|---|---:|---:|---:|
| `negative` | 154 | 0 | 154 |
| `positive` | 26 | 41 | 67 |

## Synthesis audit

| Class | Needed | Generated | Validated | Kept | Acceptance |
|---|---:|---:|---:|---:|---:|
| `positive` | 74 | 86 | 74 | 41 | 86.0% |

Acceptance = synthetic candidates that the same model re-classified as the target class (self-consistency check, [Synthetic Imputation, arxiv 2504.15160](https://arxiv.org/abs/2504.15160)).