davanstrien/imdb-classify-augment-v3-synth-test
收藏Hugging Face2026-04-28 更新2026-05-03 收录
下载链接:
https://hf-mirror.com/datasets/davanstrien/imdb-classify-augment-v3-synth-test
下载链接
链接失效反馈官方服务:
资源简介:
---
tags:
- classify-and-augment
- llm-annotated
---
# davanstrien/imdb-classify-augment-v3-synth-test
LLM-annotated dataset produced by [classify-and-augment](https://github.com/davanstrien/classify-and-augment).
## Configuration
- **Model**: `HuggingFaceTB/SmolLM3-3B`
- **Labels**: `positive`, `negative`
- **Input rows**: 20
- **Output rows**: 30
## Label distribution
| Label | Real | Synthetic | Total |
|---|---:|---:|---:|
| `negative` | 9 | 6 | 15 |
| `positive` | 11 | 4 | 15 |
## Synthesis audit
| Class | Needed | Generated | Validated | Kept | Acceptance |
|---|---:|---:|---:|---:|---:|
| `positive` | 4 | 8 | 4 | 4 | 50.0% |
| `negative` | 6 | 12 | 6 | 6 | 50.0% |
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)).
提供机构:
davanstrien



