---
license: apache-2.0
language:
- ko
- en
task_categories:
- text-generation
tags:
- korean
- synthetic
- instruction
- on-policy
- eagle3
size_categories:
- 100K<n<1M
pretty_name: Korean On-Policy QA for Gemma 4 26B-A4B (EAGLE-3)
configs:
- config_name: default
  data_files:
  - split: train
    path: data/*.jsonl
---

# Korean On-Policy QA (Gemma 4 26B-A4B) — EAGLE-3 training data

Instruction/response pairs whose **responses were regenerated on-policy by
`BCCard/gemma-4-26B-A4B-it-FP8-Dynamic`**. Built to retrain an EAGLE-3 speculator for
Korean, but also usable for general instruction-tuning / distillation.

## Structure
- **Rows**: ~150,000
- **Columns**: `instruction` (str), `output` (str, verifier-generated), `messages` (chat list)
- **Split**: `train`

## How it was made
- **Prompt source**: [sh2orc/bccard-maywell-jojo0217-markai-lcw99-kendamarron-microsoft](https://huggingface.co/datasets/sh2orc/bccard-maywell-jojo0217-markai-lcw99-kendamarron-microsoft)
  (1.71M-row Korean/English QA). Only the `instruction` column was sampled (~150k); the original answers were discarded.
- **Response regeneration**: `RedHatAI/gemma-4-26B-A4B-it-FP8-Dynamic`, thinking off, text-only.

## Usage
```python
from datasets import load_dataset
ds = load_dataset("BCCard/gemma-4-26B-A4B-korean-on-policy-150k", split="train")
print(ds[0]["instruction"], ds[0]["output"])
```

## License / provenance
**Apache 2.0**. Both the prompt-source dataset and the response-generating model Gemma 4
are Apache 2.0. Responses are synthetic data generated by Gemma 4; source-dataset
attribution is retained. (Informational, not legal advice.)