BCCard/gemma-4-26B-A4B-korean-on-policy-150k
收藏资源简介:
--- 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.)
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. The dataset contains approximately 150,000 rows with columns: instruction (str), output (str, verifier-generated), messages (chat list). It was created by sampling the instruction column from the source dataset `sh2orc/bccard-maywell-jojo0217-markai-lcw99-kendamarron-microsoft` (about 150k rows), discarding original answers, and regenerating responses using the specified model with thinking off, text-only mode. Licensed under Apache 2.0, with synthetic responses generated by Gemma 4 and source-dataset attribution retained.




