| language: | |
| - zh | |
| license: apache-2.0 | |
| tags: | |
| - roleplay | |
| - character-ai | |
| - taiwanese-mandarin | |
| - sharegpt | |
| task_categories: | |
| - text-generation | |
| # Silicon Girlfriend Dataset | |
| --- | |
| [silicon-based-girlfriend](https://huggingface.co/RX5950XTP/silicon-based-girlfriend) QLoRA 模型的訓練資料集。 | |
| --- | |
| ## Dataset Details / 資料集資訊 | |
| | 項目 | 內容 | | |
| | ------------- | ------------------------------------------ | | |
| | 筆數 | 985 筆 | | |
| | 格式 | ShareGPT(`system` + `conversations`) | | |
| | 語言 | 繁體中文(臺灣用語) | | |
| | 平均對話輪數 | ~10 輪 | | |
| | 最大 Token 數 | 8190 tokens | | |
| | 生成模型 | Kimi K2.5 | | |
| --- | |
| ## Format / 資料格式 | |
| ShareGPT 格式,每筆資料包含: | |
| ```json | |
| { | |
| "system": "角色設定系統提示詞...", | |
| "conversations": [ | |
| {"from": "human", "value": "使用者輸入"}, | |
| {"from": "gpt", "value": "角色回應"} | |
| ] | |
| } | |
| ``` | |
| --- | |
| ## Files / 檔案 | |
| | 檔案 | 說明 | | |
| | ----------------------------- | --------------------------------- | | |
| | `conversations_clean.jsonl` | 清理後的訓練資料(每行一筆 JSON) | | |
| --- | |
| ## License | |
| Apache 2.0 | |