agentlans/thomas-yanxin-MT-SFT-ShareGPT-sample

该数据集提供了[thomas-yanxin/MT-SFT-ShareGPT](https://huggingface.co/datasets/thomas-yanxin/MT-SFT-ShareGPT)数据集的样本，包含英文和中文子集。数据集内容包括：`train.jsonl`（原始数据的1/10，经过洗牌）、`EN.jsonl`（来自`train.jsonl`的英文对话）和`ZH.jsonl`（来自`train.jsonl`的中文对话）。每行代表一个对话，包含可选的系统消息，随后是人类和GPT的对话轮次。原始数据集的列被保留，缺失的列用null值填充。`EN_prompt_downsample`包含带有难度评级的英文提示，中等难度的行限制为70,000行以解决类别不平衡问题。数据按80%训练/20%测试分割，用于模型性能评估和优化。已知问题：在某些数学数据集中，部分多位数数字因类似电话或信用卡号而被编辑，需要时可排除这些行。许可证为Apache 2.0。

This dataset provides a sample of the [thomas-yanxin/MT-SFT-ShareGPT](https://huggingface.co/datasets/thomas-yanxin/MT-SFT-ShareGPT) dataset with English and Chinese subsets. The dataset contents include: `train.jsonl` (1/10 of the original data, shuffled), `EN.jsonl` (English conversations from `train.jsonl`), and `ZH.jsonl` (Chinese conversations from `train.jsonl`). Each row represents a conversation with an optional system message, followed by human and GPT turns. Columns from the original dataset are preserved, with null values for missing columns. `EN_prompt_downsample` contains English prompts with difficulty ratings, with medium difficulty rows limited to 70,000 rows to address class imbalance. The data is split 80% training / 20% test for model performance evaluation and optimization. Known issue: In some math datasets, certain multi-digit numbers have been redacted because they resemble phone or credit card numbers. Exclude those rows if needed. License: Apache 2.0.