Salesforce/GIFT-CTX

该仓库以parquet格式托管GIFT-EvalCTX数据集，高度兼容LLMs。每行数据包含以下字段：idx（样本唯一ID）、source（来源）、skill（上下文类型）、frequency（样本频率）、history_values（历史值字符串）、history_start（历史开始时间戳）、history_end（历史结束时间戳）、future_values（预测未来值字符串）、future_start（未来开始时间戳）、future_end（未来结束时间戳）、entry_sep（分隔符）、roi（关注区域）、pred_length（预测时间戳数量）、system_prompt（默认系统提示）、user_instruct（预测查询）、context_info（当前样本上下文）、prompt（完整查询提示）。所有字段均为字符串类型，需转换为适当格式（浮点数数组或日期时间）。该数据集仅用于研究目的。

This repository hosts the GIFT-EvalCTX dataset in parquet format, highly compatible with LLMs. Each row is a sample of the dataset and contain the following fields: idx (unique id of the sample), source (source of the same), skill (the types of context), frequency (frequency of the sample), history_values (a string containing values of the history), history_start (starting timestamp of the history), history_end (ending timestamp of the history), future_values (a string containing values of the forecasting future), future_start (starting timestamp of the future), future_end (ending timestamp of the future), entry_sep (separator used to split the string in history_values, future_values, and roi into array of floats), roi (region of interest, indicating the indices of the timestamp that are affected by the context and used in evaluation), pred_length (number of timestamps to forecast), system_prompt (default system prompt), user_instruct (query to forecast), context_info (context for the current sample), prompt (a complete prompt to query LLM). Note that all fields contain string only, you need to convert them into the appropriate format (array of floats, or datetime). This repository is made public for research purposes only.