synthetic_text_to_sql-ShareGPT

# synthetic_text_to_sql

ShareGPT version of [gretelai/synthetic_text_to_sql](https://huggingface.co/datasets/gretelai/synthetic_text_to_sql) using the following code:

```python
from datasets import load_dataset, DatasetDict

# Load the dataset
dataset = load_dataset('gretelai/synthetic_text_to_sql', split='all')

def format_sample(sample):
conversations = [
{
"from": "human",
"value": f"{sample['sql_context']}\n\n{sample['sql_prompt']}"
},
{
"from": "gpt",
"value": f"{sample['sql']}\n\n{sample['sql_explanation']}"
}
]
return {"conversations": conversations}

dataset = dataset.map(format_sample, remove_columns=dataset.column_names)
dataset = DatasetDict({'train': dataset})
```

It means that the `sql_context` and `sql_prompt` fields are concatenated as user instruction, and the `sql` and `sql_explanation` are concatenated as answers.
Ideally, we'd want to steer the model's answer by providing the explanations first.
However, they're not phrased in a way that would make sense if they appeared before the code, which is why I decided to append them after the `sql` field.

Let me know if you think that another formatting would be better.

# 合成文本转SQL

本实现为基于以下代码的ShareGPT格式版[gretelai/synthetic_text_to_sql](https://huggingface.co/datasets/gretelai/synthetic_text_to_sql)数据集：

python
from datasets import load_dataset, DatasetDict

# 加载数据集
dataset = load_dataset('gretelai/synthetic_text_to_sql', split='all')

def format_sample(sample):
conversations = [
{
"from": "human",
"value": f"{sample['sql_context']}

{sample['sql_prompt']}"
},
{
"from": "gpt",
"value": f"{sample['sql']}

{sample['sql_explanation']}"
}
]
return {"conversations": conversations}

dataset = dataset.map(format_sample, remove_columns=dataset.column_names)
dataset = DatasetDict({'train': dataset})

该实现的拼接逻辑为：将`sql_context`与`sql_prompt`字段拼接作为用户指令，将`sql`与`sql_explanation`字段拼接作为模型回复。

理想情况下，我们希望通过先提供解释来引导模型的输出。但原解释的措辞无法在SQL代码之前合理呈现，因此本次实现将其追加至`sql`字段之后。

若您认为另有更优的格式方案，欢迎告知。