TableLLM: Enabling Tabular Data Manipulation by LLMs in Real Office Usage Scenarios
| Paper | Training set | Github | Homepage |
We present TableLLM, a powerful large language model designed to handle tabular data manipulation tasks efficiently, whether they are embedded in spreadsheets or documents, meeting the demands of real office scenarios. The TableLLM series encompasses two distinct scales: TableLLM-7B and TableLLM-13B, which are fine-tuned based on CodeLlama-7b-Instruct-hf and CodeLlama-13b-Instruct-hf.
TableLLM generates either a code solution or a direct text answer to handle tabular data manipulation tasks based on different scenarios. Code generation is used for handling spreadsheet-embedded tabular data, which often involves the insert, delete, update, query, merge, and plot operations of tables. Text generation is used for handling document-embedded tabular data, which often involves the query operation of short tables.
Evaluation Results
We evaluate the code solution generation ability of TableLLM on three benchmarks: WikiSQL, Spider and Self-created table operation benchmark. The text answer generation ability is tested on four benchmarks: WikiTableQuestion (WikiTQ), TAT-QA, FeTaQA and OTTQA. The evaluation result is shown below:
| Model | WikiTQ | TAT-QA | FeTaQA | OTTQA | WikiSQL | Spider | Self-created | Average |
|---|---|---|---|---|---|---|---|---|
| TaPEX | 38.5 | – | – | – | 83.9 | 15.0 | / | 45.8 |
| TaPas | 31.5 | – | – | – | 74.2 | 23.1 | / | 42.92 |
| TableLlama | 24.0 | 22.2 | 20.5 | 6.4 | 43.7 | 9.0 | / | 20.7 |
| GPT3.5 | 58.5 | 72.1 | 71.2 | 60.8 | 81.7 | 67.4 | 77.1 | 69.8 |
| GPT4 | 74.1 | 77.1 | 78.4 | 69.5 | 84.0 | 69.5 | 77.8 | 75.8 |
| Llama2-Chat (13B) | 48.8 | 49.6 | 67.7 | 61.5 | – | – | – | 56.9 |
| CodeLlama (13B) | 43.4 | 47.2 | 57.2 | 49.7 | 38.3 | 21.9 | 47.6 | 43.6 |
| Deepseek-Coder (33B) | 6.5 | 11.0 | 7.1 | 7.4 | 72.5 | 58.4 | 73.9 | 33.8 |
| StructGPT (GPT3.5) | 52.5 | 27.5 | 11.8 | 14.0 | 67.8 | 84.8 | / | 48.9 |
| Binder (GPT3.5) | 61.6 | 12.8 | 6.8 | 5.1 | 78.6 | 52.6 | / | 42.5 |
| DATER (GPT3.5) | 53.4 | 28.4 | 18.3 | 13.0 | 58.2 | 26.5 | / | 37.0 |
| TableLLM-7B (Ours) | 58.8 | 66.9 | 72.6 | 63.1 | 86.6 | 82.6 | 78.8 | 72.8 |
| TableLLM-13B (Ours) | 62.4 | 68.2 | 74.5 | 62.5 | 90.7 | 83.4 | 80.8 | 74.7 |
Prompt Template
The prompts we used for generating code solutions and text answers are introduced below.
Code Solution
The prompt template for the insert, delete, update, query, and plot operations on a single table.
[INST]Below are the first few lines of a CSV file. You need to write a Python program to solve the provided question.
Header and first few lines of CSV file:
{csv_data}
Question: {question}[/INST]
The prompt template for the merge operation on two tables.
[INST]Below are the first few lines two CSV file. You need to write a Python program to solve the provided question.
Header and first few lines of CSV file 1:
{csv_data1}
Header and first few lines of CSV file 2:
{csv_data2}
Question: {question}[/INST]
The csv_data field is filled with the first few lines of your provided table file. Below is an example:
Sex,Length,Diameter,Height,Whole weight,Shucked weight,Viscera weight,Shell weight,Rings
M,0.455,0.365,0.095,0.514,0.2245,0.101,0.15,15
M,0.35,0.265,0.09,0.2255,0.0995,0.0485,0.07,7
F,0.53,0.42,0.135,0.677,0.2565,0.1415,0.21,9
M,0.44,0.365,0.125,0.516,0.2155,0.114,0.155,10
I,0.33,0.255,0.08,0.205,0.0895,0.0395,0.055,7
Text Answer
The prompt template for direct text answer generation on short tables.
[INST]Offer a thorough and accurate solution that directly addresses the Question outlined in the [Question].
### [Table Text]
{table_descriptions}
### [Table]
```
{table_in_csv}
```
### [Question]
{question}
### [Solution][INST/]
For more details about how to use TableLLM, please refer to our GitHub page: https://github.com/TableLLM/TableLLM