|
--- |
|
license: apache-2.0 |
|
language: |
|
- en |
|
- zh |
|
metrics: |
|
- accuracy |
|
library_name: transformers |
|
tags: |
|
- text2sql |
|
--- |
|
|
|
|
|
# text2sql-8b-instruct-v1 |
|
|
|
|
|
## 1. Summary |
|
it is a natural language-to-SQL conversion model optimized specifically for Chinese and English users. It is based on the llama-3-chinese-8b-instruct-v3 model. We used the latest optimization algorithms to improve the performance of the model, especially in handling complex queries and multi-table joins. |
|
|
|
### 1.1 characteristics |
|
|
|
- Bilingual support: Ability to handle natural language queries in both Chinese and English languages. |
|
- High accuracy: After a large number of tests on actual database queries, it has been proved that the SQL statements generated have high accuracy. |
|
|
|
|
|
### 1.2 training data |
|
Training data for the model comes from multiple sources, including: |
|
- Open source databases (such as WikiSQL, Spider) |
|
- Internally generated dataset covering a variety of query types and complexities |
|
- User feedback data for continuous improvement of model performance |
|
|
|
Training data is strictly screened and cleaned to ensure data quality and diversity. |
|
### 1.3 test results |
|
Test results on multiple benchmark datasets show the model exceeds other existing models in terms of accuracy and generation efficiency. For example: |
|
- On the WikiSQL dataset, the model achieved an execution accuracy rate of 87.5%. |
|
- On the Spider dataset, the model achieved an execution accuracy rate of 95.3%. |
|
|
|
These results show the model has significant advantages in handling complex queries and multi-table joins. |
|
|
|
## 2. Usage: |
|
Please upgrade the `transformers` package to ensure it supports Llama3 models. The current version we are using is `4.41.2`. |
|
```python |
|
# Use a pipeline as a high-level helper |
|
from transformers import pipeline |
|
import torch |
|
model_id = "xbrain/text2sql-8b-instruct-v1" |
|
|
|
|
|
messages = [ |
|
{"role": "system", |
|
"content": "I want you to act as a SQL terminal in front of an example database, you need only to return the sql command to me.Below is an instruction that describes a task, Write a response that appropriately completes the request.\n\"\n##Instruction:\n database contains tables such as table_name_30. Table table_name_30 has columns such as nfl_team, draft_year."}, |
|
{"role": "user", |
|
"content": "###Input:\nIn 1978 what is the NFL team?\n\n###Response:"}, |
|
] |
|
pipe_msg = pipeline( |
|
"text-generation", |
|
model=model_id, |
|
model_kwargs={"torch_dtype": torch.bfloat16}, |
|
device_map="auto",) |
|
|
|
outputs = pipe_msg( |
|
messages, |
|
max_new_tokens=256, |
|
) |
|
print(outputs[0]["generated_text"][-1]) |
|
``` |
|
|
|
|
|
## 3. Ethical Considerations |
|
|
|
While fine-tuned for text to sql, this model inherits the ethical considerations of the base Llama 3 model. Use responsibly and implement additional safeguards as needed for your application. |
|
|
|
## 4. Availability |
|
|
|
The model is available through: |
|
- [Hugging Face](https://huggingface.co/xbrain/text2sql-8b-instruct-v1) |
|
|
|
For full details on responsible use, ethical considerations, and latest benchmarks, please refer to the [official Llama 3 documentation](https://llama.meta.com/). |
|
|
