app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM
from example_strings import example1, example2, example3


template_str = """{table_schemas}
\n \n
{task_spec}
\n \n
{prompt}
\n \n
SELECT"""


def load_model(model_name: str):
    tokenizer = AutoTokenizer.from_pretrained(f"NumbersStation/{model_name}")
    model = AutoModelForCausalLM.from_pretrained(f"NumbersStation/{model_name}")
    return tokenizer, model


def build_complete_prompt(table_schemas: str, task_spec: str, prompt: str) -> str:
    return template_str.format(table_schemas=table_schemas, task_spec=task_spec, prompt=prompt)


def infer(table_schemas: str, task_spec: str, prompt: str, model_choice: str = "nsql-350M"):
    tokenizer, model = load_model(model_choice)
    
    input_text = build_complete_prompt(table_schemas, task_spec, prompt)
    
    input_ids = tokenizer(input_text, return_tensors="pt").input_ids
    generated_ids = model.generate(input_ids, max_length=500)
    return (tokenizer.decode(generated_ids[0], skip_special_tokens=True))


description = """The NSQL model family was published by [Numbers Station](https://www.numbersstation.ai/) and is available in three flavors:  
- [nsql-6B](https://huggingface.co/NumbersStation/nsql-6B)
- [nsql-2B](https://huggingface.co/NumbersStation/nsql-2B)
- [nsql-350M]((https://huggingface.co/NumbersStation/nsql-350M))  

This demo let's you choose from all of them and provides the three examples you can also find in their model cards.  

In general you should first provide the table schemas of the tables you have questions about and then prompt it with a natural language question.  
The model will then generate a SQL query that you can run against your database.
"""

iface = gr.Interface(
    title="Text to SQL with NSQL",
    description=description,
    fn=infer, 
    inputs=[gr.Text(label="Table schemas", placeholder="Insert your table schemas here"),
            gr.Text(label="Specify Task", value="Using valid SQLite, answer the following questions for the tables provided above."), 
            gr.Text(label="Prompt", placeholder="Put your natural language prompt here"), 
            gr.Dropdown(["nsql-6B", "nsql-2B", "nsql-350M"], value="nsql-6B")
            ], 
    outputs="text",
    examples=[example1, example2, example3])
iface.launch()