---
license: mit
datasets:
- b-mc2/sql-create-context
- gretelai/synthetic_text_to_sql
language:
- en
base_model: google-t5/t5-base
metrics:
- exact_match
model-index:
- name: juanfra218/text2sql
  results:
  - task:
      type: text-to-sql
    metrics:
    - name: exact_match
      type: exact_match
      value: 0.4326836917562724
tags:
- sql
library_name: transformers
---

# Fine-Tuned Google T5 Model for Text to SQL Translation

A fine-tuned version of the Google T5 model, trained for the task of translating natural language queries into SQL statements.

## Model Details

- **Architecture**: Google T5 Base (Text-to-Text Transfer Transformer)
- **Task**: Text to SQL Translation
- **Fine-Tuning Datasets**: 
  - [sql-create-context Dataset](https://huggingface.co/datasets/b-mc2/sql-create-context) 
  - [Synthetic-Text-To-SQL Dataset](https://huggingface.co/datasets/gretelai/synthetic-text-to-sql)

## Ongoing Work

Currently working to implement PICARD (Parsing Incrementally for Constrained Auto-Regressive Decoding from Language Models) to improve the results of this model. More details can be found in the original [PICARD paper](https://arxiv.org/abs/2109.05093).

## Training Parameters

```
training_args = Seq2SeqTrainingArguments(
    output_dir="./results",
    evaluation_strategy="epoch",
    learning_rate=2e-5,
    per_device_train_batch_size=8,
    per_device_eval_batch_size=8,
    weight_decay=0.01,
    save_total_limit=3,
    num_train_epochs=3,
    predict_with_generate=True,
    fp16=True,
    push_to_hub=False,
)
```

## Usage

```
import torch
from transformers import AutoTokenizer, T5ForConditionalGeneration

# Load the tokenizer and model
model_path = 'text2sql_model_path'
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = T5ForConditionalGeneration.from_pretrained(model_path)

# Function to generate SQL queries
def generate_sql(prompt, schema):
    input_text = "translate English to SQL: " + prompt + " " + schema
    inputs = tokenizer(input_text, return_tensors="pt", max_length=512, truncation=True, padding="max_length")

    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    inputs = {key: value.to(device) for key, value in inputs.items()}

    max_output_length = 1024
    outputs = model.generate(**inputs, max_length=max_output_length)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Interactive loop
print("Enter 'quit' to exit.")
while True:
    prompt = input("Insert prompt: ")
    schema = input("Insert schema: ")
    if prompt.lower() == 'quit':
        break

    sql_query = generate_sql(prompt, schema)
    print(f"Generated SQL query: {sql_query}")
    print()
```

## Files

- `optimizer.pt`: State of the optimizer.
- `training_args.bin`: Training arguments and hyperparameters.
- `tokenizer.json`: Tokenizer vocabulary and settings.
- `spiece.model`: SentencePiece model file.
- `special_tokens_map.json`: Special tokens mapping.
- `tokenizer_config.json`: Tokenizer configuration settings.
- `model.safetensors`: Trained model weights.
- `generation_config.json`: Configuration for text generation.
- `config.json`: Model architecture configuration.