README.md

---
library_name: onnx
tags:
  - text2text-generation
  - t5
  - text-to-sql
  - sql
  - spider
  - encoder-decoder
  - onnx
  - inference4j
license: apache-2.0
pipeline_tag: text2text-generation
datasets:
  - spider
  - spider-syn
---

# T5-LM-Large text2sql-spider — ONNX

ONNX export of [T5-LM-Large-text2sql-spider](https://huggingface.co/gaussalgo/T5-LM-Large-text2sql-spider) (0.8B parameters) with encoder-decoder architecture and KV cache support.

This is a T5-large model fine-tuned on the Spider and Spider-Syn datasets for text-to-SQL generation. Given a natural language question and a database schema, it produces the corresponding SQL query.

Converted for use with [inference4j](https://github.com/inference4j/inference4j), an inference-only AI library for Java.

## Original Source

- **Repository:** [gaussalgo/T5-LM-Large-text2sql-spider](https://huggingface.co/gaussalgo/T5-LM-Large-text2sql-spider)
- **Base model:** [google/t5-large-lm-adapt](https://huggingface.co/google/t5-large-lm-adapt)
- **License:** Apache 2.0

## Usage with inference4j

```java
try (var sqlGen = T5SqlGenerator.t5LargeSpider().build()) {
    String sql = sqlGen.generateSql(
        "How many employees are in each department?",
        "\"employees\" \"id\" int, \"name\" varchar, \"dept_id\" int "
        + "[SEP] \"departments\" \"id\" int, \"name\" varchar");
    System.out.println(sql);
}
```

## Schema Format

The model expects the schema in the following format:

```
"table_name" "col1" type, "col2" type, foreign_key: "table"."col" = "other"."col" primary key: "col" [SEP] "table2" ...
```

- Table and column names are double-quoted
- Columns are comma-separated with types
- Tables are separated by `[SEP]`
- Foreign keys and primary keys are declared per table

## Model Details

| Property | Value |
|----------|-------|
| Architecture | T5 encoder-decoder (0.8B parameters) |
| Task | Text-to-SQL generation |
| Training data | Spider, Spider-Syn |
| Tokenizer | SentencePiece (32,128 tokens) |
| Original framework | PyTorch (transformers) |
| Export method | Hugging Face Optimum (encoder-decoder with KV cache) |

## License

This model is licensed under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0). Original model by [Gaussalgo](https://huggingface.co/gaussalgo), base model by [Google](https://huggingface.co/google).