---
base_model: chabab/gemma-3-270m-text2sql-oracle-postgres
tags: [gguf, text-to-sql, sql, oracle, postgresql, gemma3]
license: apache-2.0
---

# gemma-3-270m-text2sql-oracle-postgres-GGUF

GGUF builds of [`chabab/gemma-3-270m-text2sql-oracle-postgres`](https://huggingface.co/chabab/gemma-3-270m-text2sql-oracle-postgres) — Gemma-3 270M fine-tuned for
Oracle / PostgreSQL text-to-SQL.

| File | Quant | Size |
|---|---|---|
| `gemma-3-270m-text2sql-f16.gguf` | f16 | 543 MB |
| `gemma-3-270m-text2sql-q8_0.gguf` | q8_0 | 292 MB |

## Ollama

```bash
huggingface-cli download chabab/gemma-3-270m-text2sql-oracle-postgres-GGUF gemma-3-270m-text2sql-q8_0.gguf Modelfile --local-dir .
ollama create text2sql -f Modelfile
ollama run text2sql "Schema:
employees(employee_id INTEGER PK, first_name VARCHAR(50), last_name VARCHAR(50), salary NUMERIC(12,2))

Question:
Show the five employees with the largest salary. Return only the SQL."
```

## llama.cpp

```bash
llama-cli -m gemma-3-270m-text2sql-q8_0.gguf --temp 0 -p "<your schema + question>"
```

Use greedy decoding (`temperature 0`). The model is trained to emit exactly one SQL
statement with no fences or commentary.