ssnym/llama-3.2-1b-text-to-sql-GGUF

Llama-3.2-1B Text-to-SQL — GGUF (LoRA adapter)

GGUF conversion of the LoRA adapter from ssnym/llama-3.2-1b-text-to-sql, for use with llama.cpp.

This repo contains only the adapter, converted to GGUF (f16) via llama.cpp's convert_lora_to_gguf.py. You'll need the base model's GGUF separately — see below.

Release date: September 12, 2026

Files

• —text-to-sql-adapter-f16.gguf — LoRA adapter (f16), ~22.5MB

Usage

1. 1.Download a GGUF build of the base model, e.g. unsloth/Llama-3.2-1B-Instruct-GGUF. Verified working with Q4_K_M.
2. 2.Download text-to-sql-adapter-f16.gguf from this repo.
3. 3.Run with llama.cpp:

llama-cli -hf unsloth/Llama-3.2-1B-Instruct-GGUF:Q4_K_M \
  --lora text-to-sql-adapter-f16.gguf \
  -p "Given the table employees(id, name, department, salary), write a SQL query to find the average salary per department."

Or with a locally converted/downloaded base GGUF file:

llama-cli -m base-model.gguf --lora text-to-sql-adapter-f16.gguf -p "..."

You can also serve it via llama-server (OpenAI-compatible API + web UI):

llama-server -m base-model.gguf --lora text-to-sql-adapter-f16.gguf --port 8080

Why a separate adapter file instead of a merged GGUF?

Keeping the adapter separate means the same ~22.5MB file can be applied on top of any GGUF build of the base model (any quant level, from anyone) without re-downloading or re-quantizing the full base model each time — the same tradeoff as the original safetensors adapter, just carried through to the GGUF/llama.cpp side.

Original model

See ssnym/llama-3.2-1b-text-to-sql for training details, evaluation results, and the safetensors/PEFT version of this adapter.

License

Built with Llama. Released under the Llama 3.2 Community License, inherited from the base model.