CodeStrux-Tech/tac-1-lora

1---2license: apache-2.03language:4  - es5library_name: peft6base_model: unsloth/qwen3-4b-instruct-2507-unsloth-bnb-4bit7pipeline_tag: text-generation8tags:9  - lora10  - qlora11  - sft12  - trl13  - unsloth14  - address-parsing15  - costa-rica16  - qwen317---18 19# tac-1-lora — QLoRA adapter for tac-120 21## Overview22 23This is the QLoRA adapter that produced [`CodeStrux-Tech/tac-1`](https://huggingface.co/CodeStrux-Tech/tac-1). **Most users want the merged tac-1 repo**, not these adapter weights. Use this repo only if you need to inspect or extend the adapter directly.24 25## Loading with PEFT26 27PEFT loading requires the base model `unsloth/qwen3-4b-instruct-2507-unsloth-bnb-4bit`.28 29## Training configuration30 31- **Base model:** Qwen/Qwen3-4B-Instruct-250732- **Method:** QLoRA (r=16, α=32)33- **Learning rate:** 2e-434- **Epochs:** 235- **Max sequence length:** 409636- **Steps:** 69237- **Final train loss:** 0.04338- **Hardware:** ~2 h 50 m on an RTX 4080 16 GB39- **Stack:** unsloth 2025.11.1 / transformers 4.57.2 / trl 0.23.040 41## Training data42 43The adapter was trained on the tac-1 corpus: 5,532 examples (seed 0), 805 heldout (seed 1); `--max-legs 4`; 22 districts ingested, 19,042 POIs, 11 griddable; holdout districts grecia, curridabat, go-guadalupe excluded from training. See [`CodeStrux-Tech/tac-1-corpus`](https://huggingface.co/CodeStrux-Tech/tac-1-corpus).44 45### Training data attribution46 47Contains information from OpenStreetMap (https://www.openstreetmap.org/copyright), which is made available under the Open Database License (ODbL) 1.0. © OpenStreetMap contributors.48 49For full architecture, evaluation, and limitations, see [`CodeStrux-Tech/tac-1`](https://huggingface.co/CodeStrux-Tech/tac-1).50 51tac-1 is a derivative work of [Qwen/Qwen3-4B-Instruct-2507](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507), Copyright 2024 Alibaba Cloud, licensed under the Apache License, Version 2.0. The upstream LICENSE is included in this repository.52