license: apache-2.0
Llama2-13B-LoRA-Toponym-Resolution
This model includes the LoRA (Low-Rank Adaptation) weights fine-tuned for toponym resolution based on the Llama2 13B architecture. It specializes in disambiguating geographic names (toponyms) to provide unambiguous references, often in the form of structured addresses (e.g., city, state, country). For example, given the toponym "Paris", the model may output "Paris, TX, US" based on the context. By further querying geocodes, such as Nominatim, GeoNames, or Google Maps API, the geo-coordinates and other properties, such as population, type can be determined.
Model Summary
- Base Model: Llama2 13B
- Fine-tuning Technique: LoRA (Low-Rank Adaptation)
- Task: Toponym Resolution
- Objective: To identify and resolve toponyms, providing specific addresses or references for geographic names within text.
Usage
This model is ideal for geoparsing and geographic information extraction tasks, especially in:
- Research papers and scientific literature
- Climate impact reports
- Social media and web data related to environmental topics
Loading the Model
You can load this model with its LoRA weights using the transformers and peft libraries as follows:
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
# Load the base model and tokenizer
model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-2-13b-chat-hf")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-13b-chat-hf")
# Load LoRA weights for toponym resolution
model = PeftModel.from_pretrained(model, "xukehu/Llama2-13B-LoRA-Toponym-Resolution")
Citation
If you use this model, please cite the following publication:
@article{hu2024toponym,
title={Toponym resolution leveraging lightweight and open-source large language models and geo-knowledge},
author={Hu, Xuke and Kersten, Jens and Klan, Friederike and Farzana, Sheikh Mastura},
journal={International Journal of Geographical Information Science},
pages={1--28},
year={2024},
publisher={Taylor & Francis}
}