You can deploy this model and start earning money today!

starwarindia/
addrparser-iaparser

Name: addrparser-iaparser
Availability: InStock

Qwen/Qwen2.5-0.5B-Instruct

adapter

Views

No views yet

AddrParser-Qwen05B (PEFT Adapter)

This repository contains 100+ training checkpoints for a PEFT-finetuned model based on Qwen2.5-0.5B-Instruct.
The model was trained for address parsing (extracting structured address components from free-text input).

📌 The final and recommended checkpoint is checkpoint-108168, but all intermediate checkpoints (checkpoint-1000 → checkpoint-108168) are included for reproducibility and research.

Model Details

Base model: Qwen/Qwen2.5-0.5B-Instruct
Method: PEFT (LoRA adapters)
Language: English + Indian addresses (mixed formatting)
Task: Address parsing (NLP → structured fields)
Author: starwarindia
License: MIT (same as base model unless specified)

Repo Structure

.
├── adapter_config.json
├── adapter_model.safetensors   # adapter weights
├── tokenizer.json / vocab.json # tokenizer files
├── training_args.bin
├── checkpoint-1000/
├── checkpoint-10000/
├── checkpoint-20000/
│   ...
├── checkpoint-108168/          # ✅ final checkpoint

Usage

You can directly load the final checkpoint (recommended):

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = "Qwen/Qwen2.5-0.5B-Instruct"
peft_model = "starwarindia/addrparser-iaparser/checkpoint-108168"

tokenizer = AutoTokenizer.from_pretrained(base_model)
model = AutoModelForCausalLM.from_pretrained(base_model)
model = PeftModel.from_pretrained(model, peft_model)

text = "Flat No 12, Green Park Apartments, MG Road, Bangalore 560001"
inputs = tokenizer(text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

If you want to explore other versions, just change the path (e.g. checkpoint-50000).

Checkpoints

✅ Final: checkpoint-108168
🧪 Intermediate: checkpoint-1000, checkpoint-10000, … checkpoint-107000

Intended Uses

Training analysis (study performance over training steps)
Research in address parsing & sequence tagging
Production use: recommended to use checkpoint-108168

⚠️ Out-of-scope: Not suitable for general-purpose reasoning or unrelated tasks.

Limitations & Risks

May not generalize perfectly on unseen global address formats
Trained primarily on English + Indian addresses
Sensitive to formatting variations (punctuation, missing fields)

Citation

If you use this work, please cite:

bibtex

@misc{addrparser2025,
  title        = {AddrParser-Qwen05B (PEFT Adapter)},
  author       = {starwarindia},
  howpublished = {\url{https://huggingface.co/starwarindia/addrparser-iaparser}},
  year         = {2025}
}

Acknowledgements

Qwen Team for the base model
Hugging Face PEFT library
Google Cloud for training infrastructure