Indian Address Parser

A 1.5B model fine-tuned from Qwen2.5-1.5B-Instruct. It turns a messy Indian address into structured JSON: flat, building, street, landmark, area, city, state, pincode, phone.

Scores

Accuracy on real customer addresses has not been measured and is likely lower. The Tamil Nadu set was written by the model's author using the generator's conventions, so it is not an independent benchmark.

Conventions

• city is the district name, or the metro name for big cities.
• area is the locality, village or post-office name. flat is the door or house number only.
• Anything not written in the address stays empty. The state is filled in when it can be inferred from the city or pincode.

Tamil Nadu examples (29 of 30 exactly right)

Inputs are made-up addresses; phone numbers are fake. Empty fields are hidden. The miss is shown in red with what was expected.

Other states (outputs only, no scoring)

Try it yourself

Fastest way, on a laptop with no GPU:

pip install india-address-parser
python -c "from india_address_parser import parse; print(parse('D.No 14/2, Gandhi Road, Pollachi, Coimbatore - 642001'))"

Node.js: npm install india-address-parser (needs Ollama). The model files are on Hugging Face (GGUF). Or load the original weights with transformers:

import json, torch
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "abr-6024/indian-address-parser-1.5B"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, dtype=torch.float16, device_map="auto")

SYSTEM = ("You are an Indian address parser. Convert the address into JSON with exactly these keys: "
          "flat, building, street, landmark, area, city, state, pincode, phone. "
          "Use an empty string for anything missing. Output only JSON.")
msgs = [{"role": "system", "content": SYSTEM},
        {"role": "user", "content": "Flat 3B, Sri Murugan Apartments, Peelamedu, Coimbatore 641004"}]
ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(ids, max_new_tokens=200, do_sample=False)
print(json.loads(tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True)))

Limitations

• Trained on synthetic address text over real India Post geography; real-world accuracy is unmeasured.
• Bare-number addresses are a known weak spot (see the miss above).
• Pincodes are extracted, not validated against the locality. English / Latin script only.