FinnAI SLM v3
On-device model for Indian bank SMS → JSON and finance coaching.
Helps expense apps extract transactions from bank SMS privately on the device, and answer simple money questions without cloud LLM calls.
| Model | finndot/finnai-slm-v3 |
| On-device INT4 | finndot/finnai-slm-v3-litertlm |
| Dataset | finndot/finnai-slm-data-v3 |
| Base | Qwen/Qwen3-1.7B |
| License | Apache 2.0 |
FinnAI model family
| Version | Weights | On-device LiteRT | Data |
|---|---|---|---|
| v4 (latest) | finnai-slm-v4 |
finnai-slm-v4-litertlm |
finnai-slm-data-v4 |
| v3 (this) | finnai-slm-v3 |
finnai-slm-v3-litertlm |
finnai-slm-data-v3 |
| v2 | finnai-slm-v2 |
— | finnai-slm-data |
For new work, prefer v4 (more banks, harder SMS, Learn/Analytics chat).
What it helps with
- Parse Indian bank SMS into structured JSON (
amount,merchant,type, …) - Ignore OTP / promo / KYC (
{}) - Short ledger-grounded finance answers
- Run on-device via the LiteRT INT4 package
Quick start
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "finndot/finnai-slm-v3"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id, torch_dtype="bfloat16", device_map="auto"
)
Use enable_thinking=False in the chat template for on-device-style decoding.
Limitations
Optimized for Indian banking SMS; not a general chatbot. See v4 for the latest quality and coverage.
License
Apache 2.0