You can deploy this model and start earning money today!

sumanbarik437/
falcon-bank77-qlora

Name: falcon-bank77-qlora
Availability: InStock

Views

No views yet

Falcon-1B QLoRA Fine-Tuned on Banking77 (Intent Classification)

This model is a Falcon-RW-1B LLM fine-tuned using QLoRA for intent classification on the Banking77 dataset.
It can classify 77 different banking-related customer intents, making it suitable for chatbots, virtual assistants, and banking NLP applications.

The model is lightweight, fast, and optimized for deployment in real-world BFSI (Banking, Financial Services, Insurance) systems.

Model Details

Model Description

Developed by: Suman Barik
Task: Intent Classification (77 classes)
Languages: English
Fine-tuned from: tiiuae/falcon-rw-1b
Dataset: Banking77
Technique: QLoRA (Parameter-Efficient Fine-Tuning)
Model Type: Causal Language Model adapted for sequence classification
License: Refer to Falcon-RW-1B license
Framework: HuggingFace Transformers

Model Sources

Repository: This Hugging Face model page
Dataset: https://huggingface.co/datasets/banking77
Base Model: https://huggingface.co/tiiuae/falcon-rw-1b

Uses

Direct Use

Classification of customer support queries
Banking domain intent identification
Routing user messages in conversational AI systems
Building intelligent chatbots or IVR systems
Fine-grained understanding of user requests in fintech platforms

Example use cases:

“I want to reset my card PIN” → card_arrival
“My transfer is pending” → transfer_pending
“What is my account balance?” → balance

Downstream Use

As a component inside:
- RAG-based chatbots
- Banking AI assistants
- Call center automation
- Complaint categorization systems
- Customer query triage pipelines

Out-of-Scope Use

Generating free-form text
Non-English tasks
Safety-critical decision-making
Using model outputs as financial advice

Bias, Risks, and Limitations

Trained solely on Banking77 → may not generalize outside banking domain
Limited to English queries
May misclassify highly ambiguous queries
Not a substitute for a compliance-checked financial system

Recommendations

Use confidence thresholds
Perform human validation in production-critical systems
Fine-tune further on organization-specific datasets

How to Get Started with the Model

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "sumanbarik437/falcon-bank77-qlora"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "How do I reset my card PIN?"
inputs = tokenizer(text, return_tensors="pt")

with torch.no_grad():
    logits = model(**inputs).logits

pred = torch.argmax(logits, dim=-1).item()
print("Predicted class ID:", pred)