banking-intent-unsloth-checkpoint

Banking Intent Classifier — Llama 3 8B (LoRA)

A LoRA adapter fine-tuned on top of Llama 3 8B Instruct (4-bit) for banking customer intent classification using the BANKING77 dataset.

Model Details

• Base model: unsloth/llama-3-8b-Instruct-bnb-4bit
• Model type: Causal LM + LoRA adapter (PEFT)
• Task: Intent classification (77 classes)
• Language: English
• Fine-tuned by: nguyenvmthien
• Repository: nguyenvmthien/banking-intent-unsloth
• Framework: Unsloth + TRL (SFT) + PEFT 0.18.1

Training Details

Dataset

Sampled from the BANKING77 dataset — 77 fine-grained intent categories for banking customer service queries.

Hyperparameters

Prompt Format

Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
Classify the intent of the following banking customer request.

### Input:
{customer_message}

### Response:

Usage

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="minhthien/banking-intent-unsloth-checkpoint",
    max_seq_length=256,
    dtype=None,
    load_in_4bit=True,
)
FastLanguageModel.for_inference(model)

Or clone the full inference/evaluation code:

git clone https://github.com/nguyenvmthien/banking-intent-unsloth.git
cd banking-intent-unsloth
pip install -r requirements.txt
python scripts/evaluate.py

Evaluation

Evaluation uses log-probability scoring over all 77 candidate intent labels (no text generation needed), which is faster and more deterministic.

Results will be updated after full evaluation run.