ShinyaJ/vicuna-7b-scam-detector

Vicuna-7B Scam Detector

A fine-tuned version of Vicuna-7B for detecting scams in Thai text.

Model Details

• Base Model: lmsys/vicuna-7b-v1.5
• Fine-tuned Model: ShinyaJ/vicuna-7b-scam-detector
• Task: Binary Classification (Scam / General)
• Language: Thai

Installation

Install required packages:

pip install transformers peft torch bitsandbytes

Usage

To use the model for scam detection:

1. Import necessary libraries:

import torch
from transformers import LlamaTokenizer, LlamaForSequenceClassification
from peft import PeftModel, PeftConfig

2. Load the model and tokenizer:

model_name = "lmsys/vicuna-7b-v1.5"
peft_model_id = "ShinyaJ/vicuna-7b-scam-detector"

tokenizer = LlamaTokenizer.from_pretrained(model_name)
base_model = LlamaForSequenceClassification.from_pretrained(model_name, num_labels=2, load_in_8bit=True, device_map="auto")
peft_config = PeftConfig.from_pretrained(peft_model_id)
model_scam = PeftModel.from_pretrained(base_model, peft_model_id)

3. Define the prediction function:

def predict_scam(texts, model, tokenizer):
    .
    .
    .

    return ["scam" if pred == 1 else "general" for pred in predicted_classes]

4. Use the model to predict:

text = "Your text here"
prediction = predict_scam(text, model_scam, tokenizer)
print(prediction)

License

Please refer to the license of the base Vicuna-7B model.