hukum-indo-qa-v1 / README.md
kodetr's picture
Update README.md
6421ced verified
metadata
library_name: transformers
license: mit
language:
  - id
metrics:
  - bleu
  - bertscore
base_model:
  - meta-llama/Llama-3.2-1B
pipeline_tag: text-generation
datasets:
  - kodetr/hukum-indo-qa-v1

Model Description

Konsultasi(Q&A) hukum di indonesia menggunakan ai

  • Developed by: Tanwir
  • Language : Indonesia

Data

Pengambilan data dilakukan dengan teknik scraping di website https://www.hukumonline.com, dari hasil scraping didapatkan data sebanyak 6.965

Training

image/png

Information Result Training

image/png

Evaluation (ROUGE)

  • ROUGE Base Model: {'rouge1': 0.010078955805785689, 'rouge2': 0.0010782076211080282, 'rougeL': 0.008359498894171894, 'rougeLsum': 0.008501491288108246}
  • ROUGE Finetuned Model: {'rouge1': 0.012345676185364394, 'rouge2': 0.0015236094542167386, 'rougeL': 0.010369116143308318, 'rougeLsum': 0.010345376512436985}

BERTSCORE

  • BERTScore Base Model - F1: 0.5445
  • BERTScore Finetuned Model - F1: 0.5520

Use with transformers

Pastikan untuk memperbarui instalasi transformer Anda melalui pip install --upgrade transformer.

import torch
from transformers import pipeline

model_id = "kodetr/hukum-indo-qa-v1"
pipe = pipeline(
    "text-generation",
    model=model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

messages = [
    {"role": "system", "content": "Undang-undang yang sifatnya lex specialis untuk DPRD, apakah UU No.17 Tahun 2014 atau UU No.23 Tahun 2014?"},
    {"role": "user", "content": ""},
]
outputs = pipe(
    messages,
    max_new_tokens=256,
)
print(outputs[0]["generated_text"][-1])