| |
| """Inference script for Faruq IndoLaw Alignment Models.""" |
|
|
| import argparse |
| import torch |
| from peft import PeftModel |
| from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
|
|
| def main() -> None: |
| parser = argparse.ArgumentParser(description="Run inference with Faruq IndoLaw models from Hugging Face.") |
| parser.add_argument("--repo-id", default="ruwwww/faruq-alignment-models", help="Hugging Face repo ID") |
| parser.add_argument("--subfolder", default="models/sft-v2-5epoch", help="Model subfolder within repository") |
| parser.add_argument("--base-model", default="Qwen/Qwen2.5-1.5B", help="Base model ID") |
| parser.add_argument("--max-new-tokens", type=int, default=512, help="Maximum generated tokens") |
| args = parser.parse_args() |
|
|
| print(f"Loading tokenizer & adapter from {args.repo_id} [{args.subfolder}]...") |
| tokenizer = AutoTokenizer.from_pretrained(args.repo_id, subfolder=args.subfolder) |
| base_model = AutoModelForCausalLM.from_pretrained( |
| args.base_model, |
| torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32, |
| device_map="auto" if torch.cuda.is_available() else None, |
| ) |
| model = PeftModel.from_pretrained(base_model, args.repo_id, subfolder=args.subfolder) |
| model.eval() |
|
|
| messages = [ |
| { |
| "role": "system", |
| "content": ( |
| "Anda adalah asisten hukum Indonesia. Gunakan hanya dakwaan dan fakta persidangan yang diberikan. " |
| "Susun pertimbangan hukum yang ringkas dan amar putusan secara setia pada sumber." |
| ), |
| }, |
| { |
| "role": "user", |
| "content": ( |
| "[SURAT DAKWAAN]\n" |
| "Bahwa terdakwa didakwa melanggar Pasal 362 KUHP tentang Pencurian karena telah mengambil satu unit sepeda motor milik korban tanpa izin.\n\n" |
| "[FAKTA PERSIDANGAN & ALAT BUKTI]\n" |
| "Saksi menerangkan melihat terdakwa mendorong sepeda motor korban di malam hari. " |
| "Barang bukti berupa 1 unit sepeda motor dan kunci kontak diamankan. Terdakwa mengakui perbuatannya di depan persidangan." |
| ), |
| }, |
| ] |
|
|
| prompt = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False) |
| inputs = tokenizer(prompt, return_tensors="pt").to(model.device) |
|
|
| print("\nGenerating court ratio decidendi and dictum...\n") |
| with torch.no_grad(): |
| outputs = model.generate( |
| **inputs, |
| max_new_tokens=args.max_new_tokens, |
| do_sample=False, |
| repetition_penalty=1.1, |
| eos_token_id=tokenizer.eos_token_id, |
| pad_token_id=tokenizer.pad_token_id, |
| ) |
|
|
| completion = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True) |
| print("=== MODEL OUTPUT ===") |
| print(completion) |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
