finetune-indoMMLU-Merak-7B-v4

Based on Merak-7B-v4 Mistral: https://huggingface.co/Ichsan2895/Merak-7B-v4
Dataset used on Fine Tuning: https://github.com/fajri91/IndoMMLU

Some training params used:

lora r=64
lora_alpha=16
lora_dropout=0.05

learning_rate = 2e-4
lr_scheduler = "constant"
optimizer = "paged_adamw_32bit"
max_seq_length = 2048

Inference:

import torch
from transformers import AutoTokenizer, AutoConfig, AutoModelForCausalLM, BitsAndBytesConfig, LlamaTokenizer
from peft import PeftModel, PeftConfig

model_name = "Ichsan2895/Merak-7B-v4"
adapter_name = "Willy030125/finetune-indoMMLU-Merak-7B-v4"

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True
)

model = PeftModel.from_pretrained(model, adapter_name)
tokenizer = LlamaTokenizer.from_pretrained(model_name)

def generate_response(question: str) -> str:
    chat = [
      {"role": "system", "content": "Anda adalah Merak, sebuah model kecerdasan buatan yang dilatih oleh Muhammad Ichsan. Mohon jawab pertanyaan berikut dengan benar, faktual, dan ramah."},
      {"role": "user", "content": question},
    ]

    prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
    inputs = tokenizer(prompt, return_tensors="pt", return_attention_mask=True)

    with torch.no_grad():
        outputs = model.generate(input_ids=inputs["input_ids"].to("cuda"),
                           attention_mask=inputs.attention_mask,
                           eos_token_id=tokenizer.eos_token_id,
                           pad_token_id=tokenizer.eos_token_id,
                           max_new_tokens=1024)
        response = tokenizer.batch_decode(outputs.detach().cpu().numpy(), skip_special_tokens=True)[0]

        assistant_start = f'''{question} \n assistant\n '''
        response_start = response.find(assistant_start)
        return response[response_start + len(assistant_start) :].strip()

prompt = """Hewan pemakan tumbuhan dinamakan ...
A. Omnivora
B. Karnivora
C. Pengurai
D. Herbivora"""

print(generate_response(prompt))