import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer

MODEL_DIR = "chihoonlee10/T3Q-LLM-MG-v1.0"
model = AutoModelForCausalLM.from_pretrained(MODEL_DIR, torch_dtype=torch.float16).to("cuda")
tokenizer = AutoTokenizer.from_pretrained(MODEL_DIR)
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

s = "한국의 수도는 어디?"
conversation = [{'role': 'user', 'content': s}]
inputs = tokenizer.apply_chat_template(
    conversation,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors='pt').to("cuda")
_ = model.generate(inputs, streamer=streamer, max_new_tokens=1024)

hf (pretrained=chihoonlee10/T3Q-LLM-MG-v1.0), limit: None, provide_description: False, num_fewshot: 0, batch_size: None

Task	Version	Metric	Value		Stderr
kobest_boolq	0	acc	0.9523	±	0.0057
		macro_f1	0.9523	±	0.0057
kobest_copa	0	acc	0.7740	±	0.0132
		macro_f1	0.7737	±	0.0133
kobest_hellaswag	0	acc	0.4980	±	0.0224
		acc_norm	0.5920	±	0.0220
		macro_f1	0.4950	±	0.0223
kobest_sentineg	0	acc	0.7254	±	0.0224
		macro_f1	0.7106	±	0.0234

T3Q-LLM/T3Q-LLM-sft1.0-dpo1.0

Task	Version	Metric	Value		Stderr
kobest_boolq	0	acc	0.9387	±	0.0064
		macro_f1	0.9387	±	0.0064
kobest_copa	0	acc	0.7590	±	0.0135
		macro_f1	0.7585	±	0.0135
kobest_hellaswag	0	acc	0.5080	±	0.0224
		acc_norm	0.5580	±	0.0222
		macro_f1	0.5049	±	0.0224
kobest_sentineg	0	acc	0.8489	±	0.0180
		macro_f1	0.8483	±	0.0180