Update README.md

README.md

@@ -31,15 +31,16 @@ A6000 * 4, Deepspeed off-load를 이용해 batch size를 극대화 시켰습니
 - Warmup min LR 1e-6
 - Zero Stage3 off-load
 
-**Perplexity**
+**Perplexity** 법령 데이터를 바탕으로 PPL을 평가했습니다.
 
+- KoSaul-8B - 2.649
+- Open-Llama3-8B (beomi/Llama-3-Open-Ko-8B) - 3.529
+- Open-Llama2-7B (beomi/llama-2-ko-7b) - 3.393
 - Solar-10.7B (chihoonlee10/T3Q-ko-solar-dpo-v1.0) - 3.161
 - EEVE-10.8B (yanolja/EEVE-Korean-Instruct-10.8B-v1.0) - 3.505
 - KULLM3 (nlpai-lab/KULLM3) - 2.903
 - MLP-KTLim (MLP-KTLim/Bllossom) - 4.385
-- Open-Llama2-7B (beomi/llama-2-ko-7b) - 3.393
-- Open-Llama3-8B (beomi/Llama-3-Open-Ko-8B) - 3.529
-- KoSaul-8B - 2.649
+
 
 **Model Architecture** Llama 3 is an auto-regressive language model