Update README.md

README.md

@@ -6,7 +6,6 @@ language:
 - ko
 library_name: transformers
 ---
-
 ## KoRAE
 
 <p align="center"><img src="https://cdn-uploads.huggingface.co/production/uploads/63e087b6a98d931aa90c1b9c/XQ-pNzRDRccd7UFgYDOrx.png", width='300', height='300'></p>
@@ -18,12 +17,13 @@ We utilized special data filtering methods which introduced in [AlpaGasus](https
 We finetuned [Korean Llama-2](https://huggingface.co/beomi/llama-2-koen-13b) that introduced by [@beomi](https://huggingface.co/beomi) on the filtered dataset.
 The Flash-Attention2 and LoRA were utilized for efficient finetuning.
 
-The KoRAE will be uploaded in [Open Ko-LLM Leaderboard](https://huggingface.co/spaces/upstage/open-ko-llm-leaderboard)!
-In addition, the DPO trained version of KoRAE will be uploaded soon too!
-Stay tuned for the update of KoRAE!
+The finding of KoRAE is as follows:
 
-## Model Details
+1. The finetuning in some epochs showed that high-quality filtered data has positive effects on model's performance. However, finetuning in a few epochs, the quantity of data is more matter than quality. It seems to be due to the lack of performance of the Korean base model. Therefore, the research to improve the Korean base model must continue.
+2. The model trained with DPO showed best performance among KoRAE variants. This shows that DPO is clearly effective in the Korean LLM.
+3. The model finetuned with filtered high-quality KoRAE showed better performance than without. Therefore, for better LLM, we should try to finetune the LLM with high-quality data.
 
+## Model Details
 - **Developed by:** [Cartinoe5930](https://huggingface.co/Cartinoe5930)
 - **Base model:** [beomi/llama-2-koen-13b](https://huggingface.co/beomi/llama-2-koen-13b)
 - **Repository:** [gauss5930/KoRAE](https://github.com/gauss5930/KoRAE)
@@ -34,6 +34,7 @@ For more details, please check the GitHub Repository!
 
 - **Hardward:** We utilized A100 80G for finetuning
 - **Training factors:** The [Transformers Trainer](https://huggingface.co/docs/transformers/main_classes/trainer) and [Huggingface PEFT](https://huggingface.co/docs/peft/index) were utilized for finetuning.
+- **Training Details:** Supervised finetuning 1 epoch on [filtered KoRAE](https://huggingface.co/datasets/Cartinoe5930/KoRAE_filtered_12k) dataset
 
 For more details, please check the GitHub Repository!
 
@@ -45,6 +46,10 @@ For more information, please refer to the [dataset card](https://huggingface.co/
 
 ## Open Ko-LLM Leaderboard
 
+|Model|Average|Ko-ARC|Ko-HellaSwag|Ko-MMLU|Ko-TruthfulQA|Ko-CommonGen V2|
+|---|---|---|---|---|---|---|
+|weak-KoRAE-13b|48.1|45.22|56.79|42|40.4|56.08|
+
 ## Prompt Template
 
 ```
@@ -73,7 +78,6 @@ messages = [
     },
     {"role": "user", "content": "스트레스를 해소하는 5가지 방법에 대해서 설명해줘."}
 ]
-
 prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
 outputs = pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
 print(outputs[0]["generated_text"])