Update README.md

README.md

@@ -6,30 +6,50 @@ license: apache-2.0
   <img src="https://github.com/MLP-Lab/KORMo-tutorial/blob/main/tutorial/attachment/kormo_logo.png?raw=true" style="width: 100%; max-width: 1100px;">
 </p>
 
-# 🦾 KORMo-10B
 
+
+## 🆕 Update News
+- 🚀 **2025-10-09 (한글날)**: Official release of KORMo-10B-base (Be aware that it's not an SFT model!!).
+---
+## 🦾 About KORMo
 **KORMo-10B** is a **10.8B parameter fully open LLM** capable of handling both **Korean and English**.  
-The model, training code, and training data are all **fully open**, allowing anyone to reproduce and extend it.
+The model, training code, and training data are all **fully open**, allowing anyone to reproduce and extend them.
 
 - 🧠 **Model Size**: 10.8B parameters  
 - 🗣️ **Languages**: Korean / English  
 - 🪄 **Training Data**: Synthetic data + public datasets  
 - 🧪 **License**: Apache 2.0 (commercial use permitted)
 
+```bash
+KORMo는 비영어권 최초의 Fully Open Source LLM으로, 공익적 활용을 목표로 탄생했습니다.
+우리는 누구나 세계 수준의 언어모델을 직접 만들고 발전시킬 수 있는 환경을 만들고자 합니다.
+KORMo의 주요 특징은 다음과 같습니다:
+
+1. From scratch 학습으로 설계된 10B급 한–영 추론 언어모델입니다.
+2. 학습 데이터, 코드, 모든 중간 모델과 튜토리얼을 100% 공개하여, 누구나 SOTA에 근접한 모델을 직접 재현하고 확장할 수 있습니다.
+3. 총 3.7T 토큰 규모의 학습 데이터를 공개합니다. 특히 지금까지 한 번도 공개된 적 없는 초고품질 전주기 한국어 데이터(사전학습, 사후학습, 일반형, 추론형, 강화학습 등)를 제공합니다.
+4. 이 모든 작업은 KAIST 문화기술대학원 MLP연구실의 학부·석사생 8명이 협력하여 진행했으며, 45장에 달하는 논문으로 정리했습니다.
+
+지금까지 한국어 모델을 써보면, 벤치마크 점수는 좋은데 실사용에서는 어딘가 이상하거나,
+튜닝만 하면 모델이 망가지는 경험을 하셨을 겁니다. 답답하셨죠?
+
+KORMo는 그런 문제를 정면으로 해결합니다.
+모든 중간 모델과 사후학습 데이터를 함께 공개하기 때문에, 사용자는 베이스 모델 위에 자신만의 데이터를 얹어 원하는 방향으로 강화학습·튜닝을 진행할 수 있습니다.
+👉 “좋은 한국어 모델을 갖고 싶다면, 이제 직접 만들어보세요. 코랩 무료 GPU로도 튜닝됩니다! 🤗”
+```
+
 ---
 
 ## 🔗 Links
 
+- 📖 **Technical Report**: [👉 Archive](https://huggingface.co/KORMo-Team)  
 - 🤗 **Hugging Face**: [👉 Model Download](https://huggingface.co/KORMo-Team)  
 - 💻 **GitHub Repository**: [👉 Training and Inference Code](https://github.com/MLP-Lab/KORMo-tutorial)
+- 🔉 **Tutorial**: [👉 Instruction Tuning over google colab](https://colab.research.google.com/github/MLP-Lab/KORMo-tutorial/blob/main/tutorial/02.sft_qlora.ipynb) [👉 Youtube Tutorial](https://www.youtube.com/@MLPLab)
 
 ---
 
-## 🆕 Update News
-- 🚀 **Oct 2025**: Official release of KORMo v1.0!
-
----
-
+<!--
 ## Model Architecture
 | Item | Description |
 |:----|:------------|
@@ -38,7 +58,7 @@ The model, training code, and training data are all **fully open**, allowing any
 | Context Length | 32K |
 | Languages | Korean, English |
 | License | Apache 2.0 |
-
+-->
 ---
 
 ## 📈 Benchmark Performance
@@ -74,11 +94,10 @@ The model, training code, and training data are all **fully open**, allowing any
 | kr_clinical_qa | 77.32 | 53.97 | 48.33 | 46.22 | 65.84 | 80.00 | 63.54 | 60.00 | 77.22 |
 | **Korean Avg.** | **58.15** | 47.37 | 35.82 | 39.34 | 60.94 | 63.35 | 49.60 | 49.60 | 60.37 |
 
----
 
-## 📝 Qualitative Evaluation (LLM-as-a-Judge)
+### 📝 Qualitative Evaluation (LLM-as-a-Judge)
 
-| Benchmark | KORMo-10B | smolLM3-3B | olmo2-7B | olmo2-13B | kanana1.5-8B | qwen3-8B | llama3.1-8B | exaone3.5-8B* | gemma3-12B |
+| Benchmark | KORMo-10B | smolLM3-3B | olmo2-7B | olmo2-13B | kanana1.5-8B | qwen3-8B | llama3.1-8B | exaone3.5-8B | gemma3-12B |
 |:----------|---------:|----------:|---------:|---------:|------------:|--------:|------------:|-------------:|-----------:|
 | MT-Bench (EN) | 8.32 | 7.15 | 7.32 | 7.64 | 8.45 | 8.70 | 6.32 | 8.15 | 8.70 |
 | KO-MT-Bench (KO) | 8.54 | - | - | - | 8.02 | 8.16 | 4.27 | 8.13 | 8.51 |
@@ -87,8 +106,123 @@ The model, training code, and training data are all **fully open**, allowing any
 
 ---
 
+## 📦 Installation
+
+### 1. Clone the repository
+```bash
+git clone https://github.com/MLP-Lab/KORMo-tutorial.git
+cd KORMo-tutorial
+```
+### 2. Create and activate a virtual environment (optional but recommended)
+```bash
+uv venv
+source .venv/bin/activate  
+```
+### 3. Install KORMo
+```bash
+uv pip install -e .
+```
+
+---
+## 🚀 Inference Example
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+model_name = "KORMo-Team/KORMo-10B-sft"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    trust_remote_code=True
+)
+
+messages = [
+    {"role": "user", "content": "What happens inside a black hole?"}
+]
+
+chat_prompt = tokenizer.apply_chat_template(
+    messages,
+    tokenize=False,
+    add_generation_prompt=True,
+    enable_thinking=False
+)
+
+inputs = tokenizer(chat_prompt, return_tensors="pt").to(model.device)
+
+with torch.no_grad():
+    output_ids = model.generate(
+        **inputs,
+        max_new_tokens=1024,
+    )
+
+response = tokenizer.decode(output_ids[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
+print("Assistant:", response)
+```
+
+## 🧠 Enabling Thinking Mode
+
+If you want to enable the **thinking** mode, simply set `enable_thinking=True`:
+
+```python
+chat_prompt = tokenizer.apply_chat_template(
+    messages,
+    tokenize=False,
+    add_generation_prompt=True,
+    enable_thinking=True
+)
+```
+---
+
+## 🪄 Using Specific Revisions (Training Checkpoints)
+
+KORMo provides multiple model revisions corresponding to different training stages and checkpoints.  
+You can load a specific revision with the `revision` parameter in `from_pretrained`.
+
+### 📍 Stage 1 Model (sft-stage1)
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+model_name = "KORMo-Team/KORMo-10B-sft"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    revision="sft-stage1",  # Load Stage 1 checkpoint
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    trust_remote_code=True
+)
+```
+
+### 🚀 Main Model (Final Checkpoint: sft-stage2-ckpt2)
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+model_name = "KORMo-Team/KORMo-10B-sft"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    revision="sft-stage2-ckpt2",  # Load Final Main Checkpoint
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    trust_remote_code=True
+)
+```
+
+> 💡 **Tip**:  
+> - Use `sft-stage1` for ablation studies or comparison experiments.  
+> - Use `sft-stage2-ckpt2` as the **main production model**.  
+
+---
+
+
 ## Contact
 - KyungTae Lim, Professor at KAIST. `[email protected]`
 
-
 ## Contributor