Update README.md

README.md

@@ -27,52 +27,191 @@ pipeline_tag: text-generation
 ---
 
 # CodeMind
-Coding Test Explanatory LLM Model.
-
-## Model Details
-- **Model Name**: CodeMind
-- **Base Model**: google/gemma-2b, google/gemma-1.1-2b-it
-- **Fine-tuning Datasets**:
-  - [kreimben/leetcode_user_submissions_only_python](https://huggingface.co/datasets/kreimben/leetcode_user_submissions_only_python)
-  - [kreimben/leetcode_with_youtube_captions](https://huggingface.co/datasets/kreimben/leetcode_with_youtube_captions)
-- **Model Type**: Language Model
-- **Language**: English
-- **License**: gemma
-- **Model Size**: 2.51B
-
-## Intended Use
-CodeMind is a fine-tuned language model specifically designed to assist users with coding test questions and provide programming education. It leverages the knowledge from LeetCode user submissions in Python and YouTube video captions related to LeetCode problems to offer guidance, explanations, and code examples.
-
-## Training Data
-The model was fine-tuned using the following datasets:
-1. **kreimben/leetcode_user_submissions_only_python**: This dataset contains a collection of LeetCode user submissions written in Python. It provides a wide range of coding solutions to various algorithmic problems commonly encountered in coding interviews.
-2. **kreimben/leetcode_with_youtube_captions**: This dataset combines LeetCode problems with corresponding YouTube video captions. The captions provide explanations, thought processes, and step-by-step guidance for solving the coding problems.
-
-## Training Procedure
-The model was fine-tuned using the Hugging Face Transformer library. The base model, gemma-1.1-2b-it, was further trained on the combined dataset of LeetCode user submissions and YouTube video captions. The fine-tuning process aimed to enhance the model's understanding of coding concepts, problem-solving strategies, and its ability to generate relevant code snippets and explanations.
-
-## Evaluation
+
+## 소개
+`CodeMind-gemma-2b`는 코딩 테스트 문제 해결 및 프로그래밍 교육을 지원하는 언어 모델입니다.
+이 모델은 LeetCode 정리글과 관련 유튜브 캡션을 활용하여 문제 해결에 대한 설명과 코드 예제를 제공합니다.
+
+## 모델 세부 정보
+- **모델 이름**: CodeMind
+- **기본 모델**: google/gemma-2b
+- **언어**: 영어
+- **모델 크기**: 2.51B 파라미터
+- **라이선스**: MIT
+
+## 주요 기능
+- 코딩 테스트 문제 해결
+- 프로그래밍 개념 설명
+- 관련 코드 스니펫 생성
+
+## 훈련 데이터
+- [**LeetCode 사용자 제출물**](https://huggingface.co/datasets/kreimben/leetcode_user_submissions): 다양한 알고리즘 문제의 파이썬 솔루션
+- [**유튜브 캡션**](https://huggingface.co/datasets/kreimben/leetcode_with_youtube_captions): LeetCode 문제에 대한 설명 및 단계별 가이드
+
+## 사용된 라이브러리
+- [transformers](https://github.com/huggingface/transformers): 자연어 처리 모델을 위한 라이브러리
+- [datasets](https://github.com/huggingface/datasets): 데이터셋 처리 및 관리 라이브러리
+- [bitsandbytes](https://github.com/TimDettmers/bitsandbytes): 최적화된 연산을 위한 라이브러리
+- [peft](https://github.com/huggingface/peft): 파인 튜닝을 위한 라이브러리
+- [trl](https://github.com/huggingface/trl): 언어 모델 튜닝을 위한 라이브러리
+- [pandas](https://github.com/pandas-dev/pandas): 데이터 조작을 위한 라이브러리
+
+## 파일 구조
+- **dataset/**: 데이터셋 파일을 포함합니다.
+- **eval/**: 평가 스크립트를 포함합니다.
+- **fine-tuning/**: fine tuning 관련 노트북 및 스크립트를 포함합니다.
+  - `gemma-1.1-2b-it peft qlora.ipynb`: fine tuning 과정에 대한 세부 사항이 포함된 노트북입니다.
+- **demo.ipynb**: 데모 노트북으로 모델 사용 예제가 포함되어 있습니다.
+- **requirements.txt**: 프로젝트 의존성 목록이 포함되어 있습니다.
+- **utils.py**: 유틸리티 함수들이 포함되어 있습니다.
+
+## 사용 방법
+이 모델은 HuggingFace의 모델 허브를 통해 액세스할 수 있으며, API를 사용하여 응용 프로그램에 통합할 수 있습니다. 코딩 문제 또는 프로그래밍 관련 질문을 제공하면 모델이 관련 설명, 코드 스니펫 또는 가이드를 생성합니다.
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+tokenizer = AutoTokenizer.from_pretrained("kreimben/CodeMind-gemma-2b")
+model = AutoModelForCausalLM.from_pretrained("kreimben/CodeMind-gemma-2b")
+
+inputs = tokenizer("코딩 문제나 질문을 여기에 입력하세요", return_tensors="pt")
+outputs = model.generate(inputs.input_ids)
+print(tokenizer.decode(outputs[0]))
+```
+
+## 훈련 과정
+
+### 모델 및 토크나이저 로드
+```python
+import os
+from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
+
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype=torch.bfloat16
+)
+
+model_id = 'google/gemma-1.1-2b-it'
+token = os.getenv('HF_READ')
+
+model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config, device_map={"": 0}, token=token)
+model.config.use_cache = False
+model.gradient_checkpointing_enable()
+
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+tokenizer.padding_side = 'right'
+tokenizer.pad_token = tokenizer.eos_token
+```
+
+### LoRA 구성 및 모델 준비
+```python
+from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training
+import bitsandbytes as bnb
+
+model = prepare_model_for_kbit_training(model)
+
+def find_all_linear_names(model):
+    cls = bnb.nn.Linear4bit
+    lora_module_names = set()
+    for name, module in model.named_modules():
+        if isinstance(module, cls):
+            names = name.split('.')
+            lora_module_names.add(names[0] if len(names) == 1 else names[-1])
+        if 'lm_head' in lora_module_names:
+            lora_module_names.remove('lm_head')
+    return list(lora_module_names)
+
+modules = find_all_linear_names(model)
+lora_config = LoraConfig(
+    r=64,
+    lora_alpha=32,
+    target_modules=modules,
+    lora_dropout=0.05,
+    bias="none",
+    task_type="CAUSAL_LM"
+)
+
+model = get_peft_model(model, lora_config)
+```
+
+### 데이터 준비
+```python
+import pandas as pd
+from datasets import Dataset
+
+submission_dataset = datasets.load_dataset('kreimben/leetcode_user_submissions_only_python', split='train').to_pandas()
+submission_dataset = submission_dataset[['title', 'question_hints', 'question_content', 'content']]
+captions_dataset = datasets.load_dataset('kreimben/leetcode_with_youtube_captions', split='train').to_pandas()
+captions_dataset = captions_dataset[['title', 'question_hints', 'question_content', 'cc_content']]
+captions_dataset.rename(columns={'cc_content': 'content'}, inplace=True)
+
+dataset = pd.concat([submission_dataset, captions_dataset])
+del submission_dataset, captions_dataset
+
+dataset = Dataset.from_pandas(dataset)
+GEMMA_2B_IT_MODEL_PREFIX_TEXT = "Below is an coding test problem. Solve the question."
+
+def generate_prompt(data_point):
+    return f"<bos><start_of_turn>user {GEMMA_2B_IT_MODEL_PREFIX_TEXT}
+
+I don't know {data_point['title']} problem. give me the insight or appoach.
+
+this is problem's hint.
+{data_point['question_hints']}
+
+here are some content of question.
+{data_point['question_content']}<end_of_turn>
+<start_of_turn>model {data_point['content']}<end_of_turn><eos>"
+
+text_column = [generate_prompt(data_point) for data_point in dataset]
+dataset = dataset.add_column("prompt", text_column)
+```
+
+### 훈련
+```python
+from trl import SFTTrainer
+import transformers
+import torch
+
+tokenizer.pad_token = tokenizer.eos_token
+torch.cuda.empty_cache()
+
+trainer = SFTTrainer(
+    model=model,
+    train_dataset=dataset,
+    dataset_text_field="prompt",
+    peft_config=lora_config,
+    data_collator=transformers.DataCollatorForLanguageModeling(tokenizer, mlm=False),
+    args=transformers.TrainingArguments(
+        output_dir='out',
+        bf16=True,
+        max_steps=100,
+        warmup_steps=50,
+        per_device_train_batch_size=1,
+        gradient_accumulation_steps=1,
+        optim="paged_adamw_8bit",
+        logging_steps=20,
+        report_to='wandb',
+    ),
+)
+
+trainer.train()
+```
+
+## 평가
+모델의 성능은 다음과 같이 평가되었습니다:
+
 | Metric       | Value  |
 |--------------|--------|
-| average      | 41.62  |
-| arc          | 41.81  |
+| Average      | 41.62  |
+| ARC          | 41.81  |
 | HellaSwag    | 59.03  |
 | MMLU         | 37.26  |
 | TruthfulQA   | 43.45  |
 | Winogrande   | 59.91  |
 | GSM8K        | 8.26   |
 
-## Bias and Limitations
-- The model's knowledge is primarily based on the LeetCode user submissions and YouTube video captions used for fine-tuning. It may have limitations in handling coding problems or concepts that are not well-represented in the training data.
-- The model's responses are generated based on patterns and information learned from the training data. It may sometimes produce incorrect or suboptimal solutions. Users should always review and verify the generated code before using it in practice.
-- The model may exhibit biases present in the training data, such as favoring certain programming styles, algorithms, or approaches. It is important to consider alternative solutions and best practices when using the model's outputs.
-
-## Ethical Considerations
-- The model should be used as a supportive tool for learning and problem-solving, not as a substitute for human expertise and critical thinking.
-- Users should be aware that the model's responses are generated based on patterns in the training data and may not always be accurate, complete, or up to date.
-- The model should not be relied upon for making critical decisions or solving real-world problems without thorough validation and testing.
-
-## Usage
-To use the CodeMind model, you can access it through the Hugging Face model hub or by integrating it into your own applications using the provided API. Provide a coding problem or a question related to programming concepts, and the model will generate relevant explanations, code snippets, or guidance based on its training.
-
-Please refer to the documentation and examples for detailed instructions on how to integrate and use the CodeMind model effectively.
+## 제한 사항 및 윤리적 고려사항
+- 모델의 출력은 학습 데이터에 기반하므로 항상 정확하지 않을 수 있습니다.
+- 중요한 결정이나 실세계 문제 해결에 모델 출력을 사용하기 전에 반드시 검증이 필요합니다.