retato-snu/focuspang_llm

FOCUSPANG_LLM

A project for 2023-Fall SNU Creative integrated design lecture (Team L)

FOCUSPANG

포커스팡은 온라인 학습과정에서 학생들의 다양한 행동 패턴을 분석하여 수업에 집중을 잘 하고 있는지 판단을 도와주고, 성적 분포를 분석하여 앞으로의 학습 계획을 도와주는 서비스이다. 이 프로젝트에서는 성적 분포를 분석하여 학생들의 전체적인 학습 수행 능력에 대한 피드백을 주는 LLM 모델을 설계하고 제작하는 것을 목표로 한다.

Training Structure

Colossal

How to run our codes

Install dependencies

Before start, Need miniconda3

1. Create anaconda environment

Create env

conda create -n env_name python=3.10.13

Install torch

conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia

Install Transfomer

pip install transformers

Install flash-attention

pip install flash-attn --no-build-isolation

Install xformers

conda install xformers -c xformers

Install Colossal & Coati

cd Colossal

pip install .

2. Install dependencies for learning

cd learning

pip install -r requirements.txt

3. Clone pretrained model (Additional Option)

git lfs install

• polyglot-ko-1.3b git clone https://huggingface.co/EleutherAI/polyglot-ko-1.3b
• polyglot-ko-5.8b git clone https://huggingface.co/EleutherAI/polyglot-ko-5.8b
• polyglot-ko-12.8b git clone https://huggingface.co/EleutherAI/polyglot-ko-12.8b

Without clone pretrained model, Parameter pretained part should be written like: EleutherAI/polyglot-ko-1.3b

This will be saved cache, so clone model and use local file is recommended.

4. Additional dependency for learning with 8bit (Additional Option)

If you want to learn with 8bit for limit of GPU memory, use this version: Quantization learning version.

You need to additional dependency for this.

pip install peft

Training

1. SFT model

In FOCUSPANG_LLM/Colossal/learning, you can use train_sft.sh script.

2. RM model

In FOCUSPANG_LLM/Colossal/learning, you can use train_rm.sh script.

3. PPO model

In FOCUSPANG_LLM/Colossal/learning, you can use train_prompt.sh script.

Inference

Example command

 CUDA_VISIBLE_DEVICES=1 python inference.py \
    --model polyglotko \
    --pretrain /mnt/hf/polyglot-ko-5.8b \
    --model_path $YOUR_PPO_MODEL_PATH \
    --input $YOUR_INPUT \

Implement Detail

Information about the files/code developed ourselved can be found in the README.md provided below.

• Colossal
• DeepSpeed
• Dataset