Downloads · 30 days
22
16% of all-time downloads
KoAlpaca-InterView-5.8B
UICHEOL-HWANG/KoAlpaca-InterView-5.8B
KoAlpaca-InterView-5.8B is a question answering model from UICHEOL-HWANG. Use it when the input is a question plus a passage.
- 이준범님 모델 기반으로 생성 - 데이터셋 : AiHub 채용 면접 인터뷰 데이터 : https://aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=71592
All-time downloads
137
Public
Parameters
5.9B
23.5 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors23.5 GB · 100%
At a glance
- Task
- Question Answering
- Model type
- gpt_neox
- Access
- Public
- Created
- Nov 21, 2024
- Updated
- Nov 27, 2024
- SHA
- 1e0ba69c
Base models
Model card
From the Hugging Face model README
@Update : 2024-11-27
KoAlpaca-Polyglot-5.8B
- 이준범님 모델 기반으로 생성
- 데이터셋 : AiHub 채용 면접 인터뷰 데이터 : https://aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=71592
'loss': 2.1778, 'grad_norm': 5.742631912231445, 'learning_rate': 1.831713150322434e-07, 'epoch': 4.86
Trainargument
epochs : 5
batch_size : 2
gradient_accumulation_steps : 32
weight_decay : 0.02
optim : paged_adamw_32bit
learning_rate: 1e-5
총론
- GCP V100 1대 사용
- 아쉽게도 배치사이즈를 늘릴 수가 없는 상황(최소 A100은 되어야 RAM 보장됨, V100으로는 훈련중 터짐 )
- Validation 불가
- 램 용량 부하로 인해 validation을 불가한 상황
- 인퍼런스 및 양자화
- 추후 재 설계 예정
- Task
- Question Answering
- Type
- gpt_neox
- Languages
- ko
- Created
- Nov 21, 2024
- Updated
- Nov 27, 2024