StableLM
Affiliation
Stability AI
Commercial
Fine-tuning Method
SFT
Note
code
• StableLM-alpha (2023-04-20)
◦ 3B & 7B (이후에 15B & 65B release 예정)
◦ Data
The Pile
◦ CC BY-SA-4.0 : Commercial
• StableLM-tuned-alpha
◦ 3B & 7B
◦ StableLM-alpha의 추가 SFT version
◦ Data
Alpaca (52k)
GPT4All (400k) : GPT-3.5 Turbo로 생성한 437,605개의 (prompts-responses) dataset
Anthropic HH : ~160k Human-rated examples (harmfulness & helpfulness 기준, response pair 중에 더 선호되는 것)
DataBricks Dolly (15k)
ShareGPT English Subset
◦ StableLM-alpha →SFT
◦ CC BY-NC-SA-4.0 : Non-commercial
The Pile
◦ CC BY-SA-4.0 : Commercial
• StableLM-tuned-alpha
◦ 3B & 7B
◦ StableLM-alpha의 추가 SFT version
◦ Data
Alpaca (52k)
GPT4All (400k) : GPT-3.5 Turbo로 생성한 437,605개의 (prompts-responses) dataset
Anthropic HH : ~160k Human-rated examples (harmfulness & helpfulness 기준, response pair 중에 더 선호되는 것)
DataBricks Dolly (15k)
ShareGPT English Subset
◦ StableLM-alpha →SFT
◦ CC BY-NC-SA-4.0 : Non-commercial데이터
The Pile
Alpaca
GPT4All (400k) : GPT-3.5 Turbo로 생성한 437,605개의 (prompts-responses) dataset
Anthropic HH : ~160k Human-rated examples (harmfulness & helpfulness 기준, response pair 중에 더 선호되는 것)
DataBricks Dolly
ShareGPT
모델 크기
3B
7B
새롭게 제공된 Resource
Model
출시일
2023-04-20