Open LLM をまとめました。
2023年7月10日時点の情報です。ライセンスなどは改めて確認してください。
Release Dateに並べ替える。
Language Model | Developer | Release Date | Checkpoints | Language | Params (B) | Tokens | Context Length | Licence | Hugging Face |
---|---|---|---|---|---|---|---|---|---|
Xgen | Salesforce | 2023/06 | XGen-7B-4K-Base, XGen-7B-8K-Base, XGen-7B-8K-Inst (for research only) | 22 languages: Bulgarian, Catalan, Czech, Danish, German, English, Spanish, French, Croatian, Hungarian, Italian, Dutch, Polish, Romanian, Russian, Slovenian, Serbian, Swedish, Ukrainian, Japanese, Chinese | 7 | 1.5T | 4K to 8K | Apache 2.0 | Salesforce |
DLite | AI Squared | 2023/05 | dlite-v2-1_5b | English | 0.124 to 1.5 | 1024 | Apache 2.0 | AI Squared | |
h2oGPT | H2O LLM Studio | 2023/05 | h2oGPT | English | 12 to 20 | 256 - 2048 | Apache 2.0 | H2O LLM Studio | |
RedPajama-INCITE | Together and leaders | 2023/05 | RedPajama-INCITE | English | 3 to 7 | 2048 | Apache 2.0 | Together and leaders | |
OpenLLaMA | OpenLM Research | 2023/05 | open_llama_7b_700bt_preview, open_llama_3b_600bt_preview | English | 3, 7, 13 | 1T | 2048 | Apache 2.0 | OpenLM Research |
Falcon | The Technology Innovation Institute (TII) | 2023/05 | Falcon-40B, Falcon-7B | 11 languages, English, German, Spanish, French (and limited capabilities in Italian, Portuguese, Polish, Dutch, Romanian, Czech, Swedish) | 7, 40 | 1T | 2048 | Apache 2.0 | TII |
MPT | MosaicML | 2023/05 | MPT-7B, MPT-7B-Instruct, MPT-30B, MPT-30B-instruct | English | 7, 30 | 1T | 8192 to 65k | Apache 2.0, CC BY-SA-3.0 | MosaicML |
Pythia | EleutherAI | 2023/04 | 70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, 12B | English | 0.07 to 12 | 2048 | Apache 2.0 | EleutherAI | |
Dolly | Databricks | 2023/04 | dolly-v2-3b, dolly-v2-7b, dolly-v2-12b | English | 3, 7, 12 | 2048 | MIT | Databricks | |
StableLM-Alpha | StableLM | 2023/04 | StableLM-Alpha | English | 3 to 7 | 4096 | CC BY-SA-4.0 | StableLM | |
FastChat-T5 | lmsys | 2023/04 | fastchat-t5-3b-v1.0 | English | 3 | 512 | Apache 2.0 | lmsys | |
Cerebras-GPT | Cerebras | 2023/03 | 111M, 256M, 590M, 1.3B, 2.7B, 6.7B, and 13B | English | 0.111 to 13 | 2048 | Apache 2.0 | Cerebras | |
Open Assistant (Pythia family) | Open-Assistant | 2023/03 | OA-Pythia-12B-SFT-8, OA-Pythia-12B-SFT-4, OA-Pythia-12B-SFT-1 | English | 12 | 2048 | Apache 2.0 | Open-Assistant | |
Bloom | BigScience | 2022/11 | Bloom | 46 languages (No Japanese) | 176 | 366B | 2048 | OpenRAIL-M v1 | BigScience |
GPT-NeoX-20B | EleutherAI | 2022/04 | GPT-NEOX-20B | English | 20 | 2048 | Apache 2.0 | EleutherAI | |
RWKV | BlinkDL | 2021/08 | RWKV, ChatRWKV | 100+ languages | 0.1 to 14 | infinity (RNN) | Apache 2.0 | BlinkDL | |
GPT-J-6B | Eleuther AI | 2021/06 | GPT-J-6B, GPT4All-J | English | 6 | 2048 | Apache 2.0 | EleutherAI | |
mT5 | 2020/11 | mT5-Small, mT5-Base, mT5-Large, mT5-XL, mT5-XXL | 101 (with Japanese) | 0.3 to 13 | 512 | Apache 2.0 | |||
T5 | 2019/10 | T5 & Flan-T5, Flan-T5-xxl (HF) | English | 0.06 to 11 | 512 | Apache 2.0 |