DaVinci’s Model Overview
Category | Model | Platform | Token limit | Use Case |
|---|---|---|---|---|
Chat | In House* (Released by MTK Research) | 8000 Tokens | Simple paperwork Answer in traditional Chinese only | |
In House* z(Released by MTK Research) | 8000 Tokens | |||
Llama2 70B | In House* | 4000 Tokens | Simple paperwork Translation, writing, etc. | |
Llama3-8b-instruct | In House* | 8000 Tokens | ||
Llama3-70b-instruct | In House* | 8000 Tokens | ||
Microsoft Azure | 8000 Tokens | |||
GPT-3.5-Turbo-4k | Microsoft Azure | 4000 Tokens | ||
GPT-3.5-Turbo-16k | 16000 Tokens | |||
GPT-4-Turbo-8k | 8000 Tokens | Difficult or complex work Data comparison, analysis, etc. | ||
File Upload | ada | Microsoft Azure | -- | Text content embedding |
OCR | -- | Extract text from images | ||
whisper | In House | -- | Audio to Text | |
*In House Model does not have PIM issues only in 03. Direct Chat / 直接對話
Conversation Style
Style | Difference | Usage Scenarios |
|---|---|---|
Precise | DaVinci provides more precise answers. | Data analysis, comparison, etc. |
Creative | DaVinci's responses are more imaginative. | Translation, writing letters, writing, brainstorming for presentations, etc. |