|
--- |
|
license: mit |
|
task_categories: |
|
- text-classification |
|
language: |
|
- vi |
|
- en |
|
tags: |
|
- textclassification |
|
- modelrouting |
|
- routing |
|
- gptrouting |
|
- fewshot |
|
size_categories: |
|
- n<1K |
|
pretty_name: gpt_routing |
|
--- |
|
### Model Description |
|
|
|
This model is designed for prompt routing to determine whether a prompt should be handled by GPT-4o or GPT-3.5. The goal is to reduce costs, as GPT-4o is significantly more expensive (10x the cost of GPT-3.5). |
|
|
|
The dataset is intended for few-shot learning, which is why it contains a small number of prompts in two languages: Vietnamese and English. |
|
|
|
Labels were manually assigned based on specific use cases. You can expand the classification to include other models such as LLaMA or GPT-4, depending on your requirements. |
|
|
|
- **gpt-4o:** Handles complex questions that require deep thinking and analysis, such as math problems, multiple-choice homework, and tests. |
|
- **gpt-3.5:** Handles simpler questions primarily for information retrieval, where answers are readily available in the documents from a Retrieval-Augmented Generation (RAG) system, as well as straightforward code and explanations. |
