Three basic parameters adapted to mainstream platforms and models

[chenxizhang]

The initial intent of this module design is to eliminate the differences between mainstream platforms and models on the market, enabling users to engage in a unified ChatGPT experience or text/image generation experiences. To achieve this goal, we have designed around three basic parameters to ensure compatibility with almost all mainstream platforms and models.

Currently, the mainstream platforms and models we support include: OpenAI, Azure OpenAI, Kimi, Zhìpǔ Qīngyán, DBRX, and even locally supported models using ollama, in addition to any other platforms or models compatible with OpenAI.

The three key parameters mentioned here are:

api_key

This refers to the key required by the platform or model. It is recommended to store it using the OPENAI_API_KEY environment variable.

• If you are using a local model, a key is generally not required, but to maintain a unified design, you still need to specify a value for the api_key, though you may choose any value arbitrarily.

endpoint

This refers to the service endpoint address used by the platform or model. It is recommended to store it using the OPENAI_API_ENDPOINT environment variable.

• If not defined, and not provided during the call, it will default to using OpenAI’s standard service address https://api.openai.com/v1/chat/completions.

model

This refers to the model you need to use, and it is recommended to store it using the OPENAI_API_MODEL environment variable with the default value being gpt-3.5-turbo.

If you have fully defined the above three environment variables, then using the chat or gpt functionalities becomes straightforward without needing to worry about these details—you simply focus on what functionality you want to achieve. Otherwise, you will need to specify them during each call—this can also be acceptable, for example, if you are using them in an automated system where you might prefer to specify parameters directly in the commands rather than relying on external environment variables. Thus, you can choose the approach most suitable for your circumstances.