[Feature] Add another OpenAISDK model using prompt completion API

Describe the feature

For now, OpenAISDK model only supports chat/completion API, which makes the infer service applies a chat template and triggers CoT, in its _generate method. CoT takes a large number of tokens, slows down the inference speed (request per minutes), and increases the probability of a answer got truncated.

To fix this, I request to add a new model, based on the original OpenAISDK, but using completion API instead of chat/completion. The completion API will only do completion by the prompt, and will give out a much high better result in simple eval tasks like choices, true/false, etc. The completion API doesn't take in messages and roles, so it cannot be used for complicated tasks like function call or multi-turn.

Will you implement it?

• I would like to implement this feature and create a PR!

Start by locating the existing OpenAISDK model and its _generate method in the repository, then trace how the infer service applies chat templates and passes messages to chat/completion. Compare that flow with the completion API requirements and existing model tests or evaluation entry points. Done means a separately usable model supports prompt completion while preserving the existing model's chat and multi-turn behavior.