Transformers-openai-api

- Add support for [accelerate](https://github.com/huggingface/accelerate).

This release is a breaking configuration change as models will no longer switch to CPU inference mode in the absence of a CUDA device -- rather, new configuration parameters `MODEL_DEVICE` and `TOKENIZER_DEVICE` must be set explicitly on each model. For more details see the "Models" section of the README.

- Remove `logit_bias` from `generate` arguments

- Fallback to CPU inference when CUDA not available (thanks Yongtae723)
- Add `/metrics` endpoint

Initial release