Create a new model identical to GPT2. Actual changes will be added later to track changes more easily. Based on https://github.com/huggingface/transformers/pull/21253, without the doc and model changes. Also using the name GPTBigCode instead because it does a lot more than MQA, but the name can be changed later.
Create a new model identical to GPT2. Actual changes will be added later to track changes more easily. Based on https://github.com/huggingface/transformers/pull/21253, without the doc and model changes. Also using the name GPTBigCode instead because it does a lot more than MQA, but the name can be changed later.
Tested with https://github.com/bigcode-project/bigcode-inference-benchmark/pull/14. Lots of things probably won't work until updated (tests, onnx, pretrained models, etc.)