hf_to_gguf support for DataDecide (AI2) checkpoints

[DeepStart]

I have tried to convert (unsuccessfully) a small DataDecide model to GGUF. I have asked for help at DataDecide's GitHub and it seems that there may be a need for some modifications of hf_to_gguf scripts to get DataDecide models converted. It was pointed that DataDecide uses this branch of the original OLMo repo. .

Out of the box support for DataDecide models/checkpoints in llama.cpp would be a nice thing IMHO, because they have a few dozens of small to medium pretrained checkpoints with known training data and training recepies. It would be interesting to compare how those checkpoints perform on llama.cpp.