Daniel Han

@takacsd The issue though timestamps can be reasonably easy to construct from text. The below could all be possible though: ``` list[list[struct[int, float]]] list[int] struct[list[datetime]] ``` Constructing Arrow dtypes from...

@takacsd agreed parsing the metadata string is the correct way. I thought about how one would go about doing it. Eg take: `list[pyarrow]` You'll have to first find the type...

Actually a simpler solution is to directly all `.replace` on the string and replace `list

@takacsd Nice work on the parser! :) Ye `struct` is the biggest issue with it being able to have column names. It gets worse if `struct` exists - yikes that'll...

@suptejas @avcode-exe Yes it should be supported now! Simply change the model name to any model from Yi. Also update Unsloth for local machines (Colab / Kaggle just refresh and...

@skmanzg Yes `packing = True` essentially combines small and long sequences into 1 example, hence it decreases

@skmanzg https://huggingface.co/docs/trl/en/sft_trainer#packing-dataset--constantlengthdataset-

I would turn it off to see if the results are better

Yes looks fine to me!

probs increase grad accumulation steps to smooth out the loss