David Koski

I think this has been fixed per https://github.com/ml-explore/mlx/pull/1506 in the mlx core repo.

@JoeJoe1313 see discussion in #297 -- I think we should go with the approach from #303, but thank you so much for working on this!

> I was planning to add support for all data formats MLX supports (besides `text`: `chat`, `tools`, and `completions`) Sounds great! > Before I proceed with implementing loading the the...

> Thanks [@davidkoski](https://github.com/davidkoski) I'll start adding the other file formats and I'll see if I run into any issues with LoRATrain along the way. > > Re: moving LorTrain.swift to...

> What data format is `LoRATrain.train(model:train:validate:optimizer:loss:tokenizer:parameters:progress:)` expec in the train and validate parameters? > > From this line in `LoadJSON(url:)` I understand that in the case of a text data...

> From your comment I understand we don't really have trainers for the different data formats, and we'll need to create separate trainers for each data format, is that correct?...

Looking at pixel_values coming in to `getInputEmbeddings` in python I see: ``` pixel_values.shape=(1, 13, 3, 512, 512) ``` vs ``` (lldb) po pixelValues!.shape ▿ 4 elements - 0 : 1...

@Blaizzy > I tried adding print statements to idefics3.swift so I can see the tensors but I can't see them in the CLI when I run it using mlx_run. I...

> My next question to be able to move on this is how to setup everything using Xcode? I struggled for 30 min and ended up doing via terminal. To...

@jolonf please make a new issue for this -- we can discuss.