jp-Julia2 is the latest in my attempt to learn TI's.

• I made this embedding with 8 stock images. Images were taken on a blank background .

• While this model is close it does not reproduce any of the training images exactly. I may do another version.

• 8 images trained for 500 steps.

• 1 vector per token.

• learning rate of .005

• Batch size of 1 with 1 gradient accumulation steps.