Commit History
Today
-
Headline said 'No optimizer state'; the body correctly says 'saved'
Verified signatureb498abb -
Remove asserted training denials and the retired 489/500 figure in percentage form
Verified signature7c91302
Yesterday
-
Say plainly that this is gradient descent
Verified signature'No Trainer, no TrainingArguments' is literally true and leads readers to conclude no gradients were used, which is false. The card now states that a hand-rolled loop computes a loss, calls backward, and updates one MLP projection per layer. Absence of a framework is not absence of training.9623308
-
Upload README.md with huggingface_hub
Verified signature8721afd
-
Initial upload
ab9bf6b