#This makes no sense and your own models agree

3 messages · Page 1 of 1 (latest)

mellow hare
#

"How many epochs are in the training process?
A general rule of thumb is: Num epochs = max_steps / file_of_training_jsonls_in_MB. For instance, if your training file is 100MB and you set max_steps=1000, the training process will roughly perform 10 epochs."

Large 2
The statement you provided is attempting to describe a general method for determining the number of epochs in a training process based on the size of the training file and the number of training steps. However, the formula provided is a bit unclear and seems to misinterpret the relationship between the number of epochs and the other parameters.

A more accurate and clear way to express this relationship would be:

Epochs: Represents the number of times the entire training dataset is passed through the neural network.
Steps: Represents the number of times the model updates its weights during training.

In summary, the initial statement is confusing because it tries to relate epochs directly to the file size in megabytes, which is not a standard or typical way to determine the number of epochs. The size of the file in MB is not directly related to the number of training samples or the number of epochs. Instead, consider the total number of samples and the batch size to determine the number of epochs more accurately.

bleak horizon
#

The formula is custom, the steps depend on parameters like batches and other details that you cannot control with the API itself, its a simple rule of thumb to approximate the amount of epochs using that formula.

And you can actually get back a dry response from the API with the actual epochs btw!!

#

The model wont know about this, since are very recent things that the model wasnt specially trained on