`mlm.history_mlm`

`StepResults`

Bases: TypedDict

Parameters:

Name	Description	Default
`x`	The input token ids.	required
`attention_mask`	The mask for the input token ids.	required
`position_ids`	The ids of the positions in the sequence.	required
`logits`	The logits for the next token.	required
`t`	The time step.	required
`current_step`	The current step.	required
`change`	The change in the logits.	required
`input_end_positions`	The end positions of the input tokens.	required

Called between examples.

We will dump the top k tokens and probs as tensors in separate files for each step.