For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Python function

causal_attention_mask

causal_attention_mask()

max.pipelines.modeling.dataprocessing.causal_attention_mask(original_start_pos, original_seq_len)

source

Builds a causal attention mask for a batch of variable-length sequences.

Parameters:

• original_start_pos (list[int]) – Per-example start position (context length) in the batch.
• original_seq_len (list[int]) – Per-example sequence length for this pass.

Returns:

Float32 mask array where visible positions are 0 and masked positions are a large negative value (so that softmax treats them as -inf).

Return type:

ndarray[tuple[Any, …], dtype[float32]]

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit