Skip to main content

Module corpus

Module corpus 

Source
Expand description

Probe text synthesis helpers.

The probe sweeps (batch, seq) shapes by submitting synthesized texts to the leader worker. Texts come from the curated benchmark corpus; we repeat/trim corpus entries to hit the target token count for each shape.

Functionsยง

load_probe_texts ๐Ÿ”’
Loads the benchmark corpus for use as probe text material.
synthesize_texts ๐Ÿ”’
Synthesizes batch texts each of approximately target_seq tokens.