Skip to content

Commit cc8ff45

Browse files
authored
Update recompute_num_layers in the 1b config so it works with the model (#12517)
Signed-off-by: John St John <[email protected]>
1 parent ceba38f commit cc8ff45

File tree

1 file changed

+1
-0
lines changed

1 file changed

+1
-0
lines changed

nemo/collections/llm/gpt/model/hyena.py

+1
Original file line numberDiff line numberDiff line change
@@ -316,6 +316,7 @@ class Hyena1bConfig(HyenaConfig):
316316

317317
hybrid_override_pattern: str = "SDH*SDHSDH*SDHSDH*SDHSDH*"
318318
num_layers: int = 25
319+
recompute_num_layers: int = 5 # needs to be a multiple of num_layers
319320
seq_length: int = 8192
320321
hidden_size: int = 1920
321322
num_groups_hyena: int = 1920

0 commit comments

Comments
 (0)