mirror of
https://github.com/wassname/discovering_latent_knowledge.git
synced 2026-09-11 12:10:11 +08:00
tidy
This commit is contained in:
@@ -58,11 +58,11 @@ def create_cache_interventions(
|
||||
model,
|
||||
tokenizer,
|
||||
cfg,
|
||||
N_fit_examples=20,
|
||||
batch_size=2,
|
||||
# N_fit_examples=20,
|
||||
# batch_size=2,
|
||||
rep_token=-1,
|
||||
n_difference=1,
|
||||
direction_method="pca",
|
||||
# direction_method="pca",
|
||||
get_negative=False,
|
||||
):
|
||||
"""
|
||||
@@ -70,6 +70,9 @@ def create_cache_interventions(
|
||||
|
||||
So we always load a cached version if possible. to make it approx repeatable use the same dataset etc
|
||||
"""
|
||||
direction_method=cfg.intervention_direction_method
|
||||
N_fit_examples=cfg.intervention_fit_examples
|
||||
batch_size=cfg.batch_size
|
||||
tokenizer_args = dict(
|
||||
padding="max_length",
|
||||
max_length=cfg.max_length,
|
||||
|
||||
Reference in New Issue
Block a user