Skip to content

Commit 5ed5a50

Browse files
committed
Add dosc for clip args for zero optim
1 parent a7ca297 commit 5ed5a50

File tree

1 file changed

+9
-6
lines changed

1 file changed

+9
-6
lines changed

colossalai/zero/gemini/gemini_optimizer.py

Lines changed: 9 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -46,12 +46,15 @@ class ZeroOptimizer(ColossalaiOptimizer):
4646
Defaults to 0.0.
4747
initial_scale (float, optional): Initial scale used by DynamicGradScaler. Defaults to 2**32.
4848
min_scale (float, optional): Min scale used by DynamicGradScaler. Defaults to 1.
49-
growth_factor (float, optional): growth_factor used by DynamicGradScaler. Defaults to 2.
50-
backoff_factor (float, optional): backoff_factor used by DynamicGradScaler. Defaults to 0.5.
51-
growth_interval (float, optional): growth_interval used by DynamicGradScaler. Defaults to 1000.
52-
hysteresis (float, optional): hysteresis used by DynamicGradScaler. Defaults to 2.
53-
max_scale (int, optional): max_scale used by DynamicGradScaler. Defaults to 2**32.
54-
"""
49+
growth_factor (float, optional): Growth_factor used by DynamicGradScaler. Defaults to 2.
50+
backoff_factor (float, optional): Backoff_factor used by DynamicGradScaler. Defaults to 0.5.
51+
growth_interval (float, optional): Growth_interval used by DynamicGradScaler. Defaults to 1000.
52+
hysteresis (float, optional): Hysteresis used by DynamicGradScaler. Defaults to 2.
53+
max_scale (int, optional): Max_scale used by DynamicGradScaler. Defaults to 2**32.
54+
clipping_norm (float, optional): The norm value used to clip gradient. Defaults to 0.0.
55+
norm_type (float, optional): The type of norm used for gradient clipping. Currently, only L2-norm (norm_type=2.0)
56+
is supported in ZeroOptimizer. Defaults to 2.0.
57+
"""
5558

5659
def __init__(self,
5760
optim: Optimizer,

0 commit comments

Comments
 (0)