Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Use default value of initial_scale_power if FP16 scaling params not provided #4986

Closed
wants to merge 4 commits into from
Closed

Use default value of initial_scale_power if FP16 scaling params not provided #4986

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
5c9c8de
Use default value as initial_scale_power if FP16 scaling params not p…
ShukantPal Jan 21, 2024
bb5eebd
Merge branch 'master' into shukant/fix-initial-scale
loadams Jan 22, 2025
0cf2bf4
Merge branch 'master' into shukant/fix-initial-scale
loadams Jan 24, 2025
c38e8fd
Merge branch 'master' into shukant/fix-initial-scale
loadams Jan 28, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions deepspeed/runtime/fp16/loss_scaler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@

import torch
from deepspeed import comm as dist
from deepspeed.runtime.constants import FP16_INITIAL_SCALE_POWER_DEFAULT
from deepspeed.utils import logger

INITIAL_LOSS_SCALE = 'init_scale'
Expand Down Expand Up @@ -109,14 +110,14 @@ class DynamicLossScaler(LossScalerBase):
always using the highest loss scale possible without incurring overflow.

Args:
init_scale (float, optional, default=2**32): Initial loss scale attempted by :class:`DynamicLossScaler.`
init_scale (float, optional, default=2**16): Initial loss scale attempted by :class:`DynamicLossScaler.`
scale_factor (float, optional, default=2.0): Factor used when adjusting the loss scale. If an overflow is encountered, the loss scale is readjusted to loss scale/``scale_factor``. If ``scale_window`` consecutive iterations take place without an overflow, the loss scale is readjusted to loss_scale*``scale_factor``.
scale_window (int, optional, default=1000): Number of consecutive iterations without an overflow to wait before increasing the loss scale.
consecutive_hysteresis (bool, optional, default=False): Whether to refill hysteresis if we reach an iteration that doesn't overflow
"""

def __init__(self,
init_scale=2**32,
init_scale=2**FP16_INITIAL_SCALE_POWER_DEFAULT,
scale_factor=2.,
scale_window=1000,
min_scale=1,
Expand Down
Loading