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

Fix (base): Updating A2Q defaults #718

Merged
merged 4 commits into from
Nov 6, 2023
Merged

Fix (base): Updating A2Q defaults #718

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
216e68e
Update A2Q defaults
i-colbert Oct 10, 2023
78ec9be
Fixing backwards compatibility
i-colbert Oct 10, 2023
6cdadcd
Pre-commit fixes
i-colbert Oct 12, 2023
b5f0e9e
Update pre_scaling.py
i-colbert Nov 2, 2023
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
3 changes: 1 addition & 2 deletions src/brevitas/core/scaling/pre_scaling.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,8 +15,7 @@
from brevitas.core.stats.stats_wrapper import _Stats
from brevitas.function import abs_binary_sign_grad

__all__ = [
"ParameterPreScalingWeightNorm",]
__all__ = ["ParameterPreScalingWeightNorm", "AccumulatorAwareParameterPreScaling"]


class ParameterPreScalingWeightNorm(brevitas.jit.ScriptModule):
Expand Down
3 changes: 1 addition & 2 deletions src/brevitas/quant/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -352,7 +352,7 @@ def scaling_init(scaling_init_impl, bit_width):
tensor_clamp_impl = TensorClamp
scaling_impl = ParameterScaling
scaling_init_impl = StatsFromParameterScaling
restrict_scaling_impl = FloatRestrictValue
restrict_scaling_impl = LogFloatRestrictValue
scaling_stats_impl = AbsMax
pre_scaling_impl = ParameterPreScalingWeightNorm
restrict_pre_scaling_impl = LogFloatRestrictValue
Expand Down Expand Up @@ -395,7 +395,6 @@ def accumulator_bit_width_impl(accumulator_bit_width):
proxy_class = DecoupledWeightQuantWithInputProxyFromInjector
tensor_quant = DecoupledRescalingIntQuantWithInput
pre_scaling_impl = AccumulatorAwareParameterPreScaling
pre_scaling_min_val = 1e-8
accumulator_bit_width = 32 # default maximum accumulator width is 32 bits
normalize_stats_impl = L1Norm # required to align with derivations in paper
float_to_int_impl = RoundToZeroSte # required to ensure no upwards rounding violates constraints
Expand Down
2 changes: 2 additions & 0 deletions src/brevitas_examples/super_resolution/models/common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,7 @@
from torch import Tensor
import torch.nn as nn

from brevitas.core.restrict_val import FloatRestrictValue
from brevitas.core.restrict_val import RestrictValueType
from brevitas.core.scaling import ScalingImplType
import brevitas.nn as qnn
Expand All @@ -25,6 +26,7 @@ class CommonIntWeightPerChannelQuant(Int8WeightPerTensorFloat):


class CommonIntAccumulatorAwareWeightQuant(Int8AccumulatorAwareWeightQuant):
restrict_scaling_impl = FloatRestrictValue # backwards compatibility
pre_scaling_min_val = 1e-10
scaling_min_val = 1e-10

Expand Down
Loading