From 58d75437ce9bed03112b37371e8edc095f7a58bc Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Aur=C3=A9lie=20saulquin?= <aurelie.saulq@gmail.com>
Date: Fri, 14 Mar 2025 11:37:04 +0100
Subject: [PATCH] add clamp to neuron model
---
modneflib/modnef/modnef_torch/model.py | 6 ++++++
.../modnef_neurons/blif_model/blif.py | 7 +++++++
.../modnef_neurons/blif_model/rblif.py | 9 +++++++++
.../modnef_neurons/modnef_torch_neuron.py | 7 +++++++
.../modnef_neurons/slif_model/rslif.py | 9 +++++++++
.../modnef_neurons/slif_model/slif.py | 6 ++++++
.../modnef_neurons/srlif_model/rshiftlif.py | 9 +++++++++
.../modnef_neurons/srlif_model/shiftlif.py | 8 ++++++++
.../modnef/quantizer/dynamic_scale_quantizer.py | 4 ----
modneflib/modnef/quantizer/quantizer.py | 16 ++++++----------
10 files changed, 67 insertions(+), 14 deletions(-)
diff --git a/modneflib/modnef/modnef_torch/model.py b/modneflib/modnef/modnef_torch/model.py
index bf63ae0..44443dc 100644
--- a/modneflib/modnef/modnef_torch/model.py
+++ b/modneflib/modnef/modnef_torch/model.py
@@ -106,6 +106,12 @@ class ModNEFModel(nn.Module):
if isinstance(m, ModNEFNeuron):
m.quantize(force_init=force_init)
+ def clamp(self):
+
+ for m in self.modules():
+ if isinstance(m, ModNEFNeuron):
+ m.clamp()
+
def train(self, mode : bool = True, quant : bool = False):
"""
Set neuron model for trainning
diff --git a/modneflib/modnef/modnef_torch/modnef_neurons/blif_model/blif.py b/modneflib/modnef/modnef_torch/modnef_neurons/blif_model/blif.py
index 3c83a40..d223c86 100644
--- a/modneflib/modnef/modnef_torch/modnef_neurons/blif_model/blif.py
+++ b/modneflib/modnef/modnef_torch/modnef_neurons/blif_model/blif.py
@@ -342,6 +342,13 @@ class BLIF(Leaky, ModNEFNeuron):
self.quantize_parameters()
self.quantization_flag = True
+ def clamp(self):
+ """
+ Clamp synaptic weight and neuron hyper-parameters
+ """
+
+ self.fc.weight.data = self.quantizer.clamp(self.fc.weight.data)
+
@classmethod
def detach_hidden(cls):
"""Returns the hidden states, detached from the current graph.
diff --git a/modneflib/modnef/modnef_torch/modnef_neurons/blif_model/rblif.py b/modneflib/modnef/modnef_torch/modnef_neurons/blif_model/rblif.py
index f0ddcdd..d48960d 100644
--- a/modneflib/modnef/modnef_torch/modnef_neurons/blif_model/rblif.py
+++ b/modneflib/modnef/modnef_torch/modnef_neurons/blif_model/rblif.py
@@ -357,6 +357,15 @@ class RBLIF(Leaky, ModNEFNeuron):
self.quantize_parameters()
self.quantization_flag = True
+ def clamp(self):
+ """
+ Clamp synaptic weight and neuron hyper-parameters
+ """
+
+ self.fc.weight.data = self.quantizer.clamp(self.fc.weight.data)
+ self.reccurent.weight.data = self.quantizer.clamp(self.reccurent.weight.data)
+
+
@classmethod
def detach_hidden(cls):
"""Returns the hidden states, detached from the current graph.
diff --git a/modneflib/modnef/modnef_torch/modnef_neurons/modnef_torch_neuron.py b/modneflib/modnef/modnef_torch/modnef_neurons/modnef_torch_neuron.py
index dc49183..7322594 100644
--- a/modneflib/modnef/modnef_torch/modnef_neurons/modnef_torch_neuron.py
+++ b/modneflib/modnef/modnef_torch/modnef_neurons/modnef_torch_neuron.py
@@ -91,6 +91,13 @@ class ModNEFNeuron():
raise NotImplementedError()
+ def clamp(self):
+ """
+ Clamp synaptic weight
+ """
+
+ raise NotImplementedError()
+
def set_quant(self, mode=False):
self.quantization_flag = mode
diff --git a/modneflib/modnef/modnef_torch/modnef_neurons/slif_model/rslif.py b/modneflib/modnef/modnef_torch/modnef_neurons/slif_model/rslif.py
index c950a76..694971e 100644
--- a/modneflib/modnef/modnef_torch/modnef_neurons/slif_model/rslif.py
+++ b/modneflib/modnef/modnef_torch/modnef_neurons/slif_model/rslif.py
@@ -375,6 +375,15 @@ class RSLIF(LIF, ModNEFNeuron):
self.quantize_weight()
self.quantize_parameters()
+
+ def clamp(self):
+ """
+ Clamp synaptic weight and neuron hyper-parameters
+ """
+
+ self.fc.weight.data = self.quantizer.clamp(self.fc.weight.data)
+ self.reccurent.weight.data = self.quantizer.clamp(self.reccurent.weight.data)
+
@classmethod
def detach_hidden(cls):
diff --git a/modneflib/modnef/modnef_torch/modnef_neurons/slif_model/slif.py b/modneflib/modnef/modnef_torch/modnef_neurons/slif_model/slif.py
index 315a07d..fa370b9 100644
--- a/modneflib/modnef/modnef_torch/modnef_neurons/slif_model/slif.py
+++ b/modneflib/modnef/modnef_torch/modnef_neurons/slif_model/slif.py
@@ -366,6 +366,12 @@ class SLIF(LIF, ModNEFNeuron):
self.quantize_weight()
self.quantize_parameters()
+ def clamp(self):
+ """
+ Clamp synaptic weight and neuron hyper-parameters
+ """
+
+ self.fc.weight.data = self.quantizer.clamp(self.fc.weight.data)
@classmethod
def detach_hidden(cls):
diff --git a/modneflib/modnef/modnef_torch/modnef_neurons/srlif_model/rshiftlif.py b/modneflib/modnef/modnef_torch/modnef_neurons/srlif_model/rshiftlif.py
index 427a1c0..0193876 100644
--- a/modneflib/modnef/modnef_torch/modnef_neurons/srlif_model/rshiftlif.py
+++ b/modneflib/modnef/modnef_torch/modnef_neurons/srlif_model/rshiftlif.py
@@ -375,6 +375,15 @@ class RShiftLIF(LIF, ModNEFNeuron):
self.quantize_weight()
self.quantize_parameters()
+ def clamp(self):
+ """
+ Clamp synaptic weight and neuron hyper-parameters
+ """
+
+ self.fc.weight.data = self.quantizer.clamp(self.fc.weight.data)
+ self.reccurent.weight.data = self.quantizer.clamp(self.reccurent.weight.data)
+
+
@classmethod
def detach_hidden(cls):
"""Returns the hidden states, detached from the current graph.
diff --git a/modneflib/modnef/modnef_torch/modnef_neurons/srlif_model/shiftlif.py b/modneflib/modnef/modnef_torch/modnef_neurons/srlif_model/shiftlif.py
index 0ffa99b..9bb6d1d 100644
--- a/modneflib/modnef/modnef_torch/modnef_neurons/srlif_model/shiftlif.py
+++ b/modneflib/modnef/modnef_torch/modnef_neurons/srlif_model/shiftlif.py
@@ -357,6 +357,14 @@ class ShiftLIF(LIF, ModNEFNeuron):
self.quantize_weight()
self.quantize_parameters()
+ def clamp(self):
+ """
+ Clamp synaptic weight and neuron hyper-parameters
+ """
+
+ self.fc.weight.data = self.quantizer.clamp(self.fc.weight.data)
+
+
@classmethod
def detach_hidden(cls):
"""Returns the hidden states, detached from the current graph.
diff --git a/modneflib/modnef/quantizer/dynamic_scale_quantizer.py b/modneflib/modnef/quantizer/dynamic_scale_quantizer.py
index 4f4b096..724a8f4 100644
--- a/modneflib/modnef/quantizer/dynamic_scale_quantizer.py
+++ b/modneflib/modnef/quantizer/dynamic_scale_quantizer.py
@@ -145,11 +145,7 @@ class DynamicScaleFactorQuantizer(Quantizer):
-------
Tensor
"""
-
- born_min = -int(self.signed)*2**(self.bitwidth-1)
- born_max = 2**(self.bitwidth-int(self.signed))-1
- #scaled = torch.clamp(data/self.scale_factor, min=born_min, max=born_max).to(dtype)
scaled = torch.round(data/self.scale_factor).to(self.dtype)
if unscale:
diff --git a/modneflib/modnef/quantizer/quantizer.py b/modneflib/modnef/quantizer/quantizer.py
index 2a8ef79..235981d 100644
--- a/modneflib/modnef/quantizer/quantizer.py
+++ b/modneflib/modnef/quantizer/quantizer.py
@@ -152,21 +152,17 @@ class Quantizer():
int | float | list | numpy.array | Tensor (depending on type of data)
"""
- b_min = -1 #(-2**(self.bitwidth-int(self.signed))*int(self.signed))
- b_max = 1 #2**(self.bitwidth-int(self.signed))-1
+ born_min = -int(self.signed)*2**(self.bitwidth-1)
+ born_max = 2**(self.bitwidth-int(self.signed))-1
if isinstance(data, (int, float)):
- return self._clamp(torch.tensor(data)).item()
+ return torch.clamp(torch.tensor(data), min=born_min, max=born_max).item()
elif isinstance(data, list):
- return self._clamp(torch.tensor(data)).tolist()
+ return torch.clamp(torch.tensor(data), min=born_min, max=born_max).tolist()
elif isinstance(data, np.ndarray):
- return self._clamp(torch.tensor(data)).numpy()
+ return torch.clamp(torch.tensor(data), min=born_min, max=born_max).numpy()
elif torch.is_tensor(data):
- return self._clamp(data).detach()
+ return torch.clamp(data, min=born_min, max=born_max).detach()
else:
raise TypeError("Unsupported data type")
-
- def _clamp(self, data):
-
- raise NotImplementedError()
\ No newline at end of file
--
GitLab