fix(mge/quant): fix TQT epoch scale change bug

GitOrigin-RevId: 6e39de9cec

python_module/megengine/module/qat/elemwise.py

@@ -17,9 +17,7 @@ class Elemwise(Float.Elemwise, QATModule):
     :param method: the elemwise method, see :class:`~.module.elemwise.Elemwise` for detail.
     """
 
-    def __init__(self, method):
-        super().__init__(method)
-        self.with_weight = False
+    with_weight = False
 
     def forward(self, *inps):
         return self.apply_quant_activation(super().forward(*inps))

python_module/megengine/module/qat/module.py

@@ -23,6 +23,9 @@ class QATModule(Module):
     :func:`~.quantize.quantize` further.
     """
 
+    with_weight = True
+    with_act = True
+
     def __init__(self):
         super().__init__()
 
@@ -32,9 +35,6 @@ class QATModule(Module):
         self.weight_fake_quant = None  # type: FakeQuantize
         self.act_fake_quant = None  # type: FakeQuantize
 
-        self.with_weight = True
-        self.with_act = True
-
     def set_qconfig(self, qconfig: QConfig):
         r"""
         Set quantization related configs with ``qconfig``, including
@@ -51,29 +51,21 @@ class QATModule(Module):
             self.weight_observer = safe_call(qconfig.weight_observer)
             self.weight_fake_quant = safe_call(qconfig.weight_fake_quant)
 
+    def _enable_exec(self, with_module, func, enable):
+        if not with_module:
+            return
+        if enable:
+            func.enable()
+        else:
+            func.disable()
+
     def set_fake_quant(self, enable):
-        if self.with_act:
-            if enable:
-                self.act_fake_quant.enable()
-            else:
-                self.act_fake_quant.disable()
-        if self.with_weight:
-            if enable:
-                self.weight_fake_quant.enable()
-            else:
-                self.weight_fake_quant.disable()
+        self._enable_exec(self.with_act, self.act_fake_quant, enable)
+        self._enable_exec(self.with_weight, self.weight_fake_quant, enable)
 
     def set_observer(self, enable):
-        if self.with_act:
-            if enable:
-                self.act_observer.enable()
-            else:
-                self.act_observer.disable()
-        if self.with_weight:
-            if enable:
-                self.weight_observer.enable()
-            else:
-                self.weight_observer.disable()
+        self._enable_exec(self.with_act, self.act_observer, enable)
+        self._enable_exec(self.with_weight, self.weight_observer, enable)
 
     def _apply_fakequant_with_observer(
         self, target: Tensor, fake_quant: FakeQuantize, observer: Observer

python_module/megengine/module/qat/quant_dequant.py

@@ -15,9 +15,7 @@ class QuantStub(Float.QuantStub, QATModule):
     input after converted to :class:`~.QuantizedModule`.
     """
 
-    def __init__(self):
-        super().__init__()
-        self.with_weight = False
+    with_weight = False
 
     def forward(self, inp):
         return self.apply_quant_activation(inp)
@@ -37,10 +35,8 @@ class DequantStub(Float.DequantStub, QATModule):
     input after converted to :class:`~.QuantizedModule`.
     """
 
-    def __init__(self):
-        super().__init__()
-        self.with_weight = False
-        self.with_act = False
+    with_weight = False
+    with_act = False
 
     def forward(self, inp):
         return inp

python_module/megengine/quantization/fake_quant.py

@@ -116,10 +116,11 @@ class TQT(_FakeQuantize):
         return TQT_Function(self.qmin, self.qmax)(inp, self.scale)
 
     def normal_foward(self, inp, q_dict):
-        # when disable, TQT will do normal forward, initialize scale weight
-        tmp_scale = F.maximum(F.abs(q_dict["min_val"]), F.abs(q_dict["max_val"]))
-        tmp_scale = F.log(tmp_scale / 127) / F.log(2)
-        F.add_update(self.scale, tmp_scale, alpha=0.0, beta=1.0, bias=0.0)
+        if q_dict["enable_observer"]:
+            # when disable, TQT will do normal forward, initialize scale weight
+            tmp_scale = F.maximum(F.abs(q_dict["min_val"]), F.abs(q_dict["max_val"]))
+            tmp_scale = F.log(tmp_scale / 127) / F.log(2)
+            F.add_update(self.scale, tmp_scale, alpha=0.0, beta=1.0, bias=0.0)
         return inp
 
     def get_qparams(self):

python_module/megengine/quantization/observer.py

@@ -102,6 +102,7 @@ class MinMaxObserver(Observer):
         q_dict = get_qparam_dict(self.mode)
         q_dict["min_val"] = inp_min_val
         q_dict["max_val"] = inp_max_val
+        q_dict["enable_observer"] = self.enable
         if self.mode == QuantMode.SYMMERTIC:
             symmetric_max_vals = F.maximum(-min_val, max_val)
             # use maximun to avoid scale too small at the begin

python_module/megengine/quantization/quantize.py

@@ -14,6 +14,7 @@ from ..module import qat as QAT
 from ..module import quantized as Quantized
 from ..module.qat import QATModule
 from ..module.quantized import QuantizedModule
+from .fake_quant import TQT
 from .qconfig import QConfig, ema_fakequant_qconfig
 
 
@@ -119,6 +120,14 @@ def quantize_qat(
     return module
 
 
+def _propagate(module: Module, func_str: str, *args, **kargs):
+    def fn(mod: Module):
+        if isinstance(mod, QATModule):
+            getattr(mod, func_str)(*args, **kargs)
+
+    module.apply(fn)
+
+
 def propagate_qconfig(module: QATModule, qconfig: QConfig):
     r"""
     Recursively set ``module``'s qconfig through :meth:`~.Module.apply`.
@@ -126,12 +135,7 @@ def propagate_qconfig(module: QATModule, qconfig: QConfig):
     :param module: root module to traverse recursively.
     :param qconfig: a instance of :class:`~.QConfig` to be set as submodules' qconfig.
     """
-
-    def fn(mod: Module):
-        if isinstance(mod, QATModule):
-            mod.set_qconfig(qconfig)
-
-    module.apply(fn)
+    _propagate(module, "set_qconfig", qconfig)
 
 
 def disable_fake_quant(module: Module):
@@ -141,11 +145,7 @@ def disable_fake_quant(module: Module):
     :param module: root module to do disable fake quantization recursively.
     """
 
-    def fn(mod: Module):
-        if isinstance(mod, QATModule):
-            mod.set_fake_quant(False)
-
-    module.apply(fn)
+    _propagate(module, "set_fake_quant", False)
 
 
 def disable_observer(module: Module):
@@ -155,11 +155,7 @@ def disable_observer(module: Module):
     :param module: root module to do disable observer recursively.
     """
 
-    def fn(mod: Module):
-        if isinstance(mod, QATModule):
-            self.set_observer(False)
-
-    module.apply(fn)
+    _propagate(module, "set_observer", False)
 
 
 def enable_fake_quant(module: Module):
@@ -169,11 +165,7 @@ def enable_fake_quant(module: Module):
     :param module: root module to do enable fake quantization recursively.
     """
 
-    def fn(mod: Module):
-        if isinstance(mod, QATModule):
-            mod.set_fake_quant(True)
-
-    module.apply(fn)
+    _propagate(module, "set_fake_quant", True)
 
 
 def enable_observer(module: Module):
@@ -183,8 +175,4 @@ def enable_observer(module: Module):
     :param module: root module to do enable observer recursively.
     """
 
-    def fn(mod: Module):
-        if isinstance(mod, QATModule):
-            mod.set_observer(True)
-
-    module.apply(fn)
+    _propagate(module, "set_observer", False)