[Performance] Memoryless observers for weights

src/llmcompressor/modifiers/quantization/calibration.py

@@ -43,6 +43,8 @@ def initialize_observer(
     The name is then used to load the observer from the registry and attached
     to the module. The name of the observer uses the base_name provided.
 
+    This function always initializes memoryless observers for weights
+
     :param module: torch.nn.Module that the observer is being attached to
     :param base_name: str used to name the observer attribute
 
@@ -57,9 +59,25 @@ def initialize_observer(
     args: QuantizationArgs = getattr_chain(
         module, f"quantization_scheme.{arg_name}", None
     )
+    observer = args.observer
+
+    # training is no longer supported: always use memoryless for weights
+    if base_name == "weight" and args.observer in ("static_minmax", "minmax"):
+        observer = "memoryless_minmax"
+        logger.warning(
+            "Overriding weight observer for lower memory usage "
+            f"({args.observer} -> {observer})"
+        )
+    if base_name == "weight" and args.observer in ("mse",):
+        observer = "memoryless_mse"
+        logger.warning(
+            "Overriding weight observer for lower memory usage "
+            f"({args.observer} -> {observer})"
+        )
+
     if args is not None and args.dynamic is not True:
         observer = Observer.load_from_registry(
-            args.observer, base_name=base_name, args=args, module=module
+            observer, base_name=base_name, args=args, module=module
         )
         module.register_module(f"{base_name}_observer", observer)