GPTQ - move calibration of quantiztion params to after hessian calibration

src/llmcompressor/modifiers/quantization/gptq/base.py

@@ -1,7 +1,12 @@
 from typing import Any, Dict, Iterable, List, Optional, Tuple, Union
 
 import torch
-from compressed_tensors.quantization import QuantizationScheme
+from compressed_tensors.quantization import (
+    QuantizationScheme,
+    disable_quantization,
+    enable_quantization,
+    freeze_module_quantization,
+)
 from loguru import logger
 from pydantic import Field
 from torch.nn import Module
@@ -163,7 +168,9 @@ def on_initialize(self, state: "State", **kwargs) -> bool:
         if not self.initialized_structure_:
             self.on_initialize_structure(state, **kwargs)
         if self.quantization_modifier_:
-            self.quantization_modifier_.initialize(state, **kwargs)
+            self.quantization_modifier_.initialize(
+                state, freeze_quantization=False, **kwargs
+            )
         if not self.quantize:
             raise ValueError("To use the GPTQModifier, quantization must be enabled.")
 
@@ -178,6 +185,7 @@ def on_initialize(self, state: "State", **kwargs) -> bool:
 
         self.initialize_compression(modifiable_model, calibration_dataloader)
         self.apply_compression(calibration_dataloader)
+        state.model.apply(freeze_module_quantization)
 
         return True
 
@@ -250,6 +258,11 @@ def apply_compression(
         logger.info(
             f"Running {class_name} calibration with " f"{len(dataloader)} samples..."
         )
+
+        # quantization scales and zp are already initialized but we do not
+        # want to calibrate wrt to these
+        self.model.apply(disable_quantization)
+
         if not self.sequential_update:
             # in non-sequential mode we run one forward batch for all modules
             run_calibration_forward(self.model, dataloader, mask_padding=True)
@@ -271,6 +284,9 @@ def apply_compression(
             layer_compressor.revert_layer_wrappers()
             torch.cuda.empty_cache()
 
+        # re-enable quantization
+        self.model.apply(enable_quantization)
+
     def _build_quant_modifier(self):
         """
         Build a quantization modifier based on the specified config_groups,

src/llmcompressor/modifiers/quantization/gptq/utils/gptq_wrapper.py

@@ -141,6 +141,14 @@ def compress(
                 elif hasattr(self.layer, "quantization_scheme"):
                     quant_scheme = self.layer.quantization_scheme
                     if quant_scheme.weights is not None:
+                        # fetch latest correct scale and ZP relevant for any changes
+                        # such as activation reordering
+                        from compressed_tensors.quantization import (
+                            update_layer_weight_quant_params,
+                        )
+
+                        update_layer_weight_quant_params(self.layer)
+
                         scale = self.layer.weight_scale
                         zero_point = self.layer.weight_zero_point
                         from compressed_tensors.quantization import QuantizationStrategy

src/llmcompressor/modifiers/quantization/quantization/base.py

@@ -61,7 +61,9 @@ def on_initialize_structure(self, state: State, **kwargs):
         self._apply_modifier_to_model(module)
         module.apply(freeze_module_quantization)
 
-    def on_initialize(self, state: State, **kwargs) -> bool:
+    def on_initialize(
+        self, state: State, freeze_quantization: bool = True, **kwargs
+    ) -> bool:
         if self.end and self.end != -1:
             raise ValueError(
                 "end_epoch is disabled for QuantizationModifier and can only be set to"
@@ -80,7 +82,8 @@ def on_initialize(self, state: State, **kwargs) -> bool:
             self._check_token_distribution(
                 module, threshold=kwargs.get("min_tokens_per_module")
             )
-            module.apply(freeze_module_quantization)
+            if freeze_quantization:
+                module.apply(freeze_module_quantization)
 
         return True