diff --git a/gptqmodel/looper/awq_processor.py b/gptqmodel/looper/awq_processor.py
index 6ad426586..3ad6b8a51 100644
--- a/gptqmodel/looper/awq_processor.py
+++ b/gptqmodel/looper/awq_processor.py
@@ -709,12 +709,18 @@ def _apply_quant(self, module, named_linears: Dict[str, NamedModule], start_time
             # records
             duration = time.time() - start_time
 
-            avg_loss = 999999999
+            avg_loss_value = None
             for _, layer_names, _, loss in scales_list:
                 if any(named_module.name in layer_name for layer_name in layer_names):
-                    avg_loss = loss
+                    avg_loss_value = loss
                     break
 
+            if avg_loss_value is None:
+                # Scaling not applied for this layer in AWQ; no meaningful loss metric.
+                loss_summary = "not applicable"
+            else:
+                loss_summary = f"{avg_loss_value:.10f}"
+
             # TODO "loss" and "nsamples" may not be consistent with the semantics of gptq quantization.
             stat = {
                 PROCESS_LOG_NAME: self.name(),
@@ -722,7 +728,7 @@ def _apply_quant(self, module, named_linears: Dict[str, NamedModule], start_time
                 PROCESS_LOG_MODULE: named_module.name,
                 MODULE_FEATURE_COLUMN: self.module_feature_summary(named_module),
                 DTYPE_SIZE_COLUMN: self.module_dtype_size_summary(named_module),
-                QUANT_LOG_LOSS: f"{avg_loss:.10f}",
+                QUANT_LOG_LOSS: loss_summary,
                 QUANT_LOG_NSAMPLES: f"{self.nsamples}",
                 # QUANT_LOG_DAMP: f"{damp_percent:.5f}",
                 PROCESS_LOG_TIME: f"{duration:.3f}",
diff --git a/tests/models/test_multi_vs_single_gpu.py b/tests/models/test_multi_vs_single_gpu.py
index 0a58bd8ca..2a8d4254d 100644
--- a/tests/models/test_multi_vs_single_gpu.py
+++ b/tests/models/test_multi_vs_single_gpu.py
@@ -9,7 +9,7 @@
 import sys
 from contextlib import ExitStack
 from dataclasses import dataclass
-from decimal import Decimal
+from decimal import Decimal, InvalidOperation
 from typing import Dict, Iterable, List, Tuple
 from unittest import mock
 
@@ -262,9 +262,14 @@ def _extract_layer_metrics(
             if loss_value is None or sample_value is None:
                 continue
 
+            try:
+                loss_decimal = Decimal(loss_value)
+            except (InvalidOperation, TypeError, ValueError):
+                continue
+
             per_layer = layer_metrics.setdefault(layer_index, {})
             per_layer[module_name] = LayerMetrics(
-                loss=Decimal(loss_value),
+                loss=loss_decimal,
                 samples=int(sample_value),
             )
         return layer_metrics