Merge branch 'main' into fix_sparsity_test

vllm-project · Oct 31, 2024 · 30c2315 · 30c2315
2 parents d4affb2 + 60c766f
commit 30c2315
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/src/llmcompressor/modifiers/quantization/gptq/utils/gptq_wrapper.py b/src/llmcompressor/modifiers/quantization/gptq/utils/gptq_wrapper.py
@@ -10,6 +10,7 @@
 
 from llmcompressor.modifiers.utils import SPARSITY_THRESHOLD
 from llmcompressor.modifiers.utils.compression_wrapper import ModuleCompressionWrapper
+from llmcompressor.observers import Observer
 from llmcompressor.pytorch.utils.helpers import tensor_sparsity
 from llmcompressor.utils import getattr_chain
 from llmcompressor.utils.metric_logging import (
@@ -303,6 +304,7 @@ def _update_quantization_parameters(self, args: QuantizationArgs, W: torch.Tenso
         :param W: weight to calculate quantization parameters from
         """
         observer = args.get_observer()
+        observer = Observer.load_from_registry(observer, quantization_args=args)
         _scale, _zero_point = observer(W, g_idx=None)
         update_parameter_data(self.layer, _scale, "weight_scale")
         update_parameter_data(self.layer, _zero_point, "weight_zero_point")