Fix issue with cuda.

Details: vllm-project#963

Author: Vladimir Leskov

src/llmcompressor/modifiers/utils/pytorch_helpers.py

@@ -102,7 +102,7 @@ def run_calibration_forward(
 
         # TODO: not ideal, figure out where we aren't freeing memory instead
         # currently without this we run OOM on the 2nd forward pass
-        torch.cuda.empty_cache()
+        # torch.cuda.empty_cache()
 
     return intermediates