propagate similar to the pipeline tests.

tests/pipelines/test_pipelines.py

@@ -2334,21 +2334,21 @@ def test_hotswapping_pipeline(self, rank0, rank1):
     def test_hotswapping_compiled_pipline_linear(self, rank0, rank1):
         # It's important to add this context to raise an error on recompilation
         target_modules = ["to_q", "to_k", "to_v", "to_out.0"]
-        with torch._dynamo.config.patch(error_on_recompile=True):
+        with torch._dynamo.config.patch(error_on_recompile=True), torch._inductor.utils.fresh_inductor_cache():
             self.check_pipeline_hotswap(do_compile=True, rank0=rank0, rank1=rank1, target_modules0=target_modules)
 
     @parameterized.expand([(11, 11), (7, 13), (13, 7)])  # important to test small to large and vice versa
     def test_hotswapping_compiled_pipline_conv2d(self, rank0, rank1):
         # It's important to add this context to raise an error on recompilation
         target_modules = ["conv", "conv1", "conv2"]
-        with torch._dynamo.config.patch(error_on_recompile=True):
+        with torch._dynamo.config.patch(error_on_recompile=True), torch._inductor.utils.fresh_inductor_cache():
             self.check_pipeline_hotswap(do_compile=True, rank0=rank0, rank1=rank1, target_modules0=target_modules)
 
     @parameterized.expand([(11, 11), (7, 13), (13, 7)])  # important to test small to large and vice versa
     def test_hotswapping_compiled_pipline_both_linear_and_conv2d(self, rank0, rank1):
         # It's important to add this context to raise an error on recompilation
         target_modules = ["to_q", "conv"]
-        with torch._dynamo.config.patch(error_on_recompile=True):
+        with torch._dynamo.config.patch(error_on_recompile=True), torch._inductor.utils.fresh_inductor_cache():
             self.check_pipeline_hotswap(do_compile=True, rank0=rank0, rank1=rank1, target_modules0=target_modules)
 
     def test_enable_lora_hotswap_called_after_adapter_added_raises(self):