fix traced model patch check

optimum/intel/ipex/modeling_base.py

@@ -74,8 +74,8 @@ def _is_patched_with_ipex(model, task):
 
     if isinstance(model, torch.jit.ScriptModule):
         for node in model.graph.nodes():
-            # Only patched model enabled tpp linear by _enable_tpp().
-            if "torch_ipex::tpp_linear" in node.__str__():
+            # Only patched model enabled fusion linear.
+            if "/fusions/" in node.__str__():
                 return True
         return False
     elif task in _IPEX_EXPORTED_GENERATION_TASKS and model.config.hidden_size < 64:
@@ -99,8 +99,6 @@ def ipex_jit_trace(model, task, use_cache):
 
     if _is_patched_with_ipex(model, task):
         model = _patch_model(model)
-        # Use Tensor Processing Primitives to accelerate linear, see https://arxiv.org/abs/2104.05755.
-        _enable_tpp()
 
     sample_inputs = _prepare_inputs_for_ipex_model(model, task, use_cache)
 
@@ -111,6 +109,8 @@ def ipex_jit_trace(model, task, use_cache):
         if not use_cache:
             sample_inputs.pop("past_key_values")
 
+    # Use Tensor Processing Primitives to accelerate linear, see https://arxiv.org/abs/2104.05755.
+    _enable_tpp()
     model = ipex.optimize(model.eval(), dtype=model.dtype, inplace=True)
     # Disable repack while jit tracing to reduce the memory
     ipex._C.disable_jit_linear_repack()