[torch.compile] PyTorch 2.6 and nightly compatibility (vllm-project#12393)

Signed-off-by: youkaichao <[email protected]>
Signed-off-by: SzymonOzog <[email protected]>

Author: youkaichao

tests/compile/piecewise/test_simple.py

@@ -92,7 +92,7 @@ def test_simple_piecewise_compile():
             num_graphs_seen=1,  # one graph for the model
             num_piecewise_graphs_seen=5,  # 2 * num_layers + 1
             num_piecewise_capturable_graphs_seen=3,  # 1 + num_layers
-            num_inductor_compilations=3,  # num_piecewise_capturable_graphs_seen
+            num_backend_compilations=3,  # num_piecewise_capturable_graphs_seen
             num_cudagraph_caputured=
             6,  # num_cudagraph_sizes * num_piecewise_capturable_graphs_seen
     ):

tests/compile/piecewise/test_toy_llama.py

@@ -322,7 +322,7 @@ def test_toy_llama():
             num_graphs_seen=0,
             num_piecewise_graphs_seen=0,
             num_piecewise_capturable_graphs_seen=0,
-            num_inductor_compilations=0,
+            num_backend_compilations=0,
             num_cudagraph_caputured=0,
     ):
         outputs.append(run_model(llama_config, use_compile=False))
@@ -332,7 +332,7 @@ def test_toy_llama():
             num_graphs_seen=1,  # one graph for the model
             num_piecewise_graphs_seen=1,
             num_piecewise_capturable_graphs_seen=1,
-            num_inductor_compilations=1,  # num_piecewise_capturable_graphs_seen
+            num_backend_compilations=1,  # num_piecewise_capturable_graphs_seen
             num_cudagraph_caputured=
             2,  # num_cudagraph_sizes * num_piecewise_capturable_graphs_seen
     ):
@@ -345,7 +345,7 @@ def test_toy_llama():
             1,  # 2 * num_layers + 1
             num_piecewise_capturable_graphs_seen=1 +
             llama_config.num_layers,  # 1 + num_layers
-            num_inductor_compilations=1 +
+            num_backend_compilations=1 +
             llama_config.num_layers,  # num_piecewise_capturable_graphs_seen
             num_cudagraph_caputured=2 *
         (1 + llama_config.num_layers