Revert "[Release 2.10] Update to Torch 2.10 - final release (#30525)"

This reverts commit f97ca67176.

tests/compile/test_aot_compile.py

@@ -90,7 +90,9 @@ def use_vllm_config(vllm_config: VllmConfig):
         yield
 
 
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 def test_no_dynamo_cache_entry(monkeypatch: pytest.MonkeyPatch):
     with monkeypatch.context() as m:
         vllm_config = make_vllm_config()
@@ -114,7 +116,9 @@ def test_no_dynamo_cache_entry(monkeypatch: pytest.MonkeyPatch):
             assert torch.allclose(actual, expected)
 
 
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 def test_force_aot_load(monkeypatch: pytest.MonkeyPatch):
     with tempfile.TemporaryDirectory() as tmpdirname, monkeypatch.context() as m:
         args = (torch.randn(10, 10),)
@@ -128,7 +132,9 @@ def test_force_aot_load(monkeypatch: pytest.MonkeyPatch):
             CompiledMod(vllm_config=vllm_config)(*args)
 
 
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 def test_save_and_load(monkeypatch: pytest.MonkeyPatch):
     with monkeypatch.context() as m:
         args = (torch.randn(10, 10),)
@@ -156,7 +162,9 @@ def test_save_and_load(monkeypatch: pytest.MonkeyPatch):
             assert torch.allclose(ret, expected)
 
 
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 def test_cache_load_returns_tuple_consistency(monkeypatch: pytest.MonkeyPatch):
     """
     Test that cache loading correctly handles the returns_tuple logic.
@@ -215,7 +223,9 @@ def test_cache_load_returns_tuple_consistency(monkeypatch: pytest.MonkeyPatch):
             )
 
 
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 def test_cache_load_returns_tuple_consistency_tuple_output(
     monkeypatch: pytest.MonkeyPatch,
 ):
@@ -284,7 +294,9 @@ def test_cache_load_returns_tuple_consistency_tuple_output(
             )
 
 
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 def test_shape_env(monkeypatch: pytest.MonkeyPatch):
     """
     Test that the shape environment is correctly serialized and preserved
@@ -321,7 +333,9 @@ def test_shape_env(monkeypatch: pytest.MonkeyPatch):
                 assert guards_string == " - s77 <= 42\n - Eq(Mod(s77, 2), 0)"
 
 
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 def test_partition_wrapper_applied_on_aot_load(
     monkeypatch: pytest.MonkeyPatch, vllm_tmp_cache: Path, mocker
 ):
@@ -412,7 +426,9 @@ def test_partition_wrapper_applied_on_aot_load(
         )
 
 
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 @create_new_process_for_each_test("spawn")
 def test_gpt2_cache_hit(monkeypatch: pytest.MonkeyPatch):
     """
@@ -476,7 +492,9 @@ def test_gpt2_cache_hit(monkeypatch: pytest.MonkeyPatch):
         symbolic_shapes_module.make_symbol = original_make_symbol
 
 
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 class TestStandaloneCompiledArtifacts:
     def test_init(self):
         cache = StandaloneCompiledArtifacts()
@@ -650,7 +668,9 @@ class TestStandaloneCompiledArtifacts:
         assert len(restored_cache.loaded_submodule_store) == 0
 
 
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 class TestStandaloneCompiledArtifactsIntegration:
     def test_add_pickle_unpickle(self):
         cache = StandaloneCompiledArtifacts()

tests/compile/test_dynamic_shapes_compilation.py

@@ -39,7 +39,9 @@ def get_test_models():
 @pytest.mark.parametrize("use_aot_compile", ["0", "1"])
 @pytest.mark.parametrize("use_bytecode_hook", [True, False])
 @pytest.mark.parametrize("evaluate_guards", [False, True])
-@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
+@pytest.mark.skipif(
+    not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
+)
 def test_dynamic_shapes_compilation(
     monkeypatch,
     model_name,

tests/kernels/moe/test_shared_fused_moe_routed_transform.py

@@ -14,7 +14,6 @@ import torch.nn as nn
 from vllm.config import VllmConfig, set_current_vllm_config
 from vllm.forward_context import set_forward_context
 from vllm.model_executor.layers.fused_moe.shared_fused_moe import SharedFusedMoE
-from vllm.utils.torch_utils import is_torch_equal_or_newer
 
 
 class SimpleLinear(nn.Module):
@@ -61,10 +60,6 @@ def setup_cuda():
 @pytest.mark.parametrize("num_tokens", [1, 32])
 @pytest.mark.parametrize("hidden_size,latent_size", [(256, 128), (128, 64)])
 @pytest.mark.parametrize("dtype", [torch.bfloat16])
-@pytest.mark.skipif(
-    is_torch_equal_or_newer("2.10.0"),
-    reason="Test fails with PyTorch 2.10.0 see: https://github.com/vllm-project/vllm/issues/33995",
-)
 def test_routed_input_transform_inside_vs_outside(
     num_tokens: int,
     hidden_size: int,