biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

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

Browse Source
This commit is contained in:
Andrey Talman
2026-02-08 16:51:09 -05:00
committed by GitHub
parent 084aa19f02
commit f97ca67176
17 changed files with 130 additions and 78 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
tests/compile/test_aot_compile.py
View File

@@ -90,9 +90,7 @@ def use_vllm_config(vllm_config: VllmConfig):
yield
@pytest.mark.skipif(
not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
def test_no_dynamo_cache_entry(monkeypatch: pytest.MonkeyPatch):
with monkeypatch.context() as m:
vllm_config = make_vllm_config()
@@ -116,9 +114,7 @@ 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.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), 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),)
@@ -132,9 +128,7 @@ 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.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
def test_save_and_load(monkeypatch: pytest.MonkeyPatch):
with monkeypatch.context() as m:
args = (torch.randn(10, 10),)
@@ -162,9 +156,7 @@ 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.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), 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.
@@ -223,9 +215,7 @@ def test_cache_load_returns_tuple_consistency(monkeypatch: pytest.MonkeyPatch):
)
@pytest.mark.skipif(
not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
def test_cache_load_returns_tuple_consistency_tuple_output(
monkeypatch: pytest.MonkeyPatch,
):
@@ -294,9 +284,7 @@ def test_cache_load_returns_tuple_consistency_tuple_output(
)
@pytest.mark.skipif(
not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
def test_shape_env(monkeypatch: pytest.MonkeyPatch):
"""
Test that the shape environment is correctly serialized and preserved
@@ -333,9 +321,7 @@ 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.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
def test_partition_wrapper_applied_on_aot_load(
monkeypatch: pytest.MonkeyPatch, vllm_tmp_cache: Path, mocker
):
@@ -426,9 +412,7 @@ def test_partition_wrapper_applied_on_aot_load(
)
@pytest.mark.skipif(
not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
@create_new_process_for_each_test("spawn")
def test_gpt2_cache_hit(monkeypatch: pytest.MonkeyPatch):
"""
@@ -492,9 +476,7 @@ 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.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
class TestStandaloneCompiledArtifacts:
def test_init(self):
cache = StandaloneCompiledArtifacts()
@@ -668,9 +650,7 @@ class TestStandaloneCompiledArtifacts:
assert len(restored_cache.loaded_submodule_store) == 0
@pytest.mark.skipif(
not is_torch_equal_or_newer("2.10.0.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
class TestStandaloneCompiledArtifactsIntegration:
def test_add_pickle_unpickle(self):
cache = StandaloneCompiledArtifacts()

4
tests/compile/test_dynamic_shapes_compilation.py
View File

@@ -39,9 +39,7 @@ 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.dev"), reason="requires torch 2.10"
)
@pytest.mark.skipif(not is_torch_equal_or_newer("2.10.0"), reason="requires torch 2.10")
def test_dynamic_shapes_compilation(
monkeypatch,
model_name,

5
tests/kernels/moe/test_shared_fused_moe_routed_transform.py
View File

@@ -14,6 +14,7 @@ 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):
@@ -60,6 +61,10 @@ 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,