[Rebase] Rebase to vLLM 0.29.0 (#7230)

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
Signed-off-by: Zhou Taichang <tzhouam@connect.ust.hk>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Zhou TaichangandClaude Opus 5 authored and GitHub committed 2026-09-10 12:08:28 +08:00
1 parent 333f14e97e
commit aff7d64948
120 files changed
+2958 -1690

No files matched your search

+2 -2
View File
@@ -1,5 +1,5 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
"""Unit tests for shared AdaLayerNorm layers used by FLUX and other models."""
import os
@@ -46,7 +46,7 @@ def _force_default_gemm(monkeypatch):
monkeypatch.setattr(
"vllm.model_executor.layers.linear.dispatch_unquantized_gemm",
lambda: default_unquantized_gemm,
lambda *_args, **_kwargs: default_unquantized_gemm,
)
+25 -2
View File
@@ -1,5 +1,5 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
"""
Shape broadcasting regression tests for RotaryEmbedding.forward_native.
@@ -16,7 +16,11 @@ import types
import pytest
import torch
from vllm_omni.diffusion.layers.rope import RotaryEmbedding, apply_rotary_emb_mindiesd
from vllm_omni.diffusion.layers.rope import (
RotaryEmbedding,
RotaryEmbeddingWan,
apply_rotary_emb_mindiesd,
)
pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
@@ -239,6 +243,25 @@ class TestRotaryEmbeddingNativeShapeRegression:
assert not torch.isnan(output).any()
def test_cuda_paths_fall_back_without_vendored_rotary_kernel() -> None:
"""Source/CPU installs need not contain the wheel-vendored CUDA kernel."""
x = torch.randn(1, 5, 2, 8)
rope = RotaryEmbedding(is_neox_style=True)
rope.apply_rotary_emb_vllm_flash_attn = None
output = rope.forward_cuda(x, torch.randn(5, 4), torch.randn(5, 4))
assert output.shape == x.shape
wan_rope = RotaryEmbeddingWan(is_neox_style=False, half_head_dim=True)
wan_rope.apply_rotary_emb_vllm_flash_attn = None
output = wan_rope.forward_cuda(
x,
torch.randn(1, 5, 1, 4),
torch.randn(1, 5, 1, 4),
)
assert output.shape == x.shape
def test_mindie_full_dim_rope_keeps_h3_cos_sin_layout(monkeypatch: pytest.MonkeyPatch) -> None:
"""H3 passes complete rotary dims to MindIE instead of the CUDA half-dim layout."""
captured: dict[str, object] = {}