[Main2Main] Upgrade to newest vLLM 0204

.github/workflows/_pre_commit.yml

@@ -38,6 +38,7 @@ jobs:
         repository: vllm-project/vllm
         path: ./vllm-empty
         ref: ${{ inputs.vllm }}
+
     - uses: dorny/paths-filter@v3
       id: filter
       with:
@@ -62,10 +63,12 @@ jobs:
       run: |
         git config --global --add safe.directory /__w/vllm-ascend/vllm-ascend
         pre-commit run --all-files --hook-stage manual --show-diff-on-failure
+
     - name: Run mypy
       run: |
         PYTHONPATH="$PYTHONPATH:$(pwd)/vllm-empty"
         export PYTHONPATH
+        env
         git config --global --add safe.directory /__w/vllm-ascend/vllm-ascend
         # Run mypy for Python 3.10, 3.11, 3.12 manually
         # Note: We are now separating mypy from pre-commit hooks for performance reasons.

.github/workflows/bot_pr_create.yaml

@@ -37,7 +37,7 @@ jobs:
     steps:
       - name: Get vLLM version
         run: |
-          VLLM_COMMIT=v0.15.0
+          VLLM_COMMIT=80f921ba4bab2ea251d149305ea0f912c6fc218a
           echo "VLLM_COMMIT=https://github.com/vllm-project/vllm/commit/$VLLM_COMMIT" >> "$GITHUB_ENV"
 
       - name: Checkout repository

.github/workflows/dockerfiles/Dockerfile.lint

@@ -27,7 +27,7 @@ RUN apt-get update -y && \
 
 ARG VLLM_REPO=https://github.com/vllm-project/vllm.git
 # For lint purpose, actually we need make a main2main matching.
-ARG VLLM_COMMIT=v0.15.0
+ARG VLLM_COMMIT=80f921ba4bab2ea251d149305ea0f912c6fc218a
 RUN git clone $VLLM_REPO /vllm-workspace/vllm && \
     cd /vllm-workspace/vllm && \
     git checkout $VLLM_COMMIT

.github/workflows/pr_test_full.yaml

@@ -75,7 +75,7 @@ jobs:
     name: e2e-full
     strategy:
       matrix:
-        vllm_version: [v0.15.0]
+        vllm_version: [80f921ba4bab2ea251d149305ea0f912c6fc218a, v0.15.0]
     needs: [changes]
     if: ${{ needs.changes.outputs.e2e_tracker == 'true' || needs.changes.outputs.e2e_tracker == true }}
     uses: ./.github/workflows/_e2e_test.yaml

.github/workflows/pr_test_light.yaml

@@ -41,7 +41,7 @@ jobs:
   lint:
     uses: ./.github/workflows/_pre_commit.yml
     with:
-      vllm: v0.15.0
+      vllm: 80f921ba4bab2ea251d149305ea0f912c6fc218a
   changes:
     runs-on: linux-aarch64-a2-0
     outputs:
@@ -87,7 +87,7 @@ jobs:
     if: ${{ needs.lint.result == 'success' && (needs.changes.outputs.e2e_tracker == 'true' || needs.changes.outputs.ut_tracker == 'true') }}
     strategy:
       matrix:
-        vllm_version: [v0.15.0]
+        vllm_version: [80f921ba4bab2ea251d149305ea0f912c6fc218a, v0.15.0]
     uses: ./.github/workflows/_unit_test.yaml
     with:
       vllm: ${{ matrix.vllm_version }}
@@ -99,7 +99,7 @@ jobs:
     name: e2e-light
     strategy:
       matrix:
-        vllm_version: [v0.15.0]
+        vllm_version: [80f921ba4bab2ea251d149305ea0f912c6fc218a, v0.15.0]
     # Note (yikun): If CI resource are limited we can split job into two chain jobs
     needs: [lint, changes]
     # only trigger e2e test after lint passed and the change is e2e related with pull request.

.github/workflows/schedule_codecov_refresh.yaml

@@ -33,7 +33,7 @@ jobs:
     name: refresh codecov
     strategy:
       matrix:
-        vllm_version: [v0.15.0]
+        vllm_version: [80f921ba4bab2ea251d149305ea0f912c6fc218a]
     uses: ./.github/workflows/_unit_test.yaml
     with:
       vllm: ${{ matrix.vllm_version }}

docs/source/community/versioning_policy.md

@@ -55,7 +55,7 @@ For main branch of vLLM Ascend, we usually make it compatible with the latest vL
 
 | vLLM Ascend | vLLM         | Python           | Stable CANN | PyTorch/torch_npu  |
 |-------------|--------------|------------------|-------------|--------------------|
-|     main    | v0.15.0 tag | >= 3.10, < 3.12   | 8.5.0 | 2.9.0 / 2.9.0 |
+|     main    | 80f921ba4bab2ea251d149305ea0f912c6fc218a, v0.15.0 tag | >= 3.10, < 3.12   | 8.5.0 | 2.9.0 / 2.9.0 |
 
 ## Release cadence
 

tests/ut/eplb/core/test_eplb_utils.py

@@ -9,6 +9,7 @@
 
 from vllm_ascend.ascend_config import init_ascend_config
 from vllm_ascend.eplb.core.eplb_utils import init_eplb_config
+from vllm_ascend.utils import vllm_version_is
 # isort: on
 
 
@@ -21,7 +22,13 @@ def setUp(self, mock_fix_incompatible_config):
             "eplb_config": {"dynamic_eplb": True, "num_redundant_experts": 2},
         }
         from vllm.model_executor.layers.fused_moe.config import RoutingMethodType
-        moe_parallel_config = FusedMoEParallelConfig(2, 0, 1, 2, 1, 1, 1, 1, True, "hccl", enable_eplb=True)
+        if vllm_version_is("0.15.0"):
+            moe_parallel_config = FusedMoEParallelConfig(
+                2, 0, 1, 2, 1, 1, 1, 1, True, "hccl", enable_eplb=True)
+        else:
+            moe_parallel_config = FusedMoEParallelConfig(
+                2, 0, 1, 2, 1, 1, 1, 1, True, "hccl",
+                is_sequence_parallel=False, enable_eplb=True)
         moe_config = FusedMoEConfig(
             num_experts=8,
             experts_per_token=8,

tests/ut/ops/test_mla.py

@@ -82,8 +82,13 @@ def setUp(self):
     @patch("vllm_ascend.ops.mla.get_tensor_model_parallel_world_size")
     def test_initialization(self, mock_tp_size, mock_ascend_config,
                             mock_get_vllm_config):
+        # Create a proper mock for MLAAttention that has the required attributes
+        mock_mla_attn = MagicMock()
+        mock_mla_attn.process_weights_after_loading = MagicMock()
+        mock_mla_attn.impl = MagicMock()
+        mock_mla_attn.impl.process_weights_after_loading = MagicMock()
 
-        with patch("vllm_ascend.ops.mla.MLAAttention", return_value=True):
+        with patch("vllm_ascend.ops.mla.MLAAttention", return_value=mock_mla_attn):
             mock_tp_size.return_value = 2
             mock_ascend_config.return_value.enable_shared_expert_dp = True
             mock_vllm_config = MagicMock(spec=VllmConfig)
@@ -126,7 +131,14 @@ def test_forward(self, mock_get_forward_context, mock_tp_size,
             num_hidden_layers=32, first_k_dense_replace=False)
         mock_get_vllm_config.return_value = mock_vllm_config
         mock_vllm_config.compilation_config = CompilationConfig()
-        with patch("vllm_ascend.ops.mla.MLAAttention", return_value=True):
+
+        # Create a proper mock for MLAAttention that has the required attributes
+        mock_mla_attn = MagicMock()
+        mock_mla_attn.process_weights_after_loading = MagicMock()
+        mock_mla_attn.impl = MagicMock()
+        mock_mla_attn.impl.process_weights_after_loading = MagicMock()
+
+        with patch("vllm_ascend.ops.mla.MLAAttention", return_value=mock_mla_attn):
             attn = AscendMultiHeadLatentAttention(
                 hidden_size=self.hidden_size,
                 num_heads=self.num_heads,

tests/ut/quantization/test_modelslim_config.py

@@ -1,14 +1,18 @@
 from unittest.mock import MagicMock, patch
 
-from vllm.attention.layer import Attention
 from vllm.model_executor.layers.fused_moe import FusedMoE
 from vllm.model_executor.layers.fused_moe.config import FusedMoEConfig
 from vllm.model_executor.layers.linear import LinearBase
 
 from tests.ut.base import TestBase
 from vllm_ascend.ops.linear import AscendUnquantizedLinearMethod
 from vllm_ascend.quantization.modelslim_config import AscendModelSlimConfig
-from vllm_ascend.utils import ASCEND_QUANTIZATION_METHOD
+from vllm_ascend.utils import ASCEND_QUANTIZATION_METHOD, vllm_version_is
+
+if vllm_version_is("v0.15.0"):
+    from vllm.attention.layer import Attention  # type: ignore
+else:
+    from vllm.model_executor.layers.attention import Attention
 
 
 class TestAscendModelSlimConfig(TestBase):

tests/ut/spec_decode/test_eagle_proposer.py

@@ -28,12 +28,15 @@ def setUp(self):
         self.vllm_config.model_config.dtype = torch.float16
         self.vllm_config.model_config.max_model_len = 2048
         self.vllm_config.model_config.uses_mrope = False
+        self.vllm_config.model_config.uses_xdrope_dim = 0
         self.vllm_config.parallel_config.tensor_parallel_size = 1
+        self.vllm_config.parallel_config.data_parallel_rank = 0
         self.vllm_config.speculative_config.draft_tensor_parallel_size = 1
         self.vllm_config.speculative_config.num_speculative_tokens = 2
         self.vllm_config.speculative_config.speculative_token_tree = str([
             (i + 1) * (0, ) for i in range(2)
         ])
+        self.vllm_config.speculative_config.draft_model_config.uses_xdrope_dim = 0
         self.vllm_config.additional_config = None
 
         self.mock_cpugpubuffer = patch(
@@ -141,12 +144,15 @@ def setUp(self):
         self.vllm_config.model_config.dtype = torch.float16
         self.vllm_config.model_config.max_model_len = 2048
         self.vllm_config.model_config.uses_mrope = False
+        self.vllm_config.model_config.uses_xdrope_dim = 0
         self.vllm_config.parallel_config.tensor_parallel_size = 1
+        self.vllm_config.parallel_config.data_parallel_rank = 0
         self.vllm_config.speculative_config.draft_tensor_parallel_size = 1
         self.vllm_config.speculative_config.num_speculative_tokens = 2
         self.vllm_config.speculative_config.speculative_token_tree = str([
             (i + 1) * (0, ) for i in range(2)
         ])
+        self.vllm_config.speculative_config.draft_model_config.uses_xdrope_dim = 0
         self.vllm_config.additional_config = None
         init_ascend_config(self.vllm_config)
 
@@ -285,12 +291,15 @@ def setUp(self):
         self.vllm_config.model_config.dtype = torch.float16
         self.vllm_config.model_config.max_model_len = 2048
         self.vllm_config.model_config.uses_mrope = False
+        self.vllm_config.model_config.uses_xdrope_dim = 0
         self.vllm_config.model_config.use_mla = False
         self.vllm_config.parallel_config.tensor_parallel_size = 1
+        self.vllm_config.parallel_config.data_parallel_rank = 0
         self.vllm_config.speculative_config.draft_tensor_parallel_size = 1
         self.vllm_config.speculative_config.speculative_token_tree = str([
             (i + 1) * (0, ) for i in range(4)
         ])
+        self.vllm_config.speculative_config.draft_model_config.uses_xdrope_dim = 0
         self.vllm_config.additional_config = None
         init_ascend_config(self.vllm_config)
 
@@ -404,12 +413,15 @@ def setUp(self):
         self.vllm_config.model_config.dtype = torch.float16
         self.vllm_config.model_config.max_model_len = 2048
         self.vllm_config.model_config.uses_mrope = False
+        self.vllm_config.model_config.uses_xdrope_dim = 0
         self.vllm_config.parallel_config.tensor_parallel_size = 1
+        self.vllm_config.parallel_config.data_parallel_rank = 0
         self.vllm_config.speculative_config.draft_tensor_parallel_size = 1
         self.vllm_config.speculative_config.num_speculative_tokens = 2
         self.vllm_config.speculative_config.speculative_token_tree = str([
             (i + 1) * (0, ) for i in range(2)
         ])
+        self.vllm_config.speculative_config.draft_model_config.uses_xdrope_dim = 0
         self.vllm_config.additional_config = None
         init_ascend_config(self.vllm_config)
 

tests/ut/spec_decode/test_mtp_proposer.py

@@ -34,6 +34,7 @@ def vllm_config(self):
         config.speculative_config.draft_model_config = MagicMock()
         config.speculative_config.draft_model_config.get_hidden_size.return_value = 4096
         config.speculative_config.draft_model_config.uses_mrope = False
+        config.speculative_config.draft_model_config.uses_xdrope_dim = 0
         config.speculative_config.speculative_token_tree = str([
             (i + 1) * (0, ) for i in range(2)
         ])
@@ -42,9 +43,11 @@ def vllm_config(self):
         config.model_config.dtype = torch.float16
         config.model_config.max_model_len = 2048
         config.model_config.uses_mrope = False
+        config.model_config.uses_xdrope_dim = 0
         config.model_config.hf_text_config = None
         config.model_config.hf_config = None
         config.parallel_config.tensor_parallel_size = 1
+        config.parallel_config.data_parallel_rank = 0
         config.speculative_config.draft_tensor_parallel_size = 1
 
         config.load_config = None

vllm_ascend/attention/mla_v1.py

@@ -1450,6 +1450,28 @@ def _mla_preprocess(self, layer_name, hidden_states, kv_cache, attn_metadata, ne
     def get_num_actual_tokens(self, attn_metadata: M):
         return attn_metadata.num_actual_tokens
 
+    def forward_mha(
+        self,
+        layer_name: str,
+        hidden_states: torch.Tensor,
+        kv_cache: tuple[torch.Tensor],
+        attn_metadata: M,
+        need_gather_q_kv: bool = False,
+        output: torch.Tensor | None = None,
+    ) -> torch.Tensor:
+        raise NotImplementedError("forward_mha is not supported for MLA attention. Use forward() instead.")
+
+    def forward_mqa(
+        self,
+        layer_name: str,
+        hidden_states: torch.Tensor,
+        kv_cache: tuple[torch.Tensor],
+        attn_metadata: M,
+        need_gather_q_kv: bool = False,
+        output: torch.Tensor | None = None,
+    ) -> torch.Tensor:
+        raise NotImplementedError("forward_mqa is not supported for MLA attention. Use forward() instead.")
+
     def forward(
         self,
         layer_name,

vllm_ascend/attention/sfa_v1.py

@@ -1062,3 +1062,24 @@ def _handle_o_proj_weight_switch_and_forward(
             torch.distributed.all_to_all_single(attn_output, send, group=get_tp_group().device_group)
 
             return attn_output, True
+
+    def forward_mha(
+        self,
+        q: torch.Tensor,
+        kv_c_normed: torch.Tensor,
+        k_pe: torch.Tensor,
+        kv_c_and_k_pe_cache: torch.Tensor,
+        attn_metadata: M,
+        k_scale: torch.Tensor,
+        output: torch.Tensor,
+    ) -> None:
+        raise NotImplementedError("forward_mha is not supported for SFA attention. Use forward() instead.")
+
+    def forward_mqa(
+        self,
+        q: torch.Tensor | tuple[torch.Tensor, torch.Tensor],
+        kv_c_and_k_pe_cache: torch.Tensor,
+        attn_metadata: M,
+        layer,
+    ) -> tuple[torch.Tensor, torch.Tensor | None]:
+        raise NotImplementedError("forward_mqa is not supported for SFA attention. Use forward() instead.")

vllm_ascend/compilation/npugraph_ex_passes/graphex_qknorm_rope_fusion_pass.py

@@ -18,7 +18,6 @@
 
 import torch
 import torchair
-from vllm.attention.layer import Attention
 from vllm.config import VllmConfig, get_layers_from_vllm_config
 from vllm.config.compilation import Range
 from vllm.logger import logger
@@ -27,6 +26,12 @@
     check_and_register_fusion_pass,
     extra_stream_scope_check,
 )
+from vllm_ascend.utils import vllm_version_is
+
+if vllm_version_is("v0.15.0"):
+    from vllm.attention.layer import Attention  # type: ignore
+else:
+    from vllm.model_executor.layers.attention import Attention
 
 
 class GraphEXQKNormRopeFusionPattern:

vllm_ascend/compilation/passes/qknorm_rope_fusion_pass.py

@@ -18,12 +18,18 @@
 import torch
 import torch._inductor.pattern_matcher as pm
 from torch._inductor.pattern_matcher import PatternMatcherPass, PatternPrettyPrinter
-from vllm.attention.layer import Attention
 from vllm.compilation.vllm_inductor_pass import VllmInductorPass
 from vllm.config import VllmConfig, get_layers_from_vllm_config
 from vllm.config.compilation import Range
 from vllm.logger import logger
 
+from vllm_ascend.utils import vllm_version_is
+
+if vllm_version_is("v0.15.0"):
+    from vllm.attention.layer import Attention  # type: ignore
+else:
+    from vllm.model_executor.layers.attention import Attention
+
 
 class QKNormRopeFusionPattern:
     def __init__(self, vllm_config, head_dim, num_heads, num_kv_heads, eps=1e-6):

vllm_ascend/distributed/kv_transfer/kv_pool/cpu_offload/cpu_offload_connector.py

@@ -10,7 +10,6 @@
 from typing import TYPE_CHECKING, Any, Optional
 
 import torch
-from vllm.attention.layer import Attention, MLAAttention
 from vllm.config import VllmConfig, get_layers_from_vllm_config
 from vllm.distributed.ec_transfer import get_ec_transfer, has_ec_transfer
 from vllm.distributed.kv_transfer.kv_connector.v1.base import KVConnectorBase_V1, KVConnectorMetadata, KVConnectorRole
@@ -27,6 +26,7 @@
     MetadataServerProc,
     MLAConfig,
 )
+from vllm_ascend.utils import vllm_version_is
 
 if TYPE_CHECKING:
     from vllm.forward_context import ForwardContext
@@ -35,6 +35,11 @@
     from vllm.v1.kv_cache_interface import KVCacheConfig
     from vllm.v1.request import Request
 
+if vllm_version_is("v0.15.0"):
+    from vllm.attention.layer import Attention, MLAAttention  # type: ignore
+else:
+    from vllm.model_executor.layers.attention import Attention, MLAAttention
+
 
 @dataclass
 class ReqMeta: