Drop vLLM 0.13.0 support (#6069)

### What this PR does / why we need it?
Drop vLLM 0.13.0 support, upgrade to 0.14.0

- vLLM version: v0.13.0
- vLLM main:
d68209402d

---------

Signed-off-by: hfadzxy <starmoon_zhang@163.com>

vllm_ascend/attention/attention_v1.py

@@ -25,6 +25,18 @@ import vllm.envs as envs_vllm
 from vllm.config import VllmConfig, get_current_vllm_config
 from vllm.forward_context import ForwardContext, get_forward_context
 from vllm.utils.math_utils import cdiv
+from vllm.v1.attention.backend import (  # type: ignore
+    AttentionBackend,
+    AttentionCGSupport,
+    AttentionImpl,
+    AttentionLayer,
+    AttentionMetadataBuilder,
+    AttentionType,
+)
+from vllm.v1.attention.backends.registry import (  # type: ignore
+    AttentionBackendEnum,
+    register_backend,
+)
 from vllm.v1.core.sched.output import SchedulerOutput
 from vllm.v1.kv_cache_interface import AttentionSpec, CrossAttentionSpec
 
@@ -44,34 +56,7 @@ from vllm_ascend.compilation.acl_graph import (
 )
 from vllm_ascend.device.device_op import DeviceOperator
 from vllm_ascend.ops.flashcomm2_oshard_manager import flashcomm2_oshard_manager
-from vllm_ascend.utils import vllm_version_is, weak_ref_tensors
-
-if vllm_version_is("0.13.0"):
-    from vllm.attention.backends.abstract import (  # type: ignore
-        AttentionBackend,
-        AttentionImpl,
-        AttentionLayer,
-        AttentionType,
-    )
-    from vllm.attention.backends.registry import (  # type: ignore
-        AttentionBackendEnum,
-        register_backend,
-    )
-    from vllm.v1.attention.backends.utils import AttentionCGSupport, AttentionMetadataBuilder
-else:
-    from vllm.v1.attention.backend import (  # type: ignore
-        AttentionBackend,
-        AttentionCGSupport,
-        AttentionImpl,
-        AttentionLayer,
-        AttentionMetadataBuilder,
-        AttentionType,
-    )
-    from vllm.v1.attention.backends.registry import (  # type: ignore
-        AttentionBackendEnum,
-        register_backend,
-    )
-
+from vllm_ascend.utils import weak_ref_tensors
 
 # default max value of sliding window size
 SWA_INT_MAX = 2147483647

vllm_ascend/attention/context_parallel/attention_cp.py

@@ -29,6 +29,7 @@ from vllm.distributed import (
     get_pcp_group,
 )
 from vllm.forward_context import ForwardContext, get_forward_context
+from vllm.v1.attention.backend import AttentionCGSupport
 from vllm.v1.kv_cache_interface import AttentionSpec
 
 from vllm_ascend.attention.attention_v1 import (
@@ -49,12 +50,7 @@ from vllm_ascend.attention.utils import (
     split_decodes_and_prefills,
 )
 from vllm_ascend.compilation.acl_graph import get_graph_params, update_graph_params_workspaces
-from vllm_ascend.utils import cp_chunkedprefill_comm_stream, vllm_version_is, weak_ref_tensors
-
-if vllm_version_is("0.13.0"):
-    from vllm.v1.attention.backends.utils import AttentionCGSupport
-else:
-    from vllm.v1.attention.backend import AttentionCGSupport
+from vllm_ascend.utils import cp_chunkedprefill_comm_stream, weak_ref_tensors
 
 
 class AscendAttentionCPMetadataBuilder(AscendAttentionMetadataBuilder):

vllm_ascend/attention/context_parallel/mla_cp.py

@@ -12,6 +12,7 @@ from vllm.distributed import (
 )
 from vllm.forward_context import ForwardContext, get_forward_context
 from vllm.utils.math_utils import cdiv
+from vllm.v1.attention.backend import AttentionCGSupport
 from vllm.v1.kv_cache_interface import AttentionSpec, MLAAttentionSpec
 
 from vllm_ascend.attention.attention_v1 import AscendAttentionState
@@ -37,12 +38,7 @@ from vllm_ascend.attention.context_parallel.common_cp import (
 )
 from vllm_ascend.attention.utils import AscendCommonAttentionMetadata
 from vllm_ascend.compilation.acl_graph import get_draft_graph_params, get_graph_params, update_graph_params_workspaces
-from vllm_ascend.utils import vllm_version_is, weak_ref_tensors
-
-if vllm_version_is("0.13.0"):
-    from vllm.v1.attention.backends.utils import AttentionCGSupport
-else:
-    from vllm.v1.attention.backend import AttentionCGSupport
+from vllm_ascend.utils import weak_ref_tensors
 
 MAX_O_PROJ_PREFETCH_SIZE = 16 * 1024 * 1024
 

vllm_ascend/attention/mla_v1.py

@@ -10,7 +10,10 @@ from vllm.forward_context import ForwardContext, get_forward_context
 from vllm.logger import logger
 from vllm.model_executor.layers.linear import UnquantizedLinearMethod
 from vllm.utils.math_utils import cdiv, round_down
+from vllm.v1.attention.backend import (  # type: ignore
+    AttentionBackend, AttentionCGSupport, MLAAttentionImpl)
 from vllm.v1.attention.backends.mla.common import MLACommonMetadataBuilder
+from vllm.v1.attention.backends.utils import PAD_SLOT_ID  # type: ignore
 from vllm.v1.kv_cache_interface import AttentionSpec, MLAAttentionSpec
 
 from vllm_ascend import envs
@@ -35,23 +38,12 @@ from vllm_ascend.ops.rotary_embedding import get_cos_and_sin_mla
 from vllm_ascend.ops.weight_prefetch import maybe_npu_prefetch
 from vllm_ascend.quantization.w8a8 import AscendW8A8LinearMethod
 from vllm_ascend.utils import (ACL_FORMAT_FRACTAL_ND, maybe_trans_nz,
-                               vllm_version_is, weak_ref_tensors)
+                               weak_ref_tensors)
 from vllm_ascend.worker.npu_input_batch import NPUInputBatch
 
 if TYPE_CHECKING:
     from vllm.v1.core.sched.output import SchedulerOutput
 
-# isort: off
-if vllm_version_is('0.13.0'):
-    from vllm.v1.attention.backends.utils import AttentionCGSupport
-    from vllm.attention.backends.abstract import (  # type: ignore
-        AttentionBackend, MLAAttentionImpl)
-    from vllm.attention.backends.utils import PAD_SLOT_ID  # type: ignore
-else:
-    from vllm.v1.attention.backend import (  # type: ignore
-        AttentionBackend, AttentionCGSupport, MLAAttentionImpl)
-    from vllm.v1.attention.backends.utils import PAD_SLOT_ID  # type: ignore
-# isort: on
 
 MAX_O_PROJ_PREFETCH_SIZE = 16 * 1024 * 1024
 BUILD_METADATA_STEP_PREFILL = 0

vllm_ascend/attention/sfa_v1.py

@@ -12,6 +12,8 @@ from vllm.forward_context import get_forward_context
 from vllm.logger import logger
 from vllm.model_executor.layers.linear import UnquantizedLinearMethod
 from vllm.triton_utils import HAS_TRITON
+from vllm.v1.attention.backend import (  # type: ignore
+    AttentionBackend, AttentionCGSupport, MLAAttentionImpl)
 from vllm.v1.attention.backends.mla.common import MLACommonMetadataBuilder
 from vllm.v1.kv_cache_interface import AttentionSpec
 
@@ -35,20 +37,11 @@ from vllm_ascend.ops.triton.rope import rope_forward_triton
 from vllm_ascend.ops.weight_prefetch import maybe_npu_prefetch
 from vllm_ascend.quantization.w8a8 import AscendW8A8LinearMethod
 from vllm_ascend.utils import (ACL_FORMAT_FRACTAL_ND, _round_up, dispose_layer,
-                               enable_dsa_cp, enable_dsa_cp_with_layer_shard, maybe_trans_nz, vllm_version_is)
+                               enable_dsa_cp, enable_dsa_cp_with_layer_shard, maybe_trans_nz)
 from vllm_ascend.worker.npu_input_batch import NPUInputBatch
 
-# isort: off
 if TYPE_CHECKING:
     from vllm.v1.core.sched.output import SchedulerOutput
-if vllm_version_is('0.13.0'):
-    from vllm.v1.attention.backends.utils import AttentionCGSupport
-    from vllm.attention.backends.abstract import (  # type: ignore
-        AttentionBackend, MLAAttentionImpl)
-else:
-    from vllm.v1.attention.backend import (  # type: ignore
-        AttentionBackend, AttentionCGSupport, MLAAttentionImpl)
-# isort: on
 
 # token count limits within bmm_transpose operator
 BMM_TRANS_MAX_SUPPORTED_TOKENS = 1024