[Attention] Move Backend enum into registry (#25893)
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
@@ -11,8 +11,8 @@ import pytest
|
||||
import torch
|
||||
|
||||
from tests.quantization.utils import is_quant_method_supported
|
||||
from tests.v1.attention.utils import _Backend
|
||||
from vllm import LLM, SamplingParams
|
||||
from vllm.attention.backends.registry import _Backend
|
||||
from vllm.attention.selector import global_force_attn_backend_context_manager
|
||||
from vllm.config import (CompilationConfig, CompilationLevel, CUDAGraphMode,
|
||||
PassConfig)
|
||||
|
||||
@@ -8,11 +8,11 @@ import torch._dynamo
|
||||
|
||||
from tests.compile.backend import LazyInitPass, TestBackend
|
||||
from tests.models.utils import check_outputs_equal
|
||||
from tests.v1.attention.utils import (BatchSpec, _Backend,
|
||||
create_common_attn_metadata)
|
||||
from tests.v1.attention.utils import BatchSpec, create_common_attn_metadata
|
||||
from vllm import LLM, SamplingParams
|
||||
from vllm._custom_ops import cutlass_scaled_fp4_mm, scaled_fp4_quant
|
||||
from vllm.attention import Attention, AttentionMetadata
|
||||
from vllm.attention.backends.registry import _Backend
|
||||
from vllm.attention.selector import global_force_attn_backend_context_manager
|
||||
from vllm.compilation.fusion import QUANT_OPS
|
||||
from vllm.compilation.fusion_attn import ATTN_OP, AttnFusionPass
|
||||
|
||||
@@ -10,8 +10,9 @@ from unittest.mock import patch
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from vllm.attention.backends.registry import _Backend
|
||||
from vllm.attention.layer import MultiHeadAttention
|
||||
from vllm.attention.selector import _Backend, _cached_get_attn_backend
|
||||
from vllm.attention.selector import _cached_get_attn_backend
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.platforms.cpu import CpuPlatform
|
||||
from vllm.platforms.cuda import CudaPlatform
|
||||
|
||||
@@ -15,10 +15,10 @@ from torch._prims_common import TensorLikeType
|
||||
|
||||
from tests.kernels.quant_utils import native_w8a8_block_matmul
|
||||
from vllm.attention import AttentionBackend, AttentionMetadata, AttentionType
|
||||
from vllm.attention.backends.registry import _Backend
|
||||
from vllm.model_executor.layers.activation import SiluAndMul
|
||||
from vllm.model_executor.layers.fused_moe.utils import (
|
||||
moe_kernel_quantize_input)
|
||||
from vllm.platforms.interface import _Backend
|
||||
from vllm.utils import (STR_BACKEND_ENV_VAR, STR_FLASH_ATTN_VAL,
|
||||
STR_XFORMERS_ATTN_VAL, make_tensor_with_pad)
|
||||
|
||||
|
||||
@@ -8,11 +8,11 @@ import pytest
|
||||
import torch
|
||||
from torch.nn.attention.flex_attention import create_block_mask, flex_attention
|
||||
|
||||
from tests.v1.attention.utils import (BatchSpec, _Backend,
|
||||
create_common_attn_metadata,
|
||||
from tests.v1.attention.utils import (BatchSpec, create_common_attn_metadata,
|
||||
create_standard_kv_cache_spec,
|
||||
create_vllm_config,
|
||||
get_attention_backend)
|
||||
from vllm.attention.backends.registry import _Backend
|
||||
from vllm.config import ModelConfig
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, cdiv, is_torch_equal_or_newer
|
||||
|
||||
@@ -6,12 +6,12 @@ from typing import Optional, Union
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from tests.v1.attention.utils import (BatchSpec, _Backend,
|
||||
create_common_attn_metadata,
|
||||
from tests.v1.attention.utils import (BatchSpec, create_common_attn_metadata,
|
||||
create_standard_kv_cache_spec,
|
||||
create_vllm_config,
|
||||
get_attention_backend)
|
||||
from vllm import _custom_ops as ops
|
||||
from vllm.attention.backends.registry import _Backend
|
||||
from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, cdiv
|
||||
from vllm.v1.attention.backends.utils import CommonAttentionMetadata
|
||||
from vllm.v1.kv_cache_interface import FullAttentionSpec
|
||||
|
||||
@@ -8,10 +8,11 @@ from typing import Optional, Union
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from vllm.attention.backends.registry import _Backend
|
||||
from vllm.config import (CacheConfig, CompilationConfig, DeviceConfig,
|
||||
LoadConfig, ModelConfig, ModelDType, ParallelConfig,
|
||||
SchedulerConfig, VllmConfig)
|
||||
from vllm.platforms import _Backend, current_platform
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.utils import resolve_obj_by_qualname
|
||||
from vllm.v1.attention.backends.utils import CommonAttentionMetadata
|
||||
from vllm.v1.kv_cache_interface import FullAttentionSpec
|
||||
|
||||
@@ -8,10 +8,10 @@ import pytest
|
||||
import torch
|
||||
|
||||
from tests.utils import get_attn_backend_list_based_on_platform
|
||||
from tests.v1.attention.utils import (BatchSpec, _Backend,
|
||||
create_common_attn_metadata,
|
||||
from tests.v1.attention.utils import (BatchSpec, create_common_attn_metadata,
|
||||
create_standard_kv_cache_spec,
|
||||
get_attention_backend)
|
||||
from vllm.attention.backends.registry import _Backend
|
||||
from vllm.config import (CacheConfig, DeviceConfig, ModelConfig,
|
||||
ParallelConfig, SchedulerConfig, SpeculativeConfig,
|
||||
VllmConfig)
|
||||
|
||||
@@ -6,10 +6,10 @@ from unittest import mock
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from tests.v1.attention.utils import (BatchSpec, _Backend,
|
||||
create_common_attn_metadata,
|
||||
from tests.v1.attention.utils import (BatchSpec, create_common_attn_metadata,
|
||||
create_standard_kv_cache_spec,
|
||||
get_attention_backend)
|
||||
from vllm.attention.backends.registry import _Backend
|
||||
from vllm.config import (CacheConfig, DeviceConfig, ModelConfig,
|
||||
ParallelConfig, SchedulerConfig, SpeculativeConfig,
|
||||
VllmConfig)
|
||||
|
||||
@@ -6,9 +6,10 @@ from typing import Optional
|
||||
|
||||
import torch
|
||||
|
||||
from tests.v1.attention.utils import (_Backend, create_standard_kv_cache_spec,
|
||||
from tests.v1.attention.utils import (create_standard_kv_cache_spec,
|
||||
create_vllm_config,
|
||||
get_attention_backend)
|
||||
from vllm.attention.backends.registry import _Backend
|
||||
from vllm.config import ParallelConfig, SpeculativeConfig
|
||||
from vllm.v1.attention.backends.utils import CommonAttentionMetadata
|
||||
|
||||
|
||||
Reference in New Issue
Block a user