Deprecations for v0.23 and v0.24 (#44992)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
@@ -6,7 +6,6 @@ from typing import TYPE_CHECKING
|
||||
|
||||
import torch
|
||||
|
||||
import vllm.envs as envs
|
||||
from vllm.logger import init_logger
|
||||
from vllm.model_executor.layers.quantization.utils.flashinfer_utils import (
|
||||
align_fp4_moe_weights_for_fi,
|
||||
@@ -15,10 +14,6 @@ from vllm.model_executor.layers.quantization.utils.flashinfer_utils import (
|
||||
from vllm.model_executor.layers.quantization.utils.nvfp4_utils import (
|
||||
swizzle_blockscale,
|
||||
)
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.utils.flashinfer import (
|
||||
has_flashinfer_cutlass_fused_moe,
|
||||
)
|
||||
|
||||
if TYPE_CHECKING:
|
||||
from vllm.model_executor.layers.fused_moe import RoutedExperts
|
||||
@@ -34,16 +29,6 @@ __all__ = [
|
||||
]
|
||||
|
||||
|
||||
def is_flashinfer_fp4_cutlass_moe_available() -> bool:
|
||||
"""Return `True` when FlashInfer CUTLASS NV-FP4 kernels can be used."""
|
||||
return (
|
||||
envs.VLLM_USE_FLASHINFER_MOE_FP4
|
||||
and has_flashinfer_cutlass_fused_moe()
|
||||
and current_platform.is_cuda()
|
||||
and current_platform.has_device_capability(100)
|
||||
)
|
||||
|
||||
|
||||
def reorder_w1w3_to_w3w1(
|
||||
weight: torch.Tensor, scale: torch.Tensor, dim: int = -2
|
||||
) -> tuple[torch.Tensor, torch.Tensor]:
|
||||
|
||||
@@ -5,10 +5,8 @@ from typing import TYPE_CHECKING
|
||||
|
||||
import torch
|
||||
|
||||
from vllm import envs
|
||||
from vllm.logger import init_logger
|
||||
from vllm.model_executor.layers.fused_moe.activation import MoEActivation
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.utils.math_utils import round_up
|
||||
|
||||
if TYPE_CHECKING:
|
||||
@@ -95,34 +93,6 @@ def rotate_weights_for_fi_trtllm_fp8_per_tensor_moe(
|
||||
)
|
||||
|
||||
|
||||
def get_flashinfer_moe_backend() -> FlashinferMoeBackend:
|
||||
backend_map = {
|
||||
"throughput": FlashinferMoeBackend.CUTLASS,
|
||||
"latency": FlashinferMoeBackend.TENSORRT_LLM,
|
||||
"masked_gemm": FlashinferMoeBackend.CUTEDSL,
|
||||
}
|
||||
|
||||
flashinfer_moe_backend = envs.VLLM_FLASHINFER_MOE_BACKEND
|
||||
if flashinfer_moe_backend in backend_map:
|
||||
if (
|
||||
flashinfer_moe_backend == "latency"
|
||||
and not current_platform.is_device_capability_family(100)
|
||||
):
|
||||
logger.info_once(
|
||||
"Flashinfer TRTLLM MOE backend is only supported on "
|
||||
"SM100 and later, using CUTLASS backend instead",
|
||||
)
|
||||
return FlashinferMoeBackend.CUTLASS
|
||||
return backend_map[flashinfer_moe_backend]
|
||||
elif current_platform.is_device_capability(90):
|
||||
return FlashinferMoeBackend.CUTLASS
|
||||
|
||||
raise ValueError(
|
||||
f"Unknown flashinfer moe backend: {flashinfer_moe_backend!r}. "
|
||||
f"Expected one of {list(backend_map.keys())}."
|
||||
)
|
||||
|
||||
|
||||
def is_flashinfer_supporting_global_sf(backend: FlashinferMoeBackend | None) -> bool:
|
||||
# TODO(shuw@nvidia): Update when new backends are added.
|
||||
backends_supporting_global_sf = (
|
||||
|
||||
Reference in New Issue
Block a user