[Bugfix] Register VLLM_BATCH_INVARIANT in envs.py to fix spurious unknown env var warning (#35007)
Signed-off-by: Ranran <1012869439@qq.com> Signed-off-by: Ranran <hzz5361@psu.edu> Signed-off-by: ran <hzz5361@psu.edu> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
@@ -7,6 +7,7 @@ import torch
|
||||
from torch.nn import Module
|
||||
from torch.utils._python_dispatch import TorchDispatchMode
|
||||
|
||||
import vllm.envs as envs
|
||||
import vllm.model_executor.layers.fused_moe.modular_kernel as mk
|
||||
from vllm import _custom_ops as ops
|
||||
from vllm._aiter_ops import rocm_aiter_ops
|
||||
@@ -17,9 +18,6 @@ from vllm.model_executor.kernels.linear import (
|
||||
)
|
||||
from vllm.model_executor.kernels.linear.scaled_mm import MarlinFP8ScaledMMLinearKernel
|
||||
from vllm.model_executor.layers.attention import Attention
|
||||
from vllm.model_executor.layers.batch_invariant import (
|
||||
vllm_is_batch_invariant,
|
||||
)
|
||||
from vllm.model_executor.layers.fused_moe import (
|
||||
FusedMoE,
|
||||
FusedMoEMethodBase,
|
||||
@@ -441,7 +439,7 @@ class Fp8LinearMethod(LinearMethodBase):
|
||||
) -> torch.Tensor:
|
||||
# if batch invariant mode is enabled, prefer DeepGEMM FP8 path
|
||||
# we will use BF16 dequant when DeepGEMM is not supported.
|
||||
if vllm_is_batch_invariant():
|
||||
if envs.VLLM_BATCH_INVARIANT:
|
||||
if self.block_quant:
|
||||
assert self.weight_block_size is not None
|
||||
return self.w8a8_block_fp8_linear.apply(
|
||||
|
||||
@@ -305,9 +305,7 @@ def _flashinfer_fp8_blockscale_gemm_impl(
|
||||
)
|
||||
return output
|
||||
|
||||
from vllm.model_executor.layers.batch_invariant import vllm_is_batch_invariant
|
||||
|
||||
if vllm_is_batch_invariant():
|
||||
if envs.VLLM_BATCH_INVARIANT:
|
||||
return run_deepgemm(input, weight, weight_scale)
|
||||
|
||||
condition = input.shape[0] < 32
|
||||
|
||||
Reference in New Issue
Block a user