xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-21 22:44:31 +08:00
Code Issues Packages Projects Releases Wiki Activity

update modelopt path

Browse Source 
Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com>
This commit is contained in:
vllmellm 2025-11-01 16:28:18 +00:00
parent dd5a70ec71
commit 52ff537459
1 changed files with 13 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
vllm/model_executor/layers/quantization/modelopt.py
View File

@ -37,6 +37,12 @@ from vllm.model_executor.layers.quantization.base_config import (
QuantizationConfig, QuantizationConfig,
QuantizeMethodBase, QuantizeMethodBase,
) )
from vllm.model_executor.layers.quantization.kernels.scaled_mm import (
init_fp8_linear_kernel,
)
from vllm.model_executor.layers.quantization.kernels.scaled_mm.ScaledMMLinearKernel import ( # noqa: E501
ScaledMMLinearQuantStrategy,
)
from vllm.model_executor.layers.quantization.kv_cache import BaseKVCacheMethod from vllm.model_executor.layers.quantization.kv_cache import BaseKVCacheMethod
from vllm.model_executor.layers.quantization.utils.flashinfer_fp4_moe import ( from vllm.model_executor.layers.quantization.utils.flashinfer_fp4_moe import (
build_flashinfer_fp4_cutlass_moe_prepare_finalize, build_flashinfer_fp4_cutlass_moe_prepare_finalize,
@ -68,7 +74,6 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
swizzle_blockscale, swizzle_blockscale,
) )
from vllm.model_executor.layers.quantization.utils.w8a8_utils import ( from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
Fp8LinearOp,
requantize_with_max_scale, requantize_with_max_scale,
) )
from vllm.model_executor.parameter import ModelWeightParameter, PerTensorScaleParameter from vllm.model_executor.parameter import ModelWeightParameter, PerTensorScaleParameter
@ -254,8 +259,12 @@ class ModelOptFp8LinearMethod(LinearMethodBase):
def __init__(self, quant_config: ModelOptFp8Config) -> None: def __init__(self, quant_config: ModelOptFp8Config) -> None:
self.quant_config = quant_config self.quant_config = quant_config
self.fp8_linear = Fp8LinearOp( self.fp8_linear = init_fp8_linear_kernel(
act_quant_static=True, act_quant_group_shape=GroupShape.PER_TENSOR act_q_static=True,
act_q_group_shape=GroupShape.PER_TENSOR,
weight_quant_strategy=ScaledMMLinearQuantStrategy.TENSOR,
out_dtype=None,
module_name=self.__class__.__name__,
) )
def create_weights( def create_weights(
@ -323,13 +332,7 @@ class ModelOptFp8LinearMethod(LinearMethodBase):
x: torch.Tensor, x: torch.Tensor,
bias: torch.Tensor | None = None, bias: torch.Tensor | None = None,
) -> torch.Tensor: ) -> torch.Tensor:
return self.fp8_linear.apply( return self.fp8_linear.apply_weights(layer, x, bias)
input=x,
weight=layer.weight,
weight_scale=layer.weight_scale,
input_scale=layer.input_scale,
bias=bias,
)
class ModelOptFp8MoEMethod(FusedMoEMethodBase): class ModelOptFp8MoEMethod(FusedMoEMethodBase):