Refactor NVFP4 Linear utils for ModelOpt and CT (#33201)

Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2026-01-30 16:37:42 -08:00
committed by GitHub
parent 2b465570e6
commit 67ebaff528
12 changed files with 462 additions and 483 deletions
@@ -5,12 +5,9 @@ from fnmatch import fnmatch
from typing import TYPE_CHECKING, Any
import torch
from torch.nn import Module
from torch.nn.parameter import Parameter
import vllm.envs as envs
import vllm.model_executor.layers.fused_moe.modular_kernel as mk
from vllm._custom_ops import cutlass_scaled_fp4_mm, scaled_fp4_quant
from vllm.logger import init_logger
from vllm.model_executor.layers.attention import Attention
from vllm.model_executor.layers.fused_moe.config import (
@@ -66,24 +63,19 @@ from vllm.model_executor.layers.quantization.utils.fp8_utils import (
from vllm.model_executor.layers.quantization.utils.marlin_utils import (
get_marlin_input_dtype,
)
from vllm.model_executor.layers.quantization.utils.marlin_utils_fp4 import (
apply_fp4_marlin_linear,
is_fp4_marlin_supported,
prepare_fp4_layer_for_marlin,
from vllm.model_executor.layers.quantization.utils.nvfp4_utils import (
apply_nvfp4_linear,
convert_to_nvfp4_linear_kernel_format,
select_nvfp4_linear_backend,
)
from vllm.model_executor.layers.quantization.utils.quant_utils import (
GroupShape,
cutlass_fp4_supported,
is_layer_skipped,
kFp8DynamicTokenSym,
kFp8StaticTensorSym,
kFp8StaticTokenSym,
kNvfp4Dynamic,
kNvfp4Static,
pad_nvfp4_activation_for_cutlass,
pad_nvfp4_weight_for_cutlass,
slice_nvfp4_output,
swizzle_blockscale,
)
from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
cutlass_block_fp8_supported,
@@ -96,11 +88,6 @@ from vllm.model_executor.parameter import (
PerTensorScaleParameter,
)
from vllm.model_executor.utils import replace_parameter
from vllm.platforms import current_platform
from vllm.utils.flashinfer import (
flashinfer_scaled_fp4_mm,
has_flashinfer,
)
if TYPE_CHECKING:
from vllm.model_executor.models.utils import WeightsMapper
@@ -498,7 +485,7 @@ class ModelOptFp8LinearMethod(LinearMethodBase):
scale[:] = torch.finfo(torch.float32).min
layer.register_parameter("input_scale", scale)
def process_weights_after_loading(self, layer: Module) -> None:
def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
weight = layer.weight
max_w_scale = layer.weight_scale.max()
if not (layer.weight_scale == layer.weight_scale[0]).all():
@@ -580,7 +567,7 @@ class ModelOptFp8PcPtLinearMethod(LinearMethodBase):
weight_scale[:] = torch.finfo(torch.float32).min
layer.register_parameter("weight_scale", weight_scale)
def process_weights_after_loading(self, layer: Module) -> None:
def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
layer.weight = Parameter(layer.weight.t(), requires_grad=False)
layer.weight_scale = Parameter(layer.weight_scale.data, requires_grad=False)
@@ -681,7 +668,7 @@ class ModelOptFp8PbWoLinearMethod(LinearMethodBase):
weight_scale[:] = torch.finfo(torch.float32).min
layer.register_parameter("weight_scale", weight_scale)
def process_weights_after_loading(self, layer: Module) -> None:
def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
# Keep weight in [out, in] layout for W8A8BlockFp8LinearOp.
layer.weight = Parameter(layer.weight.data, requires_grad=False)
@@ -1108,32 +1095,7 @@ class ModelOptNvFp4LinearMethod(LinearMethodBase):
def __init__(self, quant_config: ModelOptNvFp4Config) -> None:
self.quant_config = quant_config
self.marlin_input_dtype = None
self.backend = "none"
if envs.VLLM_NVFP4_GEMM_BACKEND is None:
if current_platform.has_device_capability(100) and has_flashinfer():
self.backend = "flashinfer-cutlass"
elif cutlass_fp4_supported():
self.backend = "cutlass"
elif is_fp4_marlin_supported():
self.backend = "marlin"
elif envs.VLLM_NVFP4_GEMM_BACKEND.startswith("flashinfer-"):
self.backend = envs.VLLM_NVFP4_GEMM_BACKEND
assert has_flashinfer(), f"FlashInfer is required for {self.backend}"
elif envs.VLLM_NVFP4_GEMM_BACKEND == "cutlass":
self.backend = "cutlass"
assert cutlass_fp4_supported(), f"Cutlass is required for {self.backend}"
elif envs.VLLM_NVFP4_GEMM_BACKEND == "marlin":
self.backend = "marlin"
assert is_fp4_marlin_supported(), f"Marlin is required for {self.backend}"
if self.backend == "none":
raise ValueError(
"No valid NVFP4 GEMM backend found. "
"Please check your platform capability."
)
logger.info_once(f"Using {self.backend} for NVFP4 GEMM")
self.backend = select_nvfp4_linear_backend()
def create_weights(
self,
@@ -1181,19 +1143,19 @@ class ModelOptNvFp4LinearMethod(LinearMethodBase):
)
layer.register_parameter("weight", weight)
# Input Weight Scale
input_scale = PerTensorScaleParameter(
# Input Global Scale
input_global_scale = PerTensorScaleParameter(
data=torch.empty(len(output_partition_sizes), dtype=torch.float32),
weight_loader=weight_loader,
)
layer.register_parameter("input_scale", input_scale)
layer.register_parameter("input_scale", input_global_scale)
# Global Weight Scale
weight_scale_2 = PerTensorScaleParameter(
# Weight Global Scale
weight_global_scale = PerTensorScaleParameter(
data=torch.empty(len(output_partition_sizes), dtype=torch.float32),
weight_loader=weight_loader,
)
layer.register_parameter("weight_scale_2", weight_scale_2)
layer.register_parameter("weight_scale_2", weight_global_scale)
# Per Block Weight Scale
weight_scale = ModelWeightParameter(
@@ -1209,65 +1171,25 @@ class ModelOptNvFp4LinearMethod(LinearMethodBase):
layer.register_parameter("weight_scale", weight_scale)
def process_weights_after_loading(self, layer: Module) -> None:
# global scales:
input_scale_2 = layer.input_scale.max().to(torch.float32)
layer.input_scale = Parameter(input_scale_2, requires_grad=False)
weight_scale_2 = layer.weight_scale_2.max().to(torch.float32)
layer.weight_scale_2 = Parameter(weight_scale_2, requires_grad=False)
def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
# Rename ModelOpt checkpoint names to standardized names
input_global_scale = layer.input_scale.max().to(torch.float32)
layer.input_global_scale = Parameter(input_global_scale, requires_grad=False)
del layer.input_scale
weight_global_scale = layer.weight_scale_2.max().to(torch.float32)
layer.weight_global_scale = Parameter(weight_global_scale, requires_grad=False)
del layer.weight_scale_2
# Pre-compute alpha and inverse for runtime quantization
layer.alpha = Parameter(
layer.input_scale * layer.weight_scale_2, requires_grad=False
layer.input_global_scale * layer.weight_global_scale, requires_grad=False
)
layer.input_global_scale_inv = Parameter(
(1.0 / layer.input_global_scale).to(torch.float32), requires_grad=False
)
# Calculate `1 / input_scale` so that we don't need to do so at runtime
layer.input_scale_inv = Parameter(
(1 / layer.input_scale).to(torch.float32), requires_grad=False
)
# Swizzle the weight blockscale.
# contracting dimension is input dimension
# block_size = 16;
assert layer.weight_scale.dtype == torch.float8_e4m3fn, (
"Weight Block scale must be represented as FP8-E4M3"
)
if self.backend == "marlin":
prepare_fp4_layer_for_marlin(layer)
del layer.alpha
del layer.input_scale
elif self.backend == "flashinfer-trtllm":
# FlashInfer TRTLLM FP4 GEMM requires a different weight layout.
# FlashInfer provides nvfp4_quantize to quantize + shuffle the
# layout but we use our own quantization so we have to call
# shuffles ourselves.
from flashinfer import shuffle_matrix_a, shuffle_matrix_sf_a
weight = layer.weight.data
weight_scale = layer.weight_scale.data
epilogue_tile_m = 128
weight = shuffle_matrix_a(weight.view(torch.uint8), epilogue_tile_m)
weight_scale = (
shuffle_matrix_sf_a(weight_scale.view(torch.uint8), epilogue_tile_m)
.reshape(weight_scale.shape)
.view(torch.float8_e4m3fn)
)
layer.weight_scale = Parameter(weight_scale, requires_grad=False)
layer.weight = Parameter(weight, requires_grad=False)
else:
# Swizzle block scales and pad the packed NVFP4 weights for kernel
# alignment (CUTLASS/FlashInfer require K and N divisible by 32).
swizzled_weight_scale = swizzle_blockscale(layer.weight_scale)
layer.weight_scale = Parameter(swizzled_weight_scale, requires_grad=False)
weight, weights_padding_cols = pad_nvfp4_weight_for_cutlass(
layer.weight.data
)
layer.weights_padding_cols = weights_padding_cols
layer.weight = Parameter(weight, requires_grad=False)
# Convert layer to NVFP4 linear kernel format
convert_to_nvfp4_linear_kernel_format(self.backend, layer)
def apply(
self,
@@ -1275,63 +1197,13 @@ class ModelOptNvFp4LinearMethod(LinearMethodBase):
x: torch.Tensor,
bias: torch.Tensor | None = None,
) -> torch.Tensor:
if self.backend == "marlin":
return apply_fp4_marlin_linear(
input=x,
weight=layer.weight,
weight_scale=layer.weight_scale,
weight_scale_2=layer.weight_scale_2,
workspace=layer.workspace,
size_n=layer.output_size_per_partition,
size_k=layer.input_size_per_partition,
bias=bias,
input_dtype=self.marlin_input_dtype,
)
output_dtype = x.dtype
# quantize BF16 or FP16 to (FP4 and interleaved block scale)
x_fp4, x_blockscale = scaled_fp4_quant(
x, layer.input_scale_inv, is_sf_swizzled_layout=True, backend=self.backend
return apply_nvfp4_linear(
backend=self.backend,
layer=layer,
x=x,
bias=bias,
)
# validate dtypes of quantized input, input block scale,
# weight and weight_blockscale
assert x_fp4.dtype == torch.uint8
assert layer.weight.dtype == torch.uint8
assert x_blockscale.dtype == torch.float8_e4m3fn
assert layer.weight_scale.dtype == torch.float8_e4m3fn
assert layer.alpha.dtype == torch.float32
# Pad activations to match weight K-dimension padding
weights_padding_cols = getattr(layer, "weights_padding_cols", 0)
output_size = layer.output_size_per_partition
output_shape = [x.shape[0], output_size]
x_fp4 = pad_nvfp4_activation_for_cutlass(x_fp4, weights_padding_cols)
mm_args = (
x_fp4,
layer.weight,
x_blockscale,
layer.weight_scale,
layer.alpha,
output_dtype,
)
if self.backend.startswith("flashinfer-"):
backend_name = self.backend[len("flashinfer-") :]
out = flashinfer_scaled_fp4_mm(*mm_args, backend=backend_name)
else:
assert self.backend == "cutlass"
out = cutlass_scaled_fp4_mm(*mm_args)
# Slice output to remove N-dimension padding
out = slice_nvfp4_output(out, output_size)
if bias is not None:
out = out + bias
return out.view(*output_shape)
class ModelOptNvFp4FusedMoE(FusedMoEMethodBase):
"""