365 lines
15 KiB
Python
365 lines
15 KiB
Python
#
|
|
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
|
# This file is a part of the vllm-ascend project.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
#
|
|
"""Ascend W4A16 quantization helpers and fused MoE method."""
|
|
|
|
from collections.abc import Callable
|
|
from typing import Any
|
|
|
|
import torch
|
|
import torch_npu
|
|
from vllm.config import get_current_vllm_config
|
|
|
|
from vllm_ascend.ascend_config import get_ascend_config
|
|
from vllm_ascend.ascend_forward_context import _EXTRA_CTX
|
|
from vllm_ascend.ops.fused_moe.experts_selector import select_experts
|
|
from vllm_ascend.ops.fused_moe.moe_runtime_args import build_fused_experts_input
|
|
|
|
from .base import AscendMoEScheme, QuantType, get_moe_num_logical_experts
|
|
from .registry import register_scheme
|
|
|
|
|
|
def unpack_from_int32(
|
|
weight: torch.Tensor,
|
|
shape: torch.Size,
|
|
num_bits: int,
|
|
packed_dim: int = 1,
|
|
) -> torch.Tensor:
|
|
"""Unpacks quantized weights from int32 format back to original bits.
|
|
|
|
:param weight: The packed int32 tensor containing quantized weights
|
|
:param shape: Original shape to restore, defaults to None
|
|
:param num_bits: The number of bits used for quantization (<= 8)
|
|
:param packed_dim: Dimension along which weights are packed (0 or 1), defaults to 1
|
|
:return: Unpacked tensor with int8 dtype after applying offset correction
|
|
"""
|
|
assert weight.dtype == torch.int32, f"Expecting `weight.dtype` is torch.int32 but got {weight.dtype}."
|
|
assert num_bits > 0, f"Expecting `num_bits` should be positive but got {num_bits}."
|
|
assert num_bits <= 8, f"Expecting `num_bits` should not be larger than 8 but got {num_bits}."
|
|
assert 32 % num_bits == 0, f"Expecting `num_bits` {num_bits} to divide 32 exactly."
|
|
assert packed_dim in [0, 1], f"Expecting `packed_dim` is 0 or 1 but got {packed_dim}."
|
|
|
|
pack_factor = 32 // num_bits
|
|
mask = (1 << num_bits) - 1
|
|
|
|
if packed_dim == 1:
|
|
unpacked_weight = torch.zeros(
|
|
(weight.shape[0], weight.shape[1] * pack_factor),
|
|
device=weight.device,
|
|
dtype=torch.int32,
|
|
)
|
|
for i in range(pack_factor):
|
|
unpacked_weight[:, i::pack_factor] = (weight >> (num_bits * i)) & mask
|
|
original_row_size = int(shape[1])
|
|
unpacked_weight = unpacked_weight[:, :original_row_size]
|
|
else:
|
|
unpacked_weight = torch.zeros(
|
|
(weight.shape[0] * pack_factor, weight.shape[1]),
|
|
device=weight.device,
|
|
dtype=torch.int32,
|
|
)
|
|
for i in range(pack_factor):
|
|
unpacked_weight[i::pack_factor, :] = (weight >> (num_bits * i)) & mask
|
|
original_row_size = int(shape[0])
|
|
unpacked_weight = unpacked_weight[:original_row_size, :]
|
|
|
|
offset = pow(2, num_bits) // 2
|
|
unpacked_weight = (unpacked_weight - offset).to(torch.int8)
|
|
|
|
return unpacked_weight
|
|
|
|
|
|
def pack_to_int32(weight: torch.Tensor) -> torch.Tensor:
|
|
"""Packs quantized weights into int32 format for storage.
|
|
|
|
:param weight: The 3D tensor to pack, must be int8 or int32 dtype
|
|
:return: Packed tensor with int32 dtype optimized for storage
|
|
"""
|
|
assert weight.dim() == 3, (
|
|
"Expecting `weight.dim()` is 3 ([expert, output_channel, input_channel] or "
|
|
"[expert, input_channel, output_channel]) but got "
|
|
f"{weight.dim()}."
|
|
)
|
|
assert weight.dtype in [torch.int8, torch.int32], (
|
|
f"Expecting `weight.dtype` is torch.int8 or torch.int32 but got {weight.dtype}."
|
|
)
|
|
|
|
if weight.dtype == torch.int32:
|
|
assert weight.shape[-1] % 8 == 0, "the last dim of weight needs to be divided by 8."
|
|
packed_weight = torch_npu.npu_convert_weight_to_int4pack(weight.flatten(0, 1))
|
|
packed_weight = packed_weight.view(weight.shape[0], weight.shape[1], -1)
|
|
else:
|
|
assert weight.shape[-1] % 4 == 0, "the last dim of weight needs to be divided by 4."
|
|
packed_weight = weight.view(torch.int32).contiguous()
|
|
|
|
return packed_weight
|
|
|
|
|
|
@register_scheme("W4A16", "moe")
|
|
class AscendW4A16FusedMoEMethod(AscendMoEScheme):
|
|
"""FusedMoE method for Ascend W4A16.
|
|
|
|
This method supports only weights generated by LLM-Compressor, for
|
|
example ``moonshotai/Kimi-K2-Thinking``.
|
|
|
|
Each original routed MoE expert in the checkpoint stores separate
|
|
LLM-Compressor tensors. The names below use ``L`` for the layer index and
|
|
``E`` for the expert index. For these 4-bit weights, ``pack_factor`` is
|
|
8, so one int32 element stores eight 4-bit weight values.
|
|
|
|
- ``model.layers.L.mlp.experts.E.gate_proj.weight_packed``:
|
|
``torch.int32``,
|
|
``[moe_intermediate_size, hidden_sizes // pack_factor]``.
|
|
- ``model.layers.L.mlp.experts.E.gate_proj.weight_scale``:
|
|
``torch.bfloat16``,
|
|
``[moe_intermediate_size, hidden_sizes // group_size]``.
|
|
- ``model.layers.L.mlp.experts.E.gate_proj.weight_shape``:
|
|
``torch.int32``, ``[2]``.
|
|
- ``model.layers.L.mlp.experts.E.up_proj.weight_packed``:
|
|
``torch.int32``,
|
|
``[moe_intermediate_size, hidden_sizes // pack_factor]``.
|
|
- ``model.layers.L.mlp.experts.E.up_proj.weight_scale``:
|
|
``torch.bfloat16``,
|
|
``[moe_intermediate_size, hidden_sizes // group_size]``.
|
|
- ``model.layers.L.mlp.experts.E.up_proj.weight_shape``:
|
|
``torch.int32``, ``[2]``.
|
|
- ``model.layers.L.mlp.experts.E.down_proj.weight_packed``:
|
|
``torch.int32``,
|
|
``[hidden_sizes, moe_intermediate_size // pack_factor]``.
|
|
- ``model.layers.L.mlp.experts.E.down_proj.weight_scale``:
|
|
``torch.bfloat16``,
|
|
``[hidden_sizes, moe_intermediate_size // group_size]``.
|
|
- ``model.layers.L.mlp.experts.E.down_proj.weight_shape``:
|
|
``torch.int32``, ``[2]``.
|
|
|
|
During loading, the gate and up projections are fused into ``w13`` and the
|
|
down projection is loaded as ``w2``. In
|
|
:meth:`process_weights_after_loading`, weight tensors are unpacked,
|
|
transposed into the data layout required by the Ascend fused MoE operator,
|
|
and repacked into the int32 dtype. The offset tensors are not loaded from the
|
|
checkpoint; they are all-zero tensors constructed because the operator
|
|
requires offset inputs.
|
|
|
|
After :meth:`process_weights_after_loading`, ``apply`` consumes:
|
|
|
|
- ``w13_weight_packed``: ``torch.int32``,
|
|
``[num_experts, hidden_sizes,
|
|
2 * moe_intermediate_size // pack_factor]``.
|
|
- ``w2_weight_packed``: ``torch.int32``,
|
|
``[num_experts, moe_intermediate_size,
|
|
hidden_sizes // pack_factor]``.
|
|
- ``w13_weight_scale``: ``torch.bfloat16``,
|
|
``[num_experts, hidden_sizes // group_size,
|
|
2 * moe_intermediate_size]``.
|
|
- ``w2_weight_scale``: ``torch.bfloat16``,
|
|
``[num_experts, moe_intermediate_size // group_size,
|
|
hidden_sizes]``.
|
|
- ``w13_weight_offset``: ``torch.bfloat16``, all zeros,
|
|
``[num_experts, hidden_sizes // group_size,
|
|
2 * moe_intermediate_size]``.
|
|
- ``w2_weight_offset``: ``torch.bfloat16``, all zeros,
|
|
``[num_experts, moe_intermediate_size // group_size,
|
|
hidden_sizes]``.
|
|
"""
|
|
|
|
quant_type: QuantType = QuantType.W4A16
|
|
|
|
def __init__(self) -> None:
|
|
self.num_bits = 4 # dtype = torch.int4
|
|
self.pack_factor = 8 # pack 8 of torch.int4 tensors to torch.int32
|
|
|
|
vllm_config = get_current_vllm_config()
|
|
self.group_size = vllm_config.quant_config.quant_description.get("group_size", 32)
|
|
self.dynamic_eplb = get_ascend_config().eplb_config.dynamic_eplb
|
|
|
|
def get_weight(
|
|
self,
|
|
num_experts: int,
|
|
intermediate_size_per_partition: int,
|
|
hidden_sizes: int,
|
|
params_dtype: torch.dtype,
|
|
) -> dict[str, Any]:
|
|
assert intermediate_size_per_partition % self.pack_factor == 0, (
|
|
f"Expecting `intermediate_size_per_partition` {intermediate_size_per_partition} "
|
|
f"can be divided by `pack_factor` {self.pack_factor}"
|
|
)
|
|
assert hidden_sizes % self.pack_factor == 0, (
|
|
f"Expecting `hidden_sizes` {hidden_sizes} can be divided by `pack_factor` {self.pack_factor}"
|
|
)
|
|
|
|
param_dict = {}
|
|
|
|
param_dict["w13_weight_packed"] = torch.empty(
|
|
num_experts, 2 * intermediate_size_per_partition, hidden_sizes // self.pack_factor, dtype=torch.int32
|
|
)
|
|
param_dict["w2_weight_packed"] = torch.empty(
|
|
num_experts, hidden_sizes, intermediate_size_per_partition // self.pack_factor, dtype=torch.int32
|
|
)
|
|
|
|
return param_dict
|
|
|
|
def get_dynamic_quant_param(
|
|
self,
|
|
num_experts: int,
|
|
intermediate_size_per_partition: int,
|
|
hidden_sizes: int,
|
|
params_dtype: torch.dtype,
|
|
) -> dict[str, Any]:
|
|
assert intermediate_size_per_partition % self.group_size == 0, (
|
|
f"Expecting `intermediate_size_per_partition` {intermediate_size_per_partition} "
|
|
f"can be divided by `group_size` {self.group_size}"
|
|
)
|
|
assert hidden_sizes % self.group_size == 0, (
|
|
f"Expecting `hidden_sizes` {hidden_sizes} can be divided by `group_size` {self.group_size}"
|
|
)
|
|
|
|
param_dict = {}
|
|
|
|
param_dict["w13_weight_scale"] = torch.empty(
|
|
num_experts, 2 * intermediate_size_per_partition, hidden_sizes // self.group_size, dtype=params_dtype
|
|
)
|
|
param_dict["w2_weight_scale"] = torch.empty(
|
|
num_experts, hidden_sizes, intermediate_size_per_partition // self.group_size, dtype=params_dtype
|
|
)
|
|
param_dict["w13_weight_shape"] = torch.empty(num_experts, 2, dtype=torch.int32)
|
|
param_dict["w2_weight_shape"] = torch.empty(num_experts, 2, dtype=torch.int32)
|
|
param_dict["w13_weight_offset"] = torch.zeros(
|
|
num_experts, 2 * intermediate_size_per_partition, hidden_sizes // self.group_size, dtype=params_dtype
|
|
)
|
|
param_dict["w2_weight_offset"] = torch.zeros(
|
|
num_experts, hidden_sizes, intermediate_size_per_partition // self.group_size, dtype=params_dtype
|
|
)
|
|
|
|
return param_dict
|
|
|
|
def apply(
|
|
self,
|
|
layer: torch.nn.Module,
|
|
x: torch.Tensor,
|
|
router_logits: torch.Tensor,
|
|
top_k: int,
|
|
renormalize: bool,
|
|
use_grouped_topk: bool = False,
|
|
num_experts: int = -1,
|
|
expert_map: torch.Tensor | None = None,
|
|
topk_group: int | None = None,
|
|
num_expert_group: int | None = None,
|
|
custom_routing_function: Callable | None = None,
|
|
scoring_func: str = "softmax",
|
|
routed_scaling_factor: float = 1.0,
|
|
e_score_correction_bias: torch.Tensor | None = None,
|
|
is_prefill: bool = True,
|
|
enable_force_load_balance: bool = True,
|
|
log2phy: torch.Tensor | None = None,
|
|
global_redundant_expert_num: int = 0,
|
|
pertoken_scale: Any | None = None,
|
|
activation: str = "silu",
|
|
apply_router_weight_on_input: bool = False,
|
|
mc2_mask: torch.Tensor | None = None,
|
|
tid2eid: torch.Tensor | None = None,
|
|
) -> torch.Tensor:
|
|
num_shared_experts = getattr(layer, "n_shared_experts", 0)
|
|
if num_shared_experts is None:
|
|
num_shared_experts = 0
|
|
num_logical_experts = get_moe_num_logical_experts(
|
|
layer,
|
|
num_experts,
|
|
global_redundant_expert_num=global_redundant_expert_num,
|
|
num_shared_experts=num_shared_experts,
|
|
)
|
|
assert router_logits.shape[1] == num_logical_experts, (
|
|
"Number of global experts mismatch (excluding redundancy): "
|
|
f"router_logits.shape[1]={router_logits.shape[1]}, num_logical_experts={num_logical_experts}"
|
|
)
|
|
|
|
topk_weights, topk_ids = select_experts(
|
|
hidden_states=x,
|
|
router_logits=router_logits,
|
|
top_k=top_k,
|
|
use_grouped_topk=use_grouped_topk,
|
|
renormalize=renormalize,
|
|
topk_group=topk_group,
|
|
num_expert_group=num_expert_group,
|
|
custom_routing_function=custom_routing_function,
|
|
scoring_func=scoring_func,
|
|
routed_scaling_factor=routed_scaling_factor,
|
|
e_score_correction_bias=e_score_correction_bias,
|
|
num_experts=num_logical_experts,
|
|
tid2eid=tid2eid,
|
|
)
|
|
|
|
topk_ids = topk_ids.to(torch.int32)
|
|
topk_weights = topk_weights.to(x.dtype)
|
|
|
|
moe_comm_method = _EXTRA_CTX.moe_comm_method
|
|
return moe_comm_method.fused_experts(
|
|
fused_experts_input=build_fused_experts_input(
|
|
hidden_states=x,
|
|
topk_weights=topk_weights,
|
|
topk_ids=topk_ids,
|
|
w1=layer.w13_weight_packed,
|
|
w2=layer.w2_weight_packed,
|
|
quant_type=self.quant_type,
|
|
dynamic_eplb=self.dynamic_eplb,
|
|
expert_map=expert_map,
|
|
global_redundant_expert_num=global_redundant_expert_num,
|
|
mc2_mask=mc2_mask,
|
|
apply_router_weight_on_input=apply_router_weight_on_input,
|
|
log2phy=log2phy,
|
|
pertoken_scale=pertoken_scale,
|
|
activation=activation,
|
|
w1_scale=layer.w13_weight_scale,
|
|
w2_scale=layer.w2_weight_scale,
|
|
w1_offset=layer.w13_weight_offset,
|
|
w2_offset=layer.w2_weight_offset,
|
|
swiglu_limit=layer.swiglu_limit,
|
|
)
|
|
)
|
|
|
|
def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
|
|
w13_shape = layer.w13_weight_packed.data.shape
|
|
w2_shape = layer.w2_weight_packed.data.shape
|
|
unpacked_w13_weight = (
|
|
unpack_from_int32(
|
|
layer.w13_weight_packed.data.flatten(0, 1),
|
|
torch.Size([w13_shape[0] * w13_shape[1], w13_shape[2] * self.pack_factor]),
|
|
self.num_bits,
|
|
)
|
|
.view(w13_shape[0], w13_shape[1], -1)
|
|
.transpose(1, 2)
|
|
.contiguous()
|
|
.int()
|
|
)
|
|
unpacked_w2_weight = (
|
|
unpack_from_int32(
|
|
layer.w2_weight_packed.data.flatten(0, 1),
|
|
torch.Size([w2_shape[0] * w2_shape[1], w2_shape[2] * self.pack_factor]),
|
|
self.num_bits,
|
|
)
|
|
.view(w2_shape[0], w2_shape[1], -1)
|
|
.transpose(1, 2)
|
|
.contiguous()
|
|
.int()
|
|
)
|
|
layer.w13_weight_packed.data = pack_to_int32(unpacked_w13_weight)
|
|
layer.w2_weight_packed.data = pack_to_int32(unpacked_w2_weight)
|
|
|
|
layer.w13_weight_scale.data = layer.w13_weight_scale.data.transpose(1, 2).contiguous()
|
|
layer.w2_weight_scale.data = layer.w2_weight_scale.data.transpose(1, 2).contiguous()
|
|
|
|
layer.w13_weight_offset.data = layer.w13_weight_offset.data.transpose(1, 2).contiguous()
|
|
layer.w2_weight_offset.data = layer.w2_weight_offset.data.transpose(1, 2).contiguous()
|