mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-10-05 00:33:03 +08:00
fbb4e0f8d1
Some checks failed
CE Compile Job / ce_job_pre_check (push) Has been cancelled
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
CE Compile Job / Show Code Archive Output (push) Has been cancelled
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
CE Compile Job / CE_UPLOAD (push) Has been cancelled
* [Feature] Support zai-org/GLM-4.5-Air BF16 model (#3928) * support glm45_air * [Feature] GLM-45-AIR Support Mix Quantization(Dense wfp8afp8 and wint8 triton_moe_backend) (#4051) * check * fix v1 load for mix and wint8 * check --quantizations 'None' * check * support RL rollout * check v1 loader * check glm rollout_model, change wfp8afp8 per_token_cast_to_fp8 to native impl * check rollout moe gate begin layer_id * check rollout e_score_correction_bias * delete infer_to_train_mapping={} * code check
103 lines
3.8 KiB
Python
103 lines
3.8 KiB
Python
"""
|
|
# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# 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.
|
|
"""
|
|
|
|
from typing import Optional
|
|
|
|
from fastdeploy.model_executor.layers.attention.attention import Attention
|
|
from fastdeploy.model_executor.layers.moe.moe import FusedMoE
|
|
|
|
from . import get_quantization_config
|
|
from .quant_base import QuantConfigBase, QuantMethodBase
|
|
|
|
|
|
class MixQuantConfig(QuantConfigBase):
|
|
"""
|
|
Quantization config for layers that has different quantization methods.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
dense_quant_type: str,
|
|
moe_quant_type: str,
|
|
kv_cache_quant_type: str = None,
|
|
image_moe_quant_type: str = None,
|
|
is_channel_wise: bool = False,
|
|
has_zero_point: bool = False,
|
|
is_permuted: bool = True,
|
|
is_checkpoint_bf16: bool = False,
|
|
) -> None:
|
|
super().__init__()
|
|
self.dense_quant_type = dense_quant_type
|
|
self.moe_quant_type = moe_quant_type
|
|
self.kv_cache_quant_type = kv_cache_quant_type
|
|
if image_moe_quant_type is None:
|
|
self.image_moe_quant_type = moe_quant_type
|
|
else:
|
|
self.image_moe_quant_type = image_moe_quant_type
|
|
self.is_channel_wise = is_channel_wise
|
|
self.has_zero_point = has_zero_point
|
|
self.quant_max_bound = 0
|
|
self.quant_min_bound = 0
|
|
self.quant_round_type = 0
|
|
self.is_permuted = is_permuted
|
|
self.is_checkpoint_bf16 = is_checkpoint_bf16
|
|
|
|
def name(self) -> str:
|
|
return "mix_quant"
|
|
|
|
@classmethod
|
|
def from_config(cls, config: dict) -> "MixQuantConfig":
|
|
return cls(
|
|
config["dense_quant_type"],
|
|
config["moe_quant_type"],
|
|
config.get("kv_cache_quant_type", None),
|
|
config.get("image_moe_quant_type", None),
|
|
config.get("is_channel_wise", False),
|
|
config.get("has_zero_point", False),
|
|
config.get("is_permuted", True),
|
|
config.get("is_checkpoint_bf16", False),
|
|
)
|
|
|
|
def get_quant_method(self, layer) -> Optional[QuantMethodBase]:
|
|
if isinstance(layer, FusedMoE):
|
|
if layer.moe_tag == "Image":
|
|
return (
|
|
get_quantization_config(self.image_moe_quant_type)
|
|
.from_config({"is_permuted": self.is_permuted, "is_checkpoint_bf16": self.is_checkpoint_bf16})
|
|
.get_quant_method(layer)
|
|
)
|
|
else:
|
|
return (
|
|
get_quantization_config(self.moe_quant_type)
|
|
.from_config({"is_permuted": self.is_permuted, "is_checkpoint_bf16": self.is_checkpoint_bf16})
|
|
.get_quant_method(layer)
|
|
)
|
|
elif isinstance(layer, Attention):
|
|
if self.kv_cache_quant_type is not None:
|
|
return (
|
|
get_quantization_config("kvcache")
|
|
.from_config(self.kv_cache_quant_type, self.is_channel_wise, self.has_zero_point)
|
|
.get_quant_method(layer)
|
|
)
|
|
else:
|
|
return None
|
|
else:
|
|
return (
|
|
get_quantization_config(self.dense_quant_type)
|
|
.from_config({"is_checkpoint_bf16": self.is_checkpoint_bf16})
|
|
.get_quant_method(layer)
|
|
)
|