Spaces:
Runtime error
Runtime error
File size: 4,462 Bytes
b37c16f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 |
# Copyright 2024 The HuggingFace Team. 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.
"AQLM (Additive Quantization of Language Model) integration file"
from ..utils import is_accelerate_available, is_aqlm_available, is_torch_available
if is_torch_available():
import torch.nn as nn
def replace_with_aqlm_linear(
model,
quantization_config=None,
linear_weights_not_to_quantize=None,
current_key_name=None,
has_been_replaced=False,
):
"""
Public method that recursively replaces the Linear layers of the given model with AQLM quantized layers.
`accelerate` is needed to use this method. Returns the converted model and a boolean that indicates if the
conversion has been successfull or not.
Args:
model (`torch.nn.Module`):
The model to convert, can be any `torch.nn.Module` instance.
quantization_config (`AqlmConfig`):
The quantization config object that contains the quantization parameters.
linear_weights_not_to_quantize (`list[str]`, *optional*):
A list of nn.Linear weights to not convert. If a parameter path is in the list (e.g. `lm_head.weight`), the corresponding module will not be
converted.
current_key_name (`list`, *optional*):
A list that contains the current key name. This is used for recursion and should not be passed by the user.
has_been_replaced (`bool`, *optional*):
A boolean that indicates if the conversion has been successful or not. This is used for recursion and
should not be passed by the user.
"""
if not is_aqlm_available():
raise ValueError("AQLM is not available. Please install it with `pip install aqlm[cpu,gpu]`")
if not is_accelerate_available():
raise ValueError("AQLM requires Accelerate to be installed: `pip install accelerate`")
if linear_weights_not_to_quantize is None:
linear_weights_not_to_quantize = []
from accelerate import init_empty_weights
from aqlm import QuantizedLinear
for name, module in model.named_children():
if current_key_name is None:
current_key_name = []
current_key_name.append(name)
if isinstance(module, nn.Linear):
# Check if the current key is not in the `linear_weights_not_to_quantize`
if ".".join(current_key_name) + ".weight" not in linear_weights_not_to_quantize:
with init_empty_weights():
in_features = module.in_features
out_features = module.out_features
model._modules[name] = QuantizedLinear(
in_features,
out_features,
bias=module.bias is not None,
in_group_size=quantization_config.in_group_size,
out_group_size=quantization_config.out_group_size,
num_codebooks=quantization_config.num_codebooks,
nbits_per_codebook=quantization_config.nbits_per_codebook,
)
has_been_replaced = True
# Store the module class in case we need to transpose the weight later
model._modules[name].source_cls = type(module)
# Force requires grad to False to avoid unexpected errors
model._modules[name].requires_grad_(False)
if len(list(module.children())) > 0:
_, has_been_replaced = replace_with_aqlm_linear(
module,
quantization_config=quantization_config,
linear_weights_not_to_quantize=linear_weights_not_to_quantize,
current_key_name=current_key_name,
has_been_replaced=has_been_replaced,
)
# Remove the last key for recursion
current_key_name.pop(-1)
return model, has_been_replaced
|