Spaces:
Paused
Paused
File size: 11,069 Bytes
2f5127c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 |
# Copyright 2020-2025 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.
import torch
from huggingface_hub import HfApi
from trl.import_utils import is_mergekit_available
if is_mergekit_available():
from mergekit.config import MergeConfiguration
from mergekit.merge import MergeOptions, run_merge
def upload_model_to_hf(folder_path: str, repo_id: str):
api = HfApi()
# Create the repository if it doesn't exist
repo = api.create_repo(repo_id, repo_type="model")
# Upload the folder to the specified repository
api.upload_folder(
folder_path=folder_path,
repo_id=repo.repo_id,
repo_type=repo.repo_type,
)
class MergeConfig:
r"""
Configuration class for merging two models using `mergekit`.
This class provides a structured way to configure and generate merge configurations for various merge methods,
such as `linear`, `ties`, `dare_ties`, and `slerp`.
Args:
method (`str`, *optional*, defaults to `"linear"`):
Merge method to use. Supported methods include:
- `"linear"`: Linearly combines two models with specified weights.
- `"ties"`: Combines two models using the TIES method with density parameters.
- `"dare_ties"`: A variant of TIES for domain adaptation.
- `"slerp"`: Combines models using spherical linear interpolation.
Note:
For more details about the merge methods and how they are implemented, see the
[MergeKit GitHub repository](https://github.com/arcee-ai/mergekit?tab=readme-ov-file#merge-methods).
Attributes:
method (`str`): The merge method to use.
policy_model_path (`str` or `None`): Path to the policy model.
target_model_path (`str` or `None`): Path to the target model.
policy_model_weight (`float`): Weight for the policy model (for `linear` and `ties` methods).
target_model_weight (`float`): Weight for the target model (for `linear` and `ties` methods).
policy_model_density (`list[float]`): Density parameters for the policy model (for `ties` and `dare_ties`).
target_model_density (`list[float]`): Density parameters for the target model (for `ties` and `dare_ties`).
normalize (`float` or `None`): Normalization factor for the TIES method.
t_values (`float` or `None`): Interpolation factor for the SLERP method.
dtype (`str`): Data type to use for merging, e.g., `"float16"`.
"""
def __init__(self, method: str = "linear"):
if not is_mergekit_available():
raise ImportError("MergeConfig requires the `mergekit` extra. To install, run `pip install mergekit`.")
self.method = method
self.policy_model_path = None
self.target_model_path = None
# Initialize relevant parameters based on the method
if method == "linear":
self.policy_model_weight = 0.5
self.target_model_weight = 0.5
self.dtype = "float16"
elif method == "ties":
self.policy_model_weight = 1.0
self.policy_model_density = [1.0, 0.7, 0.1]
self.target_model_weight = 1.0
self.target_model_density = [1.0]
self.normalize = 1.0
self.dtype = "float16"
elif method == "dare_ties":
self.policy_model_weight = 1.0
self.policy_model_density = [1.0, 0.7, 0.1]
self.target_model_weight = 1.0
self.target_model_density = [1.0]
self.normalize = 1.0
self.dtype = "float16"
elif method == "slerp":
self.t_values = 0.5
self.dtype = "float16"
else:
raise ValueError(f"Unsupported merge method: {method}")
def create_merge_config_linear(self) -> "MergeConfiguration":
"""
Creates a merge configuration for a linear merge of two models with specified weights.
"""
# Create the merge configuration dictionary
merge_config_dict = {
"dtype": self.dtype,
"merge_method": "linear",
"models": [
{"model": self.policy_model_path, "parameters": {"weight": self.policy_model_weight}},
{"model": self.target_model_path, "parameters": {"weight": self.target_model_weight}},
],
}
# Create the MergeConfiguration from the dictionary
merge_config = MergeConfiguration.model_validate(merge_config_dict)
return merge_config
def create_merge_config_ties(self) -> "MergeConfiguration":
"""
Creates a merge configuration for a TIES merge of two models, with specified weights and densities.
"""
# Create the TIES merge configuration dictionary
merge_config_dict = {
"merge_method": "ties",
"slices": None, # Optional slices if needed
"models": [
{
"model": {
"model": {"path": self.target_model_path, "revision": None},
"lora": None,
"override_architecture": None,
},
"parameters": {"density": self.target_model_density, "weight": self.target_model_weight},
},
{
"model": {
"model": {"path": self.policy_model_path, "revision": None},
"lora": None,
"override_architecture": None,
},
"parameters": {"density": self.policy_model_density, "weight": self.policy_model_weight},
},
],
"parameters": {"normalize": self.normalize},
"base_model": {
"model": {"path": self.policy_model_path, "revision": None},
"lora": None,
"override_architecture": None,
},
"dtype": self.dtype,
"tokenizer_source": None,
"tokenizer": None,
"chat_template": None,
"out_dtype": None,
}
# Create the MergeConfiguration from the dictionary
merge_config = MergeConfiguration.model_validate(merge_config_dict)
return merge_config
def create_merge_config_dare_ties(self) -> "MergeConfiguration":
"""
Creates a merge configuration for a DARE TIES merge of two models, with specified weights and densities.
"""
# Create the DARE TIES merge configuration dictionary
merge_config_dict = {
"merge_method": "dare_ties",
"slices": None, # Optional slices if needed
"models": [
{
"model": {
"model": {"path": self.target_model_path, "revision": None},
"lora": None,
"override_architecture": None,
},
"parameters": {"density": self.target_model_density, "weight": self.target_model_weight},
},
{
"model": {
"model": {"path": self.policy_model_path, "revision": None},
"lora": None,
"override_architecture": None,
},
"parameters": {"density": self.policy_model_density, "weight": self.policy_model_weight},
},
],
"parameters": {"normalize": self.normalize},
"base_model": {
"model": {"path": self.policy_model_path, "revision": None},
"lora": None,
"override_architecture": None,
},
"dtype": self.dtype,
"tokenizer_source": None,
"tokenizer": None,
"chat_template": None,
"out_dtype": None,
}
# Create the MergeConfiguration from the dictionary
merge_config = MergeConfiguration.model_validate(merge_config_dict)
return merge_config
def create_merge_config_slerp(self) -> "MergeConfiguration":
"""
Creates a merge configuration for a SLERP merge of a model with a base model.
"""
# Create the SLERP merge configuration dictionary
merge_config_dict = {
"merge_method": "slerp",
"slices": None, # Optional slices if needed
"models": [
{
"model": {
"model": {"path": self.target_model_path, "revision": None},
"lora": None,
"override_architecture": None,
},
"parameters": None, # No specific parameters for SLERP model
}
],
"parameters": {
"t": self.t_values # Set the t values for SLERP
},
"base_model": {
"model": {"path": self.policy_model_path, "revision": None},
"lora": None,
"override_architecture": None,
},
"dtype": self.dtype,
"tokenizer_source": None,
"tokenizer": None,
"chat_template": None,
"out_dtype": None,
}
# Create the MergeConfiguration from the dictionary
merge_config = MergeConfiguration.model_validate(merge_config_dict)
return merge_config
def create(self) -> "MergeConfiguration":
if self.method == "linear":
return self.create_merge_config_linear()
elif self.method == "ties":
return self.create_merge_config_ties()
elif self.method == "dare_ties":
return self.create_merge_config_dare_ties()
elif self.method == "slerp":
return self.create_merge_config_slerp()
def merge_models(config: MergeConfig, out_path: str):
"""
Merge two models using mergekit
Args:
config (`MergeConfig`): The merge configuration.
out_path (`str`): The output path for the merged model.
"""
if not is_mergekit_available():
raise ImportError("merge_models requires the `mergekit` extra. To install, run `pip install mergekit`.")
run_merge(
config,
out_path=out_path,
options=MergeOptions(
cuda=torch.cuda.is_available(),
copy_tokenizer=True,
lazy_unpickle=False,
low_cpu_memory=False,
),
)
|