File size: 9,805 Bytes
93adfea |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 |
# Copyright 2024 HuggingFace Inc. and the LlamaFactory team.
#
# This code is inspired by the HuggingFace's PEFT library.
# https://github.com/huggingface/peft/blob/v0.10.0/src/peft/peft_model.py
#
# 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 gc
import os
from typing import TYPE_CHECKING, Any, Dict, Literal, Sequence, Tuple, Union
import torch
import torch.distributed as dist
import transformers.dynamic_module_utils
from transformers import InfNanRemoveLogitsProcessor, LogitsProcessorList
from transformers.dynamic_module_utils import get_relative_imports
from transformers.utils import (
is_torch_bf16_gpu_available,
is_torch_cuda_available,
is_torch_mps_available,
is_torch_npu_available,
is_torch_xpu_available,
)
from transformers.utils.versions import require_version
from . import logging
from .packages import is_transformers_version_greater_than
_is_fp16_available = is_torch_npu_available() or is_torch_cuda_available()
try:
_is_bf16_available = is_torch_bf16_gpu_available() or (is_torch_npu_available() and torch.npu.is_bf16_supported())
except Exception:
_is_bf16_available = False
if TYPE_CHECKING:
from numpy.typing import NDArray
from ..hparams import ModelArguments
logger = logging.get_logger(__name__)
class AverageMeter:
r"""
Computes and stores the average and current value.
"""
def __init__(self):
self.reset()
def reset(self):
self.val = 0
self.avg = 0
self.sum = 0
self.count = 0
def update(self, val, n=1):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / self.count
def check_version(requirement: str, mandatory: bool = False) -> None:
r"""
Optionally checks the package version.
"""
if is_env_enabled("DISABLE_VERSION_CHECK") and not mandatory:
logger.warning_rank0_once("Version checking has been disabled, may lead to unexpected behaviors.")
return
if mandatory:
hint = f"To fix: run `pip install {requirement}`."
else:
hint = f"To fix: run `pip install {requirement}` or set `DISABLE_VERSION_CHECK=1` to skip this check."
require_version(requirement, hint)
def check_dependencies() -> None:
r"""
Checks the version of the required packages.
"""
check_version("transformers>=4.41.2,<=4.49.0,!=4.46.0,!=4.46.1,!=4.46.2,!=4.46.3,!=4.47.0,!=4.47.1,!=4.48.0")
check_version("datasets>=2.16.0,<=3.2.0")
check_version("accelerate>=0.34.0,<=1.2.1")
check_version("peft>=0.11.1,<=0.12.0")
check_version("trl>=0.8.6,<=0.9.6")
if is_transformers_version_greater_than("4.46.0") and not is_transformers_version_greater_than("4.48.1"):
logger.warning_rank0_once("There are known bugs in transformers v4.46.0-v4.48.0, please use other versions.")
def calculate_tps(dataset: Sequence[Dict[str, Any]], metrics: Dict[str, float], stage: Literal["sft", "rm"]) -> float:
r"""
Calculates effective tokens per second.
"""
effective_token_num = 0
for data in dataset:
if stage == "sft":
effective_token_num += len(data["input_ids"])
elif stage == "rm":
effective_token_num += len(data["chosen_input_ids"]) + len(data["rejected_input_ids"])
result = effective_token_num * metrics["epoch"] / metrics["train_runtime"]
return result / dist.get_world_size() if dist.is_initialized() else result
def count_parameters(model: "torch.nn.Module") -> Tuple[int, int]:
r"""
Returns the number of trainable parameters and number of all parameters in the model.
"""
trainable_params, all_param = 0, 0
for param in model.parameters():
num_params = param.numel()
# if using DS Zero 3 and the weights are initialized empty
if num_params == 0 and hasattr(param, "ds_numel"):
num_params = param.ds_numel
# Due to the design of 4bit linear layers from bitsandbytes, multiply the number of parameters by itemsize
if param.__class__.__name__ == "Params4bit":
if hasattr(param, "quant_storage") and hasattr(param.quant_storage, "itemsize"):
num_bytes = param.quant_storage.itemsize
elif hasattr(param, "element_size"): # for older pytorch version
num_bytes = param.element_size()
else:
num_bytes = 1
num_params = num_params * 2 * num_bytes
all_param += num_params
if param.requires_grad:
trainable_params += num_params
return trainable_params, all_param
def get_current_device() -> "torch.device":
r"""
Gets the current available device.
"""
if is_torch_xpu_available():
device = "xpu:{}".format(os.environ.get("LOCAL_RANK", "0"))
elif is_torch_npu_available():
device = "npu:{}".format(os.environ.get("LOCAL_RANK", "0"))
elif is_torch_mps_available():
device = "mps:{}".format(os.environ.get("LOCAL_RANK", "0"))
elif is_torch_cuda_available():
device = "cuda:{}".format(os.environ.get("LOCAL_RANK", "0"))
else:
device = "cpu"
return torch.device(device)
def get_device_count() -> int:
r"""
Gets the number of available GPU or NPU devices.
"""
if is_torch_xpu_available():
return torch.xpu.device_count()
elif is_torch_npu_available():
return torch.npu.device_count()
elif is_torch_cuda_available():
return torch.cuda.device_count()
else:
return 0
def get_logits_processor() -> "LogitsProcessorList":
r"""
Gets logits processor that removes NaN and Inf logits.
"""
logits_processor = LogitsProcessorList()
logits_processor.append(InfNanRemoveLogitsProcessor())
return logits_processor
def get_peak_memory() -> Tuple[int, int]:
r"""
Gets the peak memory usage for the current device (in Bytes).
"""
if is_torch_npu_available():
return torch.npu.max_memory_allocated(), torch.npu.max_memory_reserved()
elif is_torch_cuda_available():
return torch.cuda.max_memory_allocated(), torch.cuda.max_memory_reserved()
else:
return 0, 0
def has_tokenized_data(path: "os.PathLike") -> bool:
r"""
Checks if the path has a tokenized dataset.
"""
return os.path.isdir(path) and len(os.listdir(path)) > 0
def infer_optim_dtype(model_dtype: "torch.dtype") -> "torch.dtype":
r"""
Infers the optimal dtype according to the model_dtype and device compatibility.
"""
if _is_bf16_available and model_dtype == torch.bfloat16:
return torch.bfloat16
elif _is_fp16_available:
return torch.float16
else:
return torch.float32
def is_gpu_or_npu_available() -> bool:
r"""
Checks if the GPU or NPU is available.
"""
return is_torch_npu_available() or is_torch_cuda_available()
def is_env_enabled(env_var: str, default: str = "0") -> bool:
r"""
Checks if the environment variable is enabled.
"""
return os.getenv(env_var, default).lower() in ["true", "y", "1"]
def numpify(inputs: Union["NDArray", "torch.Tensor"]) -> "NDArray":
r"""
Casts a torch tensor or a numpy array to a numpy array.
"""
if isinstance(inputs, torch.Tensor):
inputs = inputs.cpu()
if inputs.dtype == torch.bfloat16: # numpy does not support bfloat16 until 1.21.4
inputs = inputs.to(torch.float32)
inputs = inputs.numpy()
return inputs
def skip_check_imports() -> None:
r"""
Avoids flash attention import error in custom model files.
"""
if not is_env_enabled("FORCE_CHECK_IMPORTS"):
transformers.dynamic_module_utils.check_imports = get_relative_imports
def torch_gc() -> None:
r"""
Collects GPU or NPU memory.
"""
gc.collect()
if is_torch_xpu_available():
torch.xpu.empty_cache()
elif is_torch_npu_available():
torch.npu.empty_cache()
elif is_torch_mps_available():
torch.mps.empty_cache()
elif is_torch_cuda_available():
torch.cuda.empty_cache()
def try_download_model_from_other_hub(model_args: "ModelArguments") -> str:
if (not use_modelscope() and not use_openmind()) or os.path.exists(model_args.model_name_or_path):
return model_args.model_name_or_path
if use_modelscope():
check_version("modelscope>=1.11.0", mandatory=True)
from modelscope import snapshot_download # type: ignore
revision = "master" if model_args.model_revision == "main" else model_args.model_revision
return snapshot_download(
model_args.model_name_or_path,
revision=revision,
cache_dir=model_args.cache_dir,
)
if use_openmind():
check_version("openmind>=0.8.0", mandatory=True)
from openmind.utils.hub import snapshot_download # type: ignore
return snapshot_download(
model_args.model_name_or_path,
revision=model_args.model_revision,
cache_dir=model_args.cache_dir,
)
def use_modelscope() -> bool:
return is_env_enabled("USE_MODELSCOPE_HUB")
def use_openmind() -> bool:
return is_env_enabled("USE_OPENMIND_HUB")
def use_ray() -> bool:
return is_env_enabled("USE_RAY")
|