File size: 11,139 Bytes
13362e2 |
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 |
# Copyright 2024 HuggingFace Inc., the LlamaFactory team, and the Llamole team.
#
# This code is inspired by the HuggingFace's transformers library.
# https://github.com/huggingface/transformers/blob/v4.40.0/examples/pytorch/language-modeling/run_clm.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.
from dataclasses import asdict, dataclass, field
from typing import TYPE_CHECKING, Any, Dict, Literal, Optional, Union
from typing_extensions import Self
if TYPE_CHECKING:
import torch
@dataclass
class ModelArguments:
r"""
Arguments pertaining to which model/config/tokenizer we are going to fine-tune or infer.
"""
## custom arguments
graph_decoder_path: str = field(
metadata={
"help": "Path to the model weight for graph decoder model "
},
)
graph_encoder_path: str = field(
metadata={
"help": "Path to the model weight for graph encoder model "
},
)
graph_predictor_path: str = field(
metadata={
"help": "Path to the model weight for graph predictor model "
},
)
graph_lm_connector_path: str = field(
metadata={
"help": "Path to the model weight for graph and language model connector "
},
)
model_name_or_path: str = field(
metadata={
"help": "Path to the model weight or identifier from huggingface.co/models or modelscope.cn/models."
},
)
adapter_name_or_path: Optional[str] = field(
default=None,
metadata={
"help": (
"Path to the adapter weight or identifier from huggingface.co/models. "
"Use commas to separate multiple adapters."
)
},
)
adapter_folder: Optional[str] = field(
default=None,
metadata={"help": "The folder containing the adapter weights to load."},
)
cache_dir: Optional[str] = field(
default=None,
metadata={"help": "Where to store the pre-trained models downloaded from huggingface.co or modelscope.cn."},
)
use_fast_tokenizer: bool = field(
default=True,
metadata={"help": "Whether or not to use one of the fast tokenizer (backed by the tokenizers library)."},
)
disable_graph_model_gradient: bool = field(
default=True,
metadata={"help": "Whether or not to disable the training of graph models"},
)
resize_vocab: bool = field(
default=False,
metadata={"help": "Whether or not to resize the tokenizer vocab and the embedding layers."},
)
split_special_tokens: bool = field(
default=False,
metadata={"help": "Whether or not the special tokens should be split during the tokenization process."},
)
new_special_tokens: Optional[str] = field(
default=None,
metadata={"help": "Special tokens to be added into the tokenizer. Use commas to separate multiple tokens."},
)
model_revision: str = field(
default="main",
metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
)
low_cpu_mem_usage: bool = field(
default=True,
metadata={"help": "Whether or not to use memory-efficient model loading."},
)
quantization_method: Literal["bitsandbytes", "hqq", "eetq"] = field(
default="bitsandbytes",
metadata={"help": "Quantization method to use for on-the-fly quantization."},
)
quantization_bit: Optional[int] = field(
default=None,
metadata={"help": "The number of bits to quantize the model using bitsandbytes."},
)
quantization_type: Literal["fp4", "nf4"] = field(
default="nf4",
metadata={"help": "Quantization data type to use in int4 training."},
)
double_quantization: bool = field(
default=True,
metadata={"help": "Whether or not to use double quantization in int4 training."},
)
quantization_device_map: Optional[Literal["auto"]] = field(
default=None,
metadata={"help": "Device map used to infer the 4-bit quantized model, needs bitsandbytes>=0.43.0."},
)
rope_scaling: Optional[Literal["linear", "dynamic"]] = field(
default=None,
metadata={"help": "Which scaling strategy should be adopted for the RoPE embeddings."},
)
flash_attn: Literal["auto", "disabled", "sdpa", "fa2"] = field(
default="auto",
metadata={"help": "Enable FlashAttention for faster training and inference."},
)
shift_attn: bool = field(
default=False,
metadata={"help": "Enable shift short attention (S^2-Attn) proposed by LongLoRA."},
)
mixture_of_depths: Optional[Literal["convert", "load"]] = field(
default=None,
metadata={"help": "Convert the model to mixture-of-depths (MoD) or load the MoD model."},
)
use_unsloth: bool = field(
default=False,
metadata={"help": "Whether or not to use unsloth's optimization for the LoRA training."},
)
visual_inputs: bool = field(
default=False,
metadata={"help": "Whethor or not to use multimodal LLM that accepts visual inputs."},
)
moe_aux_loss_coef: Optional[float] = field(
default=None,
metadata={"help": "Coefficient of the auxiliary router loss in mixture-of-experts model."},
)
disable_gradient_checkpointing: bool = field(
default=False,
metadata={"help": "Whether or not to disable gradient checkpointing."},
)
upcast_layernorm: bool = field(
default=False,
metadata={"help": "Whether or not to upcast the layernorm weights in fp32."},
)
upcast_lmhead_output: bool = field(
default=False,
metadata={"help": "Whether or not to upcast the output of lm_head in fp32."},
)
train_from_scratch: bool = field(
default=False,
metadata={"help": "Whether or not to randomly initialize the model weights."},
)
infer_backend: Literal["huggingface", "vllm"] = field(
default="huggingface",
metadata={"help": "Backend engine used at inference."},
)
vllm_maxlen: int = field(
default=2048,
metadata={"help": "Maximum sequence (prompt + response) length of the vLLM engine."},
)
vllm_gpu_util: float = field(
default=0.9,
metadata={"help": "The fraction of GPU memory in (0,1) to be used for the vLLM engine."},
)
vllm_enforce_eager: bool = field(
default=False,
metadata={"help": "Whether or not to disable CUDA graph in the vLLM engine."},
)
vllm_max_lora_rank: int = field(
default=32,
metadata={"help": "Maximum rank of all LoRAs in the vLLM engine."},
)
offload_folder: str = field(
default="offload",
metadata={"help": "Path to offload model weights."},
)
use_cache: bool = field(
default=True,
metadata={"help": "Whether or not to use KV cache in generation."},
)
infer_dtype: Literal["auto", "float16", "bfloat16", "float32"] = field(
default="auto",
metadata={"help": "Data type for model weights and activations at inference."},
)
hf_hub_token: Optional[str] = field(
default=None,
metadata={"help": "Auth token to log in with Hugging Face Hub."},
)
ms_hub_token: Optional[str] = field(
default=None,
metadata={"help": "Auth token to log in with ModelScope Hub."},
)
export_dir: Optional[str] = field(
default=None,
metadata={"help": "Path to the directory to save the exported model."},
)
export_size: int = field(
default=1,
metadata={"help": "The file shard size (in GB) of the exported model."},
)
export_device: Literal["cpu", "auto"] = field(
default="cpu",
metadata={"help": "The device used in model export, use `auto` to accelerate exporting."},
)
export_quantization_bit: Optional[int] = field(
default=None,
metadata={"help": "The number of bits to quantize the exported model."},
)
export_quantization_dataset: Optional[str] = field(
default=None,
metadata={"help": "Path to the dataset or dataset name to use in quantizing the exported model."},
)
export_quantization_nsamples: int = field(
default=128,
metadata={"help": "The number of samples used for quantization."},
)
export_quantization_maxlen: int = field(
default=1024,
metadata={"help": "The maximum length of the model inputs used for quantization."},
)
export_legacy_format: bool = field(
default=False,
metadata={"help": "Whether or not to save the `.bin` files instead of `.safetensors`."},
)
export_hub_model_id: Optional[str] = field(
default=None,
metadata={"help": "The name of the repository if push the model to the Hugging Face hub."},
)
print_param_status: bool = field(
default=False,
metadata={"help": "For debugging purposes, print the status of the parameters in the model."},
)
def __post_init__(self):
self.compute_dtype: Optional["torch.dtype"] = None
self.device_map: Optional[Union[str, Dict[str, Any]]] = None
self.model_max_length: Optional[int] = None
if self.split_special_tokens and self.use_fast_tokenizer:
raise ValueError("`split_special_tokens` is only supported for slow tokenizers.")
if self.visual_inputs and self.use_unsloth:
raise ValueError("Unsloth does not support MLLM yet. Stay tuned.")
if self.adapter_name_or_path is not None: # support merging multiple lora weights
self.adapter_name_or_path = [path.strip() for path in self.adapter_name_or_path.split(",")]
if self.new_special_tokens is not None: # support multiple special tokens
self.new_special_tokens = [token.strip() for token in self.new_special_tokens.split(",")]
if self.export_quantization_bit is not None and self.export_quantization_dataset is None:
raise ValueError("Quantization dataset is necessary for exporting.")
def to_dict(self) -> Dict[str, Any]:
return asdict(self)
@classmethod
def copyfrom(cls, old_arg: Self, **kwargs) -> Self:
arg_dict = old_arg.to_dict()
arg_dict.update(**kwargs)
new_arg = cls(**arg_dict)
new_arg.compute_dtype = old_arg.compute_dtype
new_arg.device_map = old_arg.device_map
new_arg.model_max_length = old_arg.model_max_length
return new_arg
|