Spaces:
Running
on
Zero
Running
on
Zero
File size: 3,080 Bytes
d6bc023 |
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 |
from dataclasses import dataclass, field
from typing import Dict, Optional, Sequence, List
import transformers
@dataclass
class ModelArguments:
model_name_or_path: Optional[str] = field(default="facebook/opt-125m")
version: Optional[str] = field(default="v0")
freeze_backbone: bool = field(default=False)
tune_mm_mlp_adapter: bool = field(default=False)
vision_tower: Optional[str] = field(default=None)
mm_vision_select_layer: Optional[int] = field(default=-1) # default to the last layer
pretrain_mm_mlp_adapter: Optional[str] = field(default=None)
mm_projector_type: Optional[str] = field(default='linear')
mm_use_im_start_end: bool = field(default=False)
mm_use_im_patch_token: bool = field(default=True)
mm_patch_merge_type: Optional[str] = field(default='flat')
mm_vision_select_feature: Optional[str] = field(default="patch")
resampler_hidden_size: Optional[int] = field(default=768)
num_queries: Optional[int] = field(default=128)
num_resampler_layers: Optional[int] = field(default=3)
tune_vision_tower: bool = field(default=False)
tune_entire_model: bool = field(default=False)
tune_vit_from_layer: Optional[int] = field(default=100)
tune_embed_tokens: Optional[int] = field(default=False)
@dataclass
class DataArguments:
data_path: str = field(default=None,
metadata={"help": "Path to the training data."})
eval_data_path: str = field(default=None,
metadata={"help": "Path to the evaluation data."})
lazy_preprocess: bool = False
is_multimodal: bool = False
image_folder: Optional[str] = field(default=None)
image_aspect_ratio: str = 'square'
@dataclass
class TrainingArguments(transformers.TrainingArguments):
cache_dir: Optional[str] = field(default=None)
optim: str = field(default="adamw_torch")
remove_unused_columns: bool = field(default=False)
freeze_mm_mlp_adapter: bool = field(default=False)
mpt_attn_impl: Optional[str] = field(default="triton")
model_max_length: int = field(
default=512,
metadata={
"help":
"Maximum sequence length. Sequences will be right padded (and possibly truncated)."
},
)
double_quant: bool = field(
default=True,
metadata={"help": "Compress the quantization statistics through double quantization."}
)
quant_type: str = field(
default="nf4",
metadata={"help": "Quantization data type to use. Should be one of `fp4` or `nf4`."}
)
bits: int = field(
default=16,
metadata={"help": "How many bits to use."}
)
lora_enable: bool = False
lora_r: int = 64
lora_alpha: int = 16
lora_dropout: float = 0.05
lora_weight_path: str = ""
lora_bias: str = "none"
mm_projector_lr: Optional[float] = None
group_by_modality_length: bool = field(default=False)
vision_tower_lr: Optional[float] = None
tune_vit_posemb_only: bool = field(default=False)
tune_vit_only: bool = field(default=False)
|