File size: 2,430 Bytes
28e2d77 d5cfa8f | 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 | from transformers import PretrainedConfig
class AuroraConfig(PretrainedConfig):
model_type = "aurora"
def __init__(
self,
token_len: int = 48,
hidden_size: int = 512,
intermediate_size: int = 1024,
num_enc_layers: int = 12,
num_dec_layers: int = 12,
num_attention_heads: int = 8,
hidden_act: str = "silu",
rope_theta: int = 10000,
dropout_rate: float = 0.2,
max_position_embeddings: int = 10000,
num_sampling_steps: int = 50,
flow_loss_depth: int = 3,
diffusion_batch_mul: int = 4,
threshold_ratio: list[float] = [0.2, 0.3, 0.4, 0.5],
mask_ratio: float = 0.5,
norm_mode: str = 'batch',
num_prototypes: int = 1024,
num_retriever_enc_layers: int = 1,
num_retriever_dec_layers: int = 1,
num_text_cross_layers: int = 1,
num_vision_cross_layers: int = 1,
num_text_connect_layers: int = 1,
num_vision_connect_layers: int = 1,
num_distill: int = 10,
**kwargs,
):
self.token_len = token_len
self.hidden_size = hidden_size
self.intermediate_size = intermediate_size
self.num_enc_layers = num_enc_layers
self.num_dec_layers = num_dec_layers
self.num_attention_heads = num_attention_heads
self.hidden_act = hidden_act
self.rope_theta = rope_theta
self.dropout_rate = dropout_rate
self.max_position_embeddings = max_position_embeddings
self.num_sampling_steps = num_sampling_steps
self.flow_loss_depth = flow_loss_depth
self.diffusion_batch_mul = diffusion_batch_mul
self.threshold_ratio = threshold_ratio
self.mask_ratio = mask_ratio
self.norm_mode = norm_mode
self.num_prototypes = num_prototypes
self.num_retriever_enc_layers = num_retriever_enc_layers
self.num_retriever_dec_layers = num_retriever_dec_layers
self.num_text_cross_layers = num_text_cross_layers
self.num_vision_cross_layers = num_vision_cross_layers
self.num_text_connect_layers = num_text_connect_layers
self.num_vision_connect_layers = num_vision_connect_layers
self.num_distill = num_distill
super().__init__(
**kwargs,
)
|