代码拉取完成,页面将自动刷新
同步操作将从 MindSpore/mindformers 强制同步,此操作会覆盖自 Fork 仓库以来所做的任何修改,且无法恢复!!!
确定后同步将在后台操作,完成时将刷新页面,请耐心等待。
model:
arch:
type: SwinForImageClassification
model_config:
type: SwinConfig
num_labels: 1000 # num classes
image_size: 224 # input image size
patch_size: 4 # patch size
num_channels: 3 # channels of input images
embed_dim: 128 # embedding dimension
depths: [2, 2, 18, 2] # number of transformer blocks for each swin layer
num_heads: [4, 8, 16, 32] # number of attention heads for each swin layer
window_size: 7 # window size for swin
mlp_ratio: 4 # ffn_hidden_size = mlp_ratio * embed_dim
qkv_bias: True # has transformer qkv bias or not
layer_norm_eps: 0.00001 # eps of layer_norm
hidden_dropout_prob: 0. # drop rate of MLP
attention_probs_dropout_prob: 0. # drop rate of Attention
drop_path_rate: 0.1 # drop path rate of transformer blocks
use_absolute_embeddings: False # if using absolute position embedding
patch_norm: True # use norm in SwinPatchEmbeddings
hidden_act: gelu # activation of MLP
weight_init: normal # weight initialize type
loss_type: SoftTargetCrossEntropy # loss type
checkpoint_name_or_path: swin_base_p4w7
processor:
type: SwinProcessor
image_processor:
type: SwinImageProcessor
size: 224 # input image size
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。