glm

configs/common/models/swin/swin_base_patch4_window12_384.py

+from libai.config import LazyCall
+from libai.models import SwinTransformer
+
+from .swin_tiny_patch4_window7_224 import cfg
+
+
+cfg.img_size = 384
+cfg.embed_dim = 128
+cfg.depths = [2, 2, 18, 2]
+cfg.num_heads = [4, 8, 16, 32]
+cfg.drop_path_rate = 0.1
+
+model = LazyCall(SwinTransformer)(cfg=cfg)

configs/common/models/swin/swin_base_patch4_window7_224.py

+from libai.config import LazyCall
+from libai.models import SwinTransformer
+
+from .swin_tiny_patch4_window7_224 import cfg
+
+
+cfg.embed_dim = 128
+cfg.depths = [2, 2, 18, 2]
+cfg.num_heads = [4, 8, 16, 32]
+cfg.drop_path_rate = 0.5
+
+model = LazyCall(SwinTransformer)(cfg=cfg)

configs/common/models/swin/swin_large_patch4_window12_384.py

+from libai.config import LazyCall
+from libai.models import SwinTransformer
+
+from .swin_tiny_patch4_window7_224 import cfg
+
+
+cfg.img_size = 384
+cfg.embed_dim = 192
+cfg.depths = [2, 2, 18, 2]
+cfg.num_heads = [6, 12, 24, 48]
+cfg.window_size = 12
+cfg.drop_path_rate = 0.1
+
+model = LazyCall(SwinTransformer)(cfg=cfg)

configs/common/models/swin/swin_large_patch4_window7_224.py

+from libai.config import LazyCall
+from libai.models import SwinTransformer
+
+from .swin_tiny_patch4_window7_224 import cfg
+
+
+cfg.embed_dim = 192
+cfg.depths = [2, 2, 18, 2]
+cfg.num_heads = [6, 12, 24, 48]
+cfg.drop_path_rate = 0.1
+
+model = LazyCall(SwinTransformer)(cfg=cfg)

configs/common/models/swin/swin_small_patch4_window7_224.py

+from libai.config import LazyCall
+from libai.models import SwinTransformer
+from .swin_tiny_patch4_window7_224 import cfg
+
+
+cfg.depths = [2, 2, 18, 2]
+cfg.drop_path_rate = 0.3
+
+model = LazyCall(SwinTransformer)(cfg=cfg)

configs/common/models/swin/swin_tiny_c24_patch4_window8_256.py

+from libai.config import LazyCall
+from libai.models import SwinTransformer
+from .swin_tiny_patch4_window7_224 import cfg
+
+
+cfg.img_size = 256
+cfg.num_heads = [4, 8, 16, 32]
+cfg.window_size = 8
+
+model = LazyCall(SwinTransformer)(cfg=cfg)

configs/common/models/swin/swin_tiny_patch4_window7_224.py

+from omegaconf import DictConfig
+from libai.config import LazyCall
+from libai.models import SwinTransformer
+
+
+cfg = dict(
+    img_size=224,
+    patch_size=4,
+    in_chans=3,
+    num_classes=1000,
+    embed_dim=96,
+    depths=[2, 2, 6, 2],
+    num_heads=[3, 6, 12, 24],
+    window_size=7,
+    mlp_ratio=4.0,
+    qkv_bias=True,
+    qk_scale=None,
+    drop_rate=0.0,
+    drop_path_rate=0.2,
+    ape=False,
+    patch_norm=True,
+    loss_func=None,
+)
+
+cfg = DictConfig(cfg)
+
+model = LazyCall(SwinTransformer)(cfg=cfg)

configs/common/models/swinv2/swinv2_base_patch4_window16_256.py

+from libai.config import LazyCall
+from libai.models import SwinTransformerV2
+from .swinv2_tiny_patch4_window8_256 import cfg
+
+cfg.window_size = 16
+cfg.depths = [2, 2, 18, 2]
+cfg.num_heads = [4, 8, 16, 32]
+cfg.drop_path_rate = 0.5
+
+model = LazyCall(SwinTransformerV2)(cfg=cfg)

configs/common/models/swinv2/swinv2_base_patch4_window8_256.py

+from libai.config import LazyCall
+from libai.models import SwinTransformerV2
+from .swinv2_tiny_patch4_window8_256 import cfg
+
+cfg.depths = [2, 2, 18, 2]
+cfg.num_heads = [4, 8, 16, 32]
+cfg.drop_path_rate = 0.5
+
+model = LazyCall(SwinTransformerV2)(cfg=cfg)

configs/common/models/swinv2/swinv2_small_patch4_window16_256.py

+from libai.config import LazyCall
+from libai.models import SwinTransformerV2
+from .swinv2_tiny_patch4_window8_256 import cfg
+
+cfg.window_size = 16
+cfg.depths = [2, 2, 18, 2]
+cfg.drop_path_rate = 0.3
+
+model = LazyCall(SwinTransformerV2)(cfg=cfg)

configs/common/models/swinv2/swinv2_small_patch4_window8_256.py

+from libai.config import LazyCall
+from libai.models import SwinTransformerV2
+from .swinv2_tiny_patch4_window8_256 import cfg
+
+cfg.depths = [2, 2, 18, 2]
+cfg.drop_path_rate = 0.3
+
+model = LazyCall(SwinTransformerV2)(cfg=cfg)

configs/common/models/swinv2/swinv2_tiny_patch4_window16_256.py

+from libai.config import LazyCall
+from libai.models import SwinTransformerV2
+from .swinv2_tiny_patch4_window8_256 import cfg
+
+cfg.window_size = 16
+
+model = LazyCall(SwinTransformerV2)(cfg=cfg)

configs/common/models/swinv2/swinv2_tiny_patch4_window8_256.py

+from omegaconf import DictConfig
+from libai.config import LazyCall
+from libai.models import SwinTransformerV2
+
+cfg = dict(
+    img_size=256,
+    patch_size=4,
+    in_chans=3,
+    num_classes=1000,
+    embed_dim=96,
+    depths=[2, 2, 6, 2],
+    num_heads=[3, 6, 12, 24],
+    window_size=8,
+    mlp_ratio=4.0,
+    qkv_bias=True,
+    drop_rate=0.0,
+    drop_path_rate=0.2,
+    ape=False,
+    patch_norm=True,
+    pretrained_window_sizes=[0, 0, 0, 0],
+    loss_func=None,
+)
+
+cfg = DictConfig(cfg)
+
+model = LazyCall(SwinTransformerV2)(cfg=cfg)

configs/common/models/t5.py

+from omegaconf import DictConfig
+from libai.config import LazyCall
+from libai.models import T5Model, T5ForPreTraining
+
+
+cfg = dict(
+    vocab_size=30522,
+    hidden_size=768,
+    hidden_layers=6,
+    num_attention_heads=16,
+    intermediate_size=1536,
+    hidden_dropout_prob=0.1,
+    attention_probs_dropout_prob=0.1,
+    max_position_embeddings=512,
+    embedding_dropout_prob=0.1,
+    initializer_range=0.02,
+    layernorm_eps=1e-5,
+    bias_gelu_fusion=True,
+    bias_dropout_fusion=True,
+    scale_mask_softmax_fusion=True,
+    apply_query_key_layer_scaling=True,
+    apply_residual_post_layernorm=False,
+    amp_enabled=False,
+)
+
+cfg = DictConfig(cfg)
+
+t5_model = LazyCall(T5Model)(cfg=cfg)
+
+pretrain_model = LazyCall(T5ForPreTraining)(cfg=cfg)

configs/common/models/vit/vit_base_patch16_224.py

+from libai.config import LazyCall
+from libai.models import VisionTransformer
+
+from .vit_tiny_patch16_224 import cfg
+
+
+cfg.patch_size = 16
+cfg.embed_dim = 768
+cfg.num_heads = 12
+
+model = LazyCall(VisionTransformer)(cfg=cfg)

configs/common/models/vit/vit_base_patch32_224.py

+from libai.config import LazyCall
+from libai.models import VisionTransformer
+
+from .vit_tiny_patch16_224 import cfg
+
+
+cfg.patch_size = 32
+cfg.embed_dim = 768
+cfg.num_heads = 12
+
+model = LazyCall(VisionTransformer)(cfg=cfg)

configs/common/models/vit/vit_giant_patch14_224.py

+from libai.config import LazyCall
+from libai.models import VisionTransformer
+
+from .vit_tiny_patch16_224 import cfg
+
+
+cfg.patch_size = 14
+cfg.embed_dim = 1408
+cfg.mlp_ratio = 48 / 11
+cfg.depth = 40
+cfg.num_heads = 16
+
+model = LazyCall(VisionTransformer)(cfg=cfg)

configs/common/models/vit/vit_gigantic_patch14_224.py

+from libai.config import LazyCall
+from libai.models import VisionTransformer
+
+from .vit_tiny_patch16_224 import cfg
+
+
+cfg.patch_size = 14
+cfg.embed_dim = 1664
+cfg.mlp_ratio = 64 / 13
+cfg.depth = 48
+cfg.num_heads = 16
+
+model = LazyCall(VisionTransformer)(cfg=cfg)

configs/common/models/vit/vit_huge_patch14_224.py

+from libai.config import LazyCall
+from libai.models import VisionTransformer
+
+from .vit_tiny_patch16_224 import cfg
+
+
+cfg.patch_size = 16
+cfg.embed_dim = 1280
+cfg.depth = 32
+cfg.num_heads = 16
+
+model = LazyCall(VisionTransformer)(cfg=cfg)

configs/common/models/vit/vit_large_patch16_224.py

+from libai.config import LazyCall
+from libai.models import VisionTransformer
+
+from .vit_tiny_patch16_224 import cfg
+
+
+cfg.patch_size = 16
+cfg.embed_dim = 1024
+cfg.depth = 24
+cfg.num_heads = 16
+
+model = LazyCall(VisionTransformer)(cfg=cfg)