Merge pull request #2281 from huggingface/convnext_zepto

convnext zepto, rmsnorm experiments

Author: rwightman

timm/layers/__init__.py

@@ -34,7 +34,7 @@
 from .mixed_conv2d import MixedConv2d
 from .mlp import Mlp, GluMlp, GatedMlp, SwiGLU, SwiGLUPacked, ConvMlp, GlobalResponseNormMlp
 from .non_local_attn import NonLocalAttn, BatNonLocalAttn
-from .norm import GroupNorm, GroupNorm1, LayerNorm, LayerNorm2d, RmsNorm
+from .norm import GroupNorm, GroupNorm1, LayerNorm, LayerNorm2d, RmsNorm, RmsNorm2d
 from .norm_act import BatchNormAct2d, GroupNormAct, GroupNorm1Act, LayerNormAct, LayerNormAct2d,\
     SyncBatchNormAct, convert_sync_batchnorm, FrozenBatchNormAct2d, freeze_batch_norm_2d, unfreeze_batch_norm_2d
 from .padding import get_padding, get_same_padding, pad_same

timm/layers/create_norm.py

@@ -10,7 +10,7 @@
 
 import torch.nn as nn
 
-from .norm import GroupNorm, GroupNorm1, LayerNorm, LayerNorm2d, RmsNorm
+from .norm import GroupNorm, GroupNorm1, LayerNorm, LayerNorm2d, RmsNorm, RmsNorm2d
 from torchvision.ops.misc import FrozenBatchNorm2d
 
 _NORM_MAP = dict(
@@ -22,6 +22,7 @@
     layernorm=LayerNorm,
     layernorm2d=LayerNorm2d,
     rmsnorm=RmsNorm,
+    rmsnorm2d=RmsNorm2d,
     frozenbatchnorm2d=FrozenBatchNorm2d,
 )
 _NORM_TYPES = {m for n, m in _NORM_MAP.items()}

timm/layers/norm.py

@@ -152,3 +152,41 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
         # Since there is no built-in PyTorch impl, always use APEX RmsNorm if is installed.
         x = fast_rms_norm(x, self.normalized_shape, self.weight, self.eps)
         return x
+
+
+class RmsNorm2d(nn.Module):
+    """ RmsNorm w/ fast (apex) norm if available
+    """
+    __constants__ = ['normalized_shape', 'eps', 'elementwise_affine']
+    normalized_shape: Tuple[int, ...]
+    eps: float
+    elementwise_affine: bool
+
+    def __init__(self, channels, eps=1e-6, affine=True, device=None, dtype=None) -> None:
+        factory_kwargs = {'device': device, 'dtype': dtype}
+        super().__init__()
+        normalized_shape = channels
+        if isinstance(normalized_shape, numbers.Integral):
+            # mypy error: incompatible types in assignment
+            normalized_shape = (normalized_shape,)  # type: ignore[assignment]
+        self.normalized_shape = tuple(normalized_shape)  # type: ignore[arg-type]
+        self.eps = eps
+        self.elementwise_affine = affine
+        if self.elementwise_affine:
+            self.weight = nn.Parameter(torch.empty(self.normalized_shape, **factory_kwargs))
+        else:
+            self.register_parameter('weight', None)
+
+        self.reset_parameters()
+
+    def reset_parameters(self) -> None:
+        if self.elementwise_affine:
+            nn.init.ones_(self.weight)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = x.permute(0, 2, 3, 1)
+        # NOTE fast norm fallback needs our rms norm impl, so both paths through here.
+        # Since there is no built-in PyTorch impl, always use APEX RmsNorm if is installed.
+        x = fast_rms_norm(x, self.normalized_shape, self.weight, self.eps)
+        x = x.permute(0, 3, 1, 2)
+        return x

timm/models/convnext.py

@@ -45,7 +45,7 @@
 
 from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD, OPENAI_CLIP_MEAN, OPENAI_CLIP_STD
 from timm.layers import trunc_normal_, AvgPool2dSame, DropPath, Mlp, GlobalResponseNormMlp, \
-    LayerNorm2d, LayerNorm, create_conv2d, get_act_layer, make_divisible, to_ntuple
+    LayerNorm2d, LayerNorm, RmsNorm2d, RmsNorm, create_conv2d, get_act_layer, get_norm_layer, make_divisible, to_ntuple
 from timm.layers import NormMlpClassifierHead, ClassifierHead
 from ._builder import build_model_with_cfg
 from ._features import feature_take_indices
@@ -289,24 +289,27 @@ def __init__(
         super().__init__()
         assert output_stride in (8, 16, 32)
         kernel_sizes = to_ntuple(4)(kernel_sizes)
-        if norm_layer is None:
-            norm_layer = LayerNorm2d
-            norm_layer_cl = norm_layer if conv_mlp else LayerNorm
+        use_rms = isinstance(norm_layer, str) and norm_layer.startswith('rmsnorm')
+        if norm_layer is None or use_rms:
+            norm_layer = RmsNorm2d if use_rms else LayerNorm2d
+            norm_layer_cl = norm_layer if conv_mlp else (RmsNorm if use_rms else LayerNorm)
             if norm_eps is not None:
                 norm_layer = partial(norm_layer, eps=norm_eps)
                 norm_layer_cl = partial(norm_layer_cl, eps=norm_eps)
         else:
             assert conv_mlp,\
                 'If a norm_layer is specified, conv MLP must be used so all norm expect rank-4, channels-first input'
+            norm_layer = get_norm_layer(norm_layer)
             norm_layer_cl = norm_layer
             if norm_eps is not None:
                 norm_layer_cl = partial(norm_layer_cl, eps=norm_eps)
+        act_layer = get_act_layer(act_layer)
 
         self.num_classes = num_classes
         self.drop_rate = drop_rate
         self.feature_info = []
 
-        assert stem_type in ('patch', 'overlap', 'overlap_tiered')
+        assert stem_type in ('patch', 'overlap', 'overlap_tiered', 'overlap_act')
         if stem_type == 'patch':
             # NOTE: this stem is a minimal form of ViT PatchEmbed, as used in SwinTransformer w/ patch_size = 4
             self.stem = nn.Sequential(
@@ -316,11 +319,12 @@ def __init__(
             stem_stride = patch_size
         else:
             mid_chs = make_divisible(dims[0] // 2) if 'tiered' in stem_type else dims[0]
-            self.stem = nn.Sequential(
+            self.stem = nn.Sequential(*filter(None, [
                 nn.Conv2d(in_chans, mid_chs, kernel_size=3, stride=2, padding=1, bias=conv_bias),
+                act_layer() if 'act' in stem_type else None,
                 nn.Conv2d(mid_chs, dims[0], kernel_size=3, stride=2, padding=1, bias=conv_bias),
                 norm_layer(dims[0]),
-            )
+            ]))
             stem_stride = 4
 
         self.stages = nn.Sequential()
@@ -592,6 +596,13 @@ def _cfgv2(url='', **kwargs):
         hf_hub_id='timm/',
         crop_pct=0.95, test_input_size=(3, 288, 288), test_crop_pct=1.0),
 
+    'convnext_zepto_rms.ra4_e3600_r224_in1k': _cfg(
+        hf_hub_id='timm/',
+        mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5)),
+    'convnext_zepto_rms_ols.untrained': _cfg(
+        # hf_hub_id='timm/',
+        mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5),
+        test_input_size=(3, 256, 256), test_crop_pct=0.95),
     'convnext_atto.d2_in1k': _cfg(
         url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/convnext_atto_d2-01bb0f51.pth',
         hf_hub_id='timm/',
@@ -600,6 +611,9 @@ def _cfgv2(url='', **kwargs):
         url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/convnext_atto_ols_a2-78d1c8f3.pth',
         hf_hub_id='timm/',
         test_input_size=(3, 288, 288), test_crop_pct=0.95),
+    'convnext_atto_rms.untrained': _cfg(
+        #hf_hub_id='timm/',
+        test_input_size=(3, 256, 256), test_crop_pct=0.95),
     'convnext_femto.d1_in1k': _cfg(
         url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-rsb-weights/convnext_femto_d1-d71d5b4c.pth',
         hf_hub_id='timm/',
@@ -968,6 +982,23 @@ def _cfgv2(url='', **kwargs):
 })
 
 
+@register_model
+def convnext_zepto_rms(pretrained=False, **kwargs) -> ConvNeXt:
+    # timm femto variant (NOTE: still tweaking depths, will vary between 3-4M param, current is 3.7M
+    model_args = dict(depths=(2, 2, 4, 2), dims=(32, 64, 128, 256), conv_mlp=True, norm_layer='rmsnorm2d')
+    model = _create_convnext('convnext_zepto_rms', pretrained=pretrained, **dict(model_args, **kwargs))
+    return model
+
+
+@register_model
+def convnext_zepto_rms_ols(pretrained=False, **kwargs) -> ConvNeXt:
+    # timm femto variant (NOTE: still tweaking depths, will vary between 3-4M param, current is 3.7M
+    model_args = dict(
+        depths=(2, 2, 4, 2), dims=(32, 64, 128, 256), conv_mlp=True, norm_layer='rmsnorm2d', stem_type='overlap_act')
+    model = _create_convnext('convnext_zepto_rms_ols', pretrained=pretrained, **dict(model_args, **kwargs))
+    return model
+
+
 @register_model
 def convnext_atto(pretrained=False, **kwargs) -> ConvNeXt:
     # timm femto variant (NOTE: still tweaking depths, will vary between 3-4M param, current is 3.7M
@@ -984,6 +1015,14 @@ def convnext_atto_ols(pretrained=False, **kwargs) -> ConvNeXt:
     return model
 
 
+@register_model
+def convnext_atto_rms(pretrained=False, **kwargs) -> ConvNeXt:
+    # timm femto variant (NOTE: still tweaking depths, will vary between 3-4M param, current is 3.7M
+    model_args = dict(depths=(2, 2, 6, 2), dims=(40, 80, 160, 320), conv_mlp=True, norm_layer='rmsnorm2d')
+    model = _create_convnext('convnext_atto_rms', pretrained=pretrained, **dict(model_args, **kwargs))
+    return model
+
+
 @register_model
 def convnext_femto(pretrained=False, **kwargs) -> ConvNeXt:
     # timm femto variant

timm/models/vision_transformer.py

@@ -2019,7 +2019,7 @@ def _cfg(url: str = '', **kwargs) -> Dict[str, Any]:
         hf_hub_id='timm/',
         input_size=(3, 160, 160), crop_pct=0.95),
     'test_vit3.r160_in1k': _cfg(
-        #hf_hub_id='timm/',
+        hf_hub_id='timm/',
         input_size=(3, 160, 160), crop_pct=0.95),
 }
 
@@ -3238,7 +3238,7 @@ def test_vit3(pretrained: bool = False, **kwargs) -> VisionTransformer:
     """ ViT Test
     """
     model_args = dict(
-        patch_size=16, embed_dim=96, depth=10, num_heads=3, mlp_ratio=2,
+        patch_size=16, embed_dim=96, depth=9, num_heads=3, mlp_ratio=2,
         class_token=False, reg_tokens=1, global_pool='map', init_values=1e-5)
     model = _create_vision_transformer('test_vit3', pretrained=pretrained, **dict(model_args, **kwargs))
     return model