-
Notifications
You must be signed in to change notification settings - Fork 35
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Call to unwrap_fabric before p2e test
- Loading branch information
Showing
5 changed files
with
203 additions
and
8 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,197 @@ | ||
| import math | ||
| from copy import deepcopy | ||
| from typing import Dict, Optional, Sequence, Tuple | ||
|
|
||
| import torch as th | ||
| import torch.nn as nn | ||
| import torch.nn.functional as F | ||
|
|
||
|
|
||
| class FanInInitReLULayer(nn.Module): | ||
| def __init__( | ||
| self, | ||
| inchan: int, | ||
| outchan: int, | ||
| layer_type: str = "conv", | ||
| init_scale: float = 1.0, | ||
| batch_norm: bool = False, | ||
| batch_norm_kwargs: Dict = {}, | ||
| group_norm_groups: Optional[int] = None, | ||
| layer_norm: bool = False, | ||
| use_activation: bool = True, | ||
| **layer_kwargs, | ||
| ): | ||
| super().__init__() | ||
|
|
||
| # Normalization | ||
| self.norm = None | ||
| if batch_norm: | ||
| self.norm = nn.BatchNorm2d(inchan, **batch_norm_kwargs) | ||
| elif group_norm_groups is not None: | ||
| self.norm = nn.GroupNorm(group_norm_groups, inchan) | ||
| elif layer_norm: | ||
| self.norm = nn.LayerNorm(inchan) | ||
|
|
||
| # Layer | ||
| layer = dict(conv=nn.Conv2d, conv3d=nn.Conv3d, linear=nn.Linear)[layer_type] | ||
| self.layer = layer(inchan, outchan, bias=self.norm is None, **layer_kwargs) | ||
| self.use_activation = use_activation | ||
|
|
||
| # Initialization | ||
| self.layer.weight.data *= init_scale / self.layer.weight.norm( | ||
| dim=tuple(range(1, self.layer.weight.data.ndim)), p=2, keepdim=True | ||
| ) | ||
| if self.layer.bias is not None: | ||
| self.layer.bias.data *= 0 | ||
|
|
||
| def forward(self, x: th.Tensor): | ||
| if self.norm is not None: | ||
| x = self.norm(x) | ||
| x = self.layer(x) | ||
| if self.use_activation: | ||
| x = F.relu(x, inplace=True) | ||
| return x | ||
|
|
||
|
|
||
| class CnnBasicBlock(nn.Module): | ||
| def __init__( | ||
| self, | ||
| inchan: int, | ||
| init_scale: float = 1.0, | ||
| init_norm_kwargs: Dict = {}, | ||
| ): | ||
| super().__init__() | ||
|
|
||
| # Layers | ||
| s = math.sqrt(init_scale) | ||
| self.conv0 = FanInInitReLULayer( | ||
| inchan, | ||
| inchan, | ||
| kernel_size=3, | ||
| padding=1, | ||
| init_scale=s, | ||
| **init_norm_kwargs, | ||
| ) | ||
| self.conv1 = FanInInitReLULayer( | ||
| inchan, | ||
| inchan, | ||
| kernel_size=3, | ||
| padding=1, | ||
| init_scale=s, | ||
| **init_norm_kwargs, | ||
| ) | ||
|
|
||
| def forward(self, x: th.Tensor) -> th.Tensor: | ||
| x = x + self.conv1(self.conv0(x)) | ||
| return x | ||
|
|
||
|
|
||
| class CnnDownStack(nn.Module): | ||
| def __init__( | ||
| self, | ||
| inchan: int, | ||
| nblock: int, | ||
| outchan: int, | ||
| init_scale: float = 1.0, | ||
| pool: bool = True, | ||
| post_pool_groups: Optional[int] = None, | ||
| init_norm_kwargs: Dict = {}, | ||
| first_conv_norm: bool = False, | ||
| **kwargs, | ||
| ): | ||
| super().__init__() | ||
|
|
||
| # Params | ||
| self.inchan = inchan | ||
| self.outchan = outchan | ||
| self.pool = pool | ||
|
|
||
| # Layers | ||
| first_conv_init_kwargs = deepcopy(init_norm_kwargs) | ||
| if not first_conv_norm: | ||
| first_conv_init_kwargs["group_norm_groups"] = None | ||
| first_conv_init_kwargs["batch_norm"] = False | ||
| self.firstconv = FanInInitReLULayer( | ||
| inchan, | ||
| outchan, | ||
| kernel_size=3, | ||
| padding=1, | ||
| **first_conv_init_kwargs, | ||
| ) | ||
| self.post_pool_groups = post_pool_groups | ||
| if post_pool_groups is not None: | ||
| self.n = nn.GroupNorm(post_pool_groups, outchan) | ||
| self.blocks = nn.ModuleList( | ||
| [ | ||
| CnnBasicBlock( | ||
| outchan, | ||
| init_scale=init_scale / math.sqrt(nblock), | ||
| init_norm_kwargs=init_norm_kwargs, | ||
| **kwargs, | ||
| ) | ||
| for _ in range(nblock) | ||
| ] | ||
| ) | ||
|
|
||
| def forward(self, x: th.Tensor) -> th.Tensor: | ||
| x = self.firstconv(x) | ||
| if self.pool: | ||
| x = F.max_pool2d(x, kernel_size=3, stride=2, padding=1) | ||
| if self.post_pool_groups is not None: | ||
| x = self.n(x) | ||
| for block in self.blocks: | ||
| x = block(x) | ||
| return x | ||
|
|
||
| def output_shape(self, inshape: Sequence[int]) -> Tuple[int, int, int]: | ||
| c, h, w = inshape | ||
| assert c == self.inchan | ||
| if self.pool: | ||
| return (self.outchan, (h + 1) // 2, (w + 1) // 2) | ||
| else: | ||
| return (self.outchan, h, w) | ||
|
|
||
|
|
||
| class ImpalaCNN(nn.Module): | ||
| def __init__( | ||
| self, | ||
| inshape: Sequence[int], | ||
| chans: Sequence[int], | ||
| outsize: int, | ||
| nblock: int, | ||
| init_norm_kwargs: Dict = {}, | ||
| dense_init_norm_kwargs: Dict = {}, | ||
| first_conv_norm: bool = False, | ||
| **kwargs, | ||
| ): | ||
| super().__init__() | ||
|
|
||
| # Layers | ||
| curshape = inshape | ||
| self.stacks = nn.ModuleList() | ||
| for i, outchan in enumerate(chans): | ||
| stack = CnnDownStack( | ||
| curshape[0], | ||
| nblock=nblock, | ||
| outchan=outchan, | ||
| init_scale=1.0 / math.sqrt(len(chans)), | ||
| init_norm_kwargs=init_norm_kwargs, | ||
| first_conv_norm=first_conv_norm if i == 0 else True, | ||
| **kwargs, | ||
| ) | ||
| self.stacks.append(stack) | ||
| curshape = stack.output_shape(curshape) | ||
| self.dense = FanInInitReLULayer( | ||
| math.prod(curshape), | ||
| outsize, | ||
| layer_type="linear", | ||
| init_scale=1.4, | ||
| **dense_init_norm_kwargs, | ||
| ) | ||
|
|
||
| def forward(self, x: th.Tensor) -> th.Tensor: | ||
| for stack in self.stacks: | ||
| x = stack(x) | ||
| x = x.reshape(x.size(0), -1) | ||
| x = self.dense(x) | ||
| return x |