Fix hardcoded input dim in DiffusionModelEncoder (#8514)

Fixes #8496 

### Description

A few sentences describing the changes proposed in this pull request.

### Types of changes
<!--- Put an `x` in all the boxes that apply, and remove the not
applicable items -->
- [x] Non-breaking change (fix or new feature that would not break
existing functionality).
- [ ] Breaking change (fix or new feature that would cause existing
functionality to change).
- [ ] New tests added to cover the changes.
- [ ] Integration tests passed locally by running `./runtests.sh -f -u
--net --coverage`.
- [ ] Quick tests passed locally by running `./runtests.sh --quick
--unittests --disttests`.
- [ ] In-line docstrings updated.
- [ ] Documentation updated, tested `make html` command in the `docs/`
folder.

Signed-off-by: IamTingTing <[email protected]>
Co-authored-by: YunLiu <[email protected]>
Co-authored-by: Eric Kerfoot <[email protected]>

Author: 3 people

monai/networks/nets/diffusion_model_unet.py

@@ -33,6 +33,7 @@
 
 import math
 from collections.abc import Sequence
+from typing import Optional
 
 import torch
 from torch import nn
@@ -2006,7 +2007,7 @@ def __init__(
 
             self.down_blocks.append(down_block)
 
-        self.out = nn.Sequential(nn.Linear(4096, 512), nn.ReLU(), nn.Dropout(0.1), nn.Linear(512, self.out_channels))
+        self.out: Optional[nn.Module] = None
 
     def forward(
         self,
@@ -2049,6 +2050,12 @@ def forward(
             h, _ = downsample_block(hidden_states=h, temb=emb, context=context)
 
         h = h.reshape(h.shape[0], -1)
+
+        # 5. out
+        if self.out is None:
+            self.out = nn.Sequential(
+                nn.Linear(h.shape[1], 512), nn.ReLU(), nn.Dropout(0.1), nn.Linear(512, self.out_channels)
+            )
         output: torch.Tensor = self.out(h)
 
         return output