Assorted fixes and addressing comments

sharktank/sharktank/models/vae/layers.py

@@ -197,13 +197,12 @@ def __init__(
         attn_groups: Optional[int] = None,
     ):
         super().__init__(theta)
-        attentions = []
 
         resnet_groups = resnet_groups if resnet_time_scale_shift == "default" else None
 
         # there is always at least one resnet
         if resnet_time_scale_shift == "spatial":
-            # TODO
+            # TODO Implement ResnetBlockCondNorm2d block for spatial time scale shift
             raise AssertionError(f"ResnetBlockCondNorm2d not yet implemented")
         else:
             resnets = [
@@ -218,24 +217,22 @@ def __init__(
                     dropout=dropout,
                 )
             ]
-        for _ in range(num_layers):
-            if add_attention:
-                attentions.append(
-                    AttentionLayer(
-                        theta("attentions")(0),
-                        heads=1,
-                        dim_head=attention_head_dim,
-                        rescale_output_factor=1.0,
-                        eps=resnet_eps,
-                        norm_num_groups=attn_groups,
-                        residual_connection=True,
-                    )
-                )
-            else:
-                attentions.append(None)
+        # TODO: loop through num_layers properly. Works for sdxl vae specifically but removed for export reasons
+        if add_attention:
+            self.attention = AttentionLayer(
+                theta("attentions")(0),
+                heads=1,
+                dim_head=attention_head_dim,
+                rescale_output_factor=1.0,
+                eps=resnet_eps,
+                norm_num_groups=attn_groups,
+                residual_connection=True,
+            )
+        else:
+            self.attention = None
 
         if resnet_time_scale_shift == "spatial":
-            # TODO
+            # TODO Implement ResnetBlock2D for spatial time scale shift support
             raise AssertionError(
                 f"ResnetBlock2D spatial time scale shift not yet implemented"
             )
@@ -252,16 +249,13 @@ def __init__(
                     dropout=dropout,
                 )
             )
-        self.attentions = nn.ModuleList(attentions)
-        self.resnets = resnets
+        self.resnets = nn.ModuleList(resnets)
 
     def forward(
         self, hidden_states: torch.Tensor, temb: Optional[torch.Tensor] = None
     ) -> torch.Tensor:
         hidden_states = self.resnets[0](hidden_states, temb)
-        for attn, resnet in zip(self.attentions, self.resnets[1:]):
-            if attn is not None:
-                hidden_states = attn(hidden_states)
-            hidden_states = resnet(hidden_states, temb)
-
+        if self.attention is not None:
+            hidden_states = self.attention(hidden_states)
+        hidden_states = self.resnets[1](hidden_states, temb)
         return hidden_states

sharktank/sharktank/models/vae/tools/diffuser_ref.py

@@ -44,6 +44,7 @@ def __init__(
                 )
 
     def decode(self, inp):
+        # The reference vae decode does not do scaling and leaves it for the sdxl pipeline. We integrate it into vae for pipeline performance so using the hardcoded values from the config.json here
         img = 1 / 0.13025 * inp
         x = self.vae.decode(img, return_dict=False)[0]
         return (x / 2 + 0.5).clamp(0, 1)

sharktank/sharktank/models/vae/tools/run_vae.py

@@ -103,8 +103,10 @@ def main(argv):
         inputs = get_random_inputs(dtype=dtype, device=device, bs=args.bs)
 
     if args.export:
+        # TODO move export from a run_vae file
         output = export_vae(mdl, inputs, args.decomp_attn)
         output.save_mlir(args.export)
+        print("exported VAE model. Skipping eager execution")
     else:
         # Save intermediates.
         intermediates_saver = None

sharktank/sharktank/models/vae/tools/sample_data.py

@@ -12,7 +12,6 @@
 
 
 def get_random_inputs(dtype, device, bs: int = 2):
-    torch.random.manual_seed(42)
     height = 1024
     width = 1024
     return torch.rand(bs, 4, width // 8, height // 8, dtype=dtype).to(device)

sharktank/sharktank/ops/signatures.py

@@ -55,13 +55,13 @@
     "sharded_cat",
     "sharded_sum",
     "softmax",
+    "squeeze",
     "to",
     "transfer_to_logical_device",
     "transpose",
     "unflatten",
     "unshard",
     "unsqueeze",
-    "squeeze",
     "view",
     "view_as_complex",
     "view_as_real",

sharktank/sharktank/types/tensors.py

@@ -382,6 +382,11 @@ def size(self, dim: Optional[int] = None) -> tuple[int]:
         if dim is None:
             return tuple(self.shape)
         return self.shape[dim]
+
+    def squeeze(self, dim: Optional[int] = None) -> "AnyTensor":
+        from ..ops import squeeze
+
+        return squeeze(self, dim)
 
     def transpose(self, dim0: int, dim1: int) -> "AnyTensor":
         from ..ops import transpose
@@ -398,11 +403,6 @@ def unsqueeze(self, dim: int) -> "AnyTensor":
 
         return unsqueeze(self, dim)
 
-    def squeeze(self, dim: Optional[int] = None) -> "AnyTensor":
-        from ..ops import squeeze
-
-        return squeeze(self, dim)
-
     def view(self, *args: Union[List[List[int]], List[int]]) -> "AnyTensor":
         from ..ops import view