Add configurable dropout to ResBlocks

Wire a dropout hyperparameter (default 0.1) through the config, model,
training pipeline, and CLI. Persisted in saved model_config so checkpoints
reconstruct the architecture correctly.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

giant/model/network.py

@@ -49,18 +49,20 @@ class ConditionEncoder(nn.Module):
 
 
 class ResBlock(nn.Module):
-    def __init__(self, dim: int, cond_dim: int) -> None:
+    def __init__(self, dim: int, cond_dim: int, dropout: float = 0.1) -> None:
         super().__init__()
         self.norm = nn.LayerNorm(dim)
         self.linear1 = nn.Linear(dim, dim)
         self.cond_proj = nn.Linear(cond_dim, dim, bias=False)
         self.act = nn.SiLU()
+        self.dropout = nn.Dropout(dropout)
         self.linear2 = nn.Linear(dim, dim)
 
     def forward(self, x: torch.Tensor, cond: torch.Tensor) -> torch.Tensor:
         h = self.norm(x)
         h = self.linear1(h) + self.cond_proj(cond)
         h = self.act(h)
+        h = self.dropout(h)
         h = self.linear2(h)
         return x + h
 
@@ -76,6 +78,7 @@ class DenoisingMLP(nn.Module):
         time_dim: int = 64,
         cond_out_dim: int = 128,
         x_dim: int = X_DIM,
+        dropout: float = 0.1,
     ) -> None:
         super().__init__()
         self.time_emb = SinusoidalEmbedding(time_dim)
@@ -88,7 +91,7 @@ class DenoisingMLP(nn.Module):
         merged_cond_dim = time_dim + cond_out_dim
         self.input_proj = nn.Linear(x_dim, hidden_dim)
         self.blocks = nn.ModuleList([
-            ResBlock(hidden_dim, merged_cond_dim) for _ in range(n_blocks)
+            ResBlock(hidden_dim, merged_cond_dim, dropout=dropout) for _ in range(n_blocks)
         ])
         self.out_proj = nn.Linear(hidden_dim, x_dim)