Update moondream.py

fix: retry fix for batched point generation

moondream.py

@@ -906,8 +906,7 @@ class MoondreamModel(nn.Module):
         pos_vec: torch.Tensor,       # (B,) next write pos per row after prefill
         include_size: bool = True,
         max_objects: int = 50,
-        lora=None
-    ):
+        lora=None):
         """
         Batched decode loop for multi-label detection.
     
@@ -948,7 +947,7 @@ class MoondreamModel(nn.Module):
                 mask[alive, :, pos] = 1
                 logits, hidden = self._decode_one_tok(
                     x_emb,
-                    mask.unsqueeze(2),                              # (B,1,1,max_ctx) 
+                    mask.unsqueeze(2),                              # (B,1,1,max_ctx) ✅
                     torch.tensor([pos], device=device, dtype=torch.long),
                     lora,
                 )
@@ -966,7 +965,7 @@ class MoondreamModel(nn.Module):
                 mask[alive, :, pos] = 1
                 logits, hidden = self._decode_one_tok(
                     y_emb,
-                    mask.unsqueeze(2),                              # (B,1,1,max_ctx) 
+                    mask.unsqueeze(2),                              # (B,1,1,max_ctx) ✅
                     torch.tensor([pos], device=device, dtype=torch.long),
                     lora,
                 )
@@ -1028,6 +1027,7 @@ class MoondreamModel(nn.Module):
 
 
 
+
     def detect_multi(self, image, objects, settings=None):
         """
         Parallel multi-label detection.