Upload modeling.py with huggingface_hub

modeling.py

@@ -372,18 +372,15 @@ class NewEmbeddings(nn.Module):
         embeddings = inputs_embeds
 
         # Set and unpad position_ids
-        # Always regenerate internally for robustness — avoids corrupt position_ids
-        # from external callers (e.g. sentence-transformers on some environments)
-        if seq_length > self.position_ids.size(0):
-            self.register_buffer(
-                "position_ids", torch.arange(seq_length, device=embeddings.device), persistent=False
-            )
+        # Create fresh position_ids each call for robustness — the persistent buffer
+        # can become corrupted on certain environments (Python 3.13 / HF Spaces)
+        _pos = torch.arange(seq_length, dtype=torch.long, device=embeddings.device)
         if unpad_inputs:
             # [1, cumsum_seq_len]
-            position_ids = torch.cat([self.position_ids[:l] for l in length]).unsqueeze(0)
+            position_ids = torch.cat([_pos[:l] for l in length]).unsqueeze(0)
         else:
             # [bs, seq_len]
-            position_ids = self.position_ids[:seq_length].expand(batch_size, -1)
+            position_ids = _pos.unsqueeze(0).expand(batch_size, -1)
 
         # Compute rotary embedding
         if self.position_embedding_type == 'rope':