[mm] Handle per-item embeddings in cache misses (#32498)

Co-authored-by: cctry <cctry@meta.com>

python/sglang/srt/managers/mm_utils.py

@@ -671,11 +671,24 @@ def _batch_encode_per_image_misses(
         if not _can_skip_pre_embed_feature_move(data_embedding_func):
             _move_items_to_device(miss_items, device)
         all_miss_embedding = data_embedding_func(miss_items)
-        all_miss_embedding = all_miss_embedding.reshape(
-            -1, all_miss_embedding.shape[-1]
-        )
 
-        split_embeddings = torch.split(all_miss_embedding, token_counts, dim=0)
+        if isinstance(all_miss_embedding, list):
+            # Per-item embeddings: no split needed, and each cache entry owns
+            # its storage (a torch.split view would pin the whole concatenated
+            # buffer for as long as any single item stays cached). Mirrors
+            # _get_chunked_embedding_by_item.
+            assert len(all_miss_embedding) == len(miss_items), (
+                f"per-item embedding count {len(all_miss_embedding)} != "
+                f"cache-miss item count {len(miss_items)}"
+            )
+            split_embeddings = [
+                emb.reshape(-1, emb.shape[-1]) for emb in all_miss_embedding
+            ]
+        else:
+            all_miss_embedding = all_miss_embedding.reshape(
+                -1, all_miss_embedding.shape[-1]
+            )
+            split_embeddings = torch.split(all_miss_embedding, token_counts, dim=0)
         for h, emb in zip(ordered_hashes, split_embeddings):
             embedding_cache.set(h, EmbeddingResult(embedding=emb))
             # Keep a local ref (no extra GPU memory) so assembly never fails due to LRU eviction.