[Feature][MM] split the images of one request into multiparts (#11828)

Signed-off-by: Xuchun Shang <xuchun.shang@linux.alibaba.com>
Signed-off-by: Kun(llfl) <i@imux.top>
Signed-off-by: Xuchun Shang <xuchun.shang@gmail.com>
Co-authored-by: Teng Ma <sima.mt@alibaba-inc.com>
Co-authored-by: Kun(llfl) <llfl@linux.alibaba.com>
Co-authored-by: Kun(llfl) <i@imux.top>
Co-authored-by: liusy58 <liusy58@linux.alibaba.com>

python/sglang/srt/managers/schedule_batch.py

@@ -334,8 +334,18 @@ class MultimodalInputs:
 
     @staticmethod
     def from_dict(obj: dict):
+        # Check if MM splitting is enabled
+        if not envs.SGLANG_ENABLE_MM_SPLITTING.get():
+            mm_items = obj["mm_items"]
+        else:
+            from sglang.srt.managers.mm_utils import get_new_expanded_mm_items
+
+            original_mm_items = obj["mm_items"]
+            # Now, `mm_items` contains one item per image.
+            mm_items = get_new_expanded_mm_items(original_mm_items)
+
         ret = MultimodalInputs(
-            mm_items=obj["mm_items"],
+            mm_items=mm_items,
         )
 
         assert isinstance(ret.mm_items, list)