From a2dcf557484a0c86f116b0dfcb83cd0215849811 Mon Sep 17 00:00:00 2001
From: VanyMo <91475099+VanyMo@users.noreply.github.com>
Date: Sun, 16 Aug 2026 22:46:25 +0800
Subject: [PATCH] fix: preserve text-only multimodal messages

---
 slime/utils/data.py              |  3 +++
 tests/test_filter_long_prompt.py | 30 ++++++++++++++++++++++++++++--
 2 files changed, 31 insertions(+), 2 deletions(-)

diff --git a/slime/utils/data.py b/slime/utils/data.py
index adac4b95a..d35c7568e 100644
--- a/slime/utils/data.py
+++ b/slime/utils/data.py
@@ -160,6 +160,9 @@ def _build_messages(data: dict, prompt_key: str, as_conversation: bool, multimod
                 if multimodal_data is not None:
                     multimodals[mt.placeholder] = (mt, list(multimodal_data))
 
+        if not multimodals:
+            return prompt
+
         pattern = "(" + "|".join(re.escape(p) for p in multimodals.keys()) + ")"
 
         for message in prompt:
diff --git a/tests/test_filter_long_prompt.py b/tests/test_filter_long_prompt.py
index 803a443e0..6a4841a6e 100644
--- a/tests/test_filter_long_prompt.py
+++ b/tests/test_filter_long_prompt.py
@@ -18,10 +18,9 @@
 
 import pytest
 
-from slime.utils.data import filter_long_prompt
+from slime.utils.data import _build_messages, filter_long_prompt
 from slime.utils.types import Sample
 
-
 NUM_GPUS = 0
 
 
@@ -110,3 +109,30 @@ def test_no_processor_path_still_preserves_order():
     kept = filter_long_prompt(samples, _Tokenizer(), None, max_length=100)
 
     assert [s.prompt for s in kept] == ["p0:5", "p2:5"]
+
+
+@pytest.mark.unit
+def test_text_only_row_keeps_string_content_with_multimodal_config():
+    data = {"prompt": "hello"}
+
+    prompt = _build_messages(data, "prompt", as_conversation=True, multimodal_keys={"image": "images"})
+
+    assert prompt == [{"role": "user", "content": "hello"}]
+
+
+@pytest.mark.unit
+def test_multimodal_placeholder_expansion_is_unchanged():
+    data = {"prompt": "look <image> here", "images": ["image.png"]}
+
+    prompt = _build_messages(data, "prompt", as_conversation=True, multimodal_keys={"image": "images"})
+
+    assert prompt == [
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "look "},
+                {"type": "image", "image": "image.png"},
+                {"type": "text", "text": " here"},
+            ],
+        }
+    ]