From b2035f456296b5a94b6561e038b542ff7057e7ff Mon Sep 17 00:00:00 2001
From: cklxx <1293822641@qq.com>
Date: Fri, 26 Dec 2025 14:38:14 +0800
Subject: [PATCH] Handle deepscaler answers without markers

---
 slime/rollout/rm_hub/deepscaler.py | 15 ++++++++++-----
 tests/test_math_reward_utils.py    |  6 ++++++
 2 files changed, 16 insertions(+), 5 deletions(-)
 create mode 100644 tests/test_math_reward_utils.py

diff --git a/slime/rollout/rm_hub/deepscaler.py b/slime/rollout/rm_hub/deepscaler.py
index 39d4de383..38db16b3c 100644
--- a/slime/rollout/rm_hub/deepscaler.py
+++ b/slime/rollout/rm_hub/deepscaler.py
@@ -1,12 +1,17 @@
 from .math_utils import extract_answer, grade_answer_mathd, grade_answer_sympy
 
 
-def get_deepscaler_rule_based_reward(response, label):
+def _extract_solution_text(response: str) -> str | None:
     if "</think>" in response:
-        model_solution = response.split("</think>")[-1]
-    elif "###Response" in response:
-        model_solution = response.split("###Response")[1]
-    else:
+        return response.split("</think>")[-1]
+    if "###Response" in response:
+        return response.split("###Response")[1]
+    return response
+
+
+def get_deepscaler_rule_based_reward(response, label):
+    model_solution = _extract_solution_text(response)
+    if not model_solution:
         return 0
 
     model_answer = extract_answer(model_solution)
diff --git a/tests/test_math_reward_utils.py b/tests/test_math_reward_utils.py
new file mode 100644
index 000000000..d85e56a1e
--- /dev/null
+++ b/tests/test_math_reward_utils.py
@@ -0,0 +1,6 @@
+from slime.rollout.rm_hub.deepscaler import get_deepscaler_rule_based_reward
+
+
+def test_deepscaler_reward_fallback_without_markers():
+    response = "Answer: \\boxed{42}"
+    assert get_deepscaler_rule_based_reward(response, "42") == 1