Merge remote-tracking branch 'origin/dev' into kernels_unittest_fix_ljl

Author: meichangsu1

cookbook/client/tinker/short_math_grpo.py

@@ -74,7 +74,7 @@ def __call__(self, sample):
             return Trajectory(messages=[], user_data=[])
 
         def get_boxed_answer(text):
-            match = re.search(r'\\boxed\{([^}]+)\}', text)
+            match = re.search(r'\\boxed{([^}]*)}', text)
             return match.group(1) if match else None
 
         ground_truth = get_boxed_answer(sample['solution'])

src/twinkle/metric/completion_and_reward.py

@@ -20,15 +20,16 @@ def reset(self):
         self.rewards = {}
         self.completion_lengths = []
 
-    def accumulate(self,
-                   inputs=None, # ignore
-                   outputs=None,# ignore
-                   *,
-                   rewards=None,
-                   completion_lengths=None,
-                   generate_time: float = None,
-                   weight_sync_time: float = None,
-                   **kwargs):
+    def accumulate(
+            self,
+            inputs=None,  # ignore
+            outputs=None,  # ignore
+            *,
+            rewards=None,
+            completion_lengths=None,
+            generate_time: float = None,
+            weight_sync_time: float = None,
+            **kwargs):
         if completion_lengths is None:
             completion_lengths = []
         if rewards is None:

src/twinkle/sampler/vllm_sampler/vllm_engine.py

@@ -410,7 +410,7 @@ async def _get_or_load_lora(
             await self.engine.add_lora(lora_request)
             self._lora_request_cache[lora_path] = lora_request
             return lora_request
-        except Exception as e:
+        except Exception:  # noqa
             return None
 
     async def sleep(self, level: int = 2) -> None: