fix: refine config for rerank input is too long

src/memos/api/config.py

@@ -410,6 +410,8 @@ def get_feedback_reranker_config() -> dict[str, Any]:
                     "url": os.getenv("MOS_RERANKER_URL", "localhost:8000/v1/rerank"),
                     "model": os.getenv("MOS_FEEDBACK_RERANKER_MODEL", "bge-reranker-v2-m3"),
                     "timeout": 10,
+                    "max_query_tokens": int(os.getenv("MOS_RERANKER_MAX_TOKENS", 8000)),
+                    "concate_len": int(os.getenv("MOS_RERANKER_CONCAT_LEN", 1000)),
                     "headers_extra": json.loads(os.getenv("MOS_RERANKER_HEADERS_EXTRA", "{}")),
                     "rerank_source": os.getenv("MOS_RERANK_SOURCE"),
                     "reranker_strategy": os.getenv("MOS_RERANKER_STRATEGY", "single_turn"),

src/memos/reranker/factory.py

@@ -42,6 +42,8 @@ def from_config(cfg: RerankerConfigFactory | None) -> BaseReranker | None:
                 reranker_url=c.get("url") or c.get("endpoint") or c.get("reranker_url"),
                 model=c.get("model", "bge-reranker-v2-m3"),
                 timeout=int(c.get("timeout", 10)),
+                max_query_tokens=min(max(c.get("max_query_tokens", 8000), 100), 8000),
+                concate_len=min(max(c.get("concate_len", 1000), 4), 8000),
                 headers_extra=headers_extra,
                 rerank_source=c.get("rerank_source"),
             )
@@ -60,6 +62,8 @@ def from_config(cfg: RerankerConfigFactory | None) -> BaseReranker | None:
                 reranker_url=c.get("url") or c.get("endpoint") or c.get("reranker_url"),
                 model=c.get("model", "bge-reranker-v2-m3"),
                 timeout=int(c.get("timeout", 10)),
+                max_query_tokens=min(max(c.get("max_query_tokens", 8000), 100), 8000),
+                concate_len=min(max(c.get("concate_len", 1000), 4), 8000),
                 headers_extra=headers_extra,
                 rerank_source=c.get("rerank_source"),
                 reranker_strategy=c.get("reranker_strategy"),

src/memos/reranker/http_bge.py

@@ -80,6 +80,8 @@ def __init__(
         token: str = "",
         model: str = "bge-reranker-v2-m3",
         timeout: int = 10,
+        max_query_tokens: int | None = None,
+        concate_len: int | None = None,
         headers_extra: dict | None = None,
         rerank_source: str | None = None,
         boost_weights: dict[str, float] | None = None,
@@ -107,6 +109,8 @@ def __init__(
         self.token = token or ""
         self.model = model
         self.timeout = timeout
+        self.max_query_tokens = max_query_tokens
+        self.concate_len = concate_len
         self.headers_extra = headers_extra or {}
         self.rerank_source = rerank_source
 
@@ -155,6 +159,10 @@ def rerank(
             Re-ranked items with scores, sorted descending by score.
         """
 
+        if self.max_query_tokens and len(query) > self.max_query_tokens:
+            single_concate_len = self.concate_len // 2
+            query = query[:single_concate_len] + "\n" + query[-single_concate_len:]
+
         if not graph_results:
             return []
 

src/memos/reranker/http_bge_strategy.py

@@ -80,6 +80,8 @@ def __init__(
         token: str = "",
         model: str = "bge-reranker-v2-m3",
         timeout: int = 10,
+        max_query_tokens: int | None = None,
+        concate_len: int | None = None,
         headers_extra: dict | None = None,
         rerank_source: str | None = None,
         boost_weights: dict[str, float] | None = None,
@@ -108,6 +110,8 @@ def __init__(
         self.token = token or ""
         self.model = model
         self.timeout = timeout
+        self.max_query_tokens = max_query_tokens
+        self.concate_len = concate_len
         self.headers_extra = headers_extra or {}
 
         self.boost_weights = (
@@ -149,6 +153,10 @@ def rerank(
         list[tuple[TextualMemoryItem, float]]
             Re-ranked items with scores, sorted descending by score.
         """
+        if self.max_query_tokens and len(query) > self.max_query_tokens:
+            single_concate_len = self.concate_len // 2
+            query = query[:single_concate_len] + "\n" + query[-single_concate_len:]
+
         if not graph_results:
             return []