sycommon-python-lib 0.2.6a3__tar.gz → 0.2.6a4__tar.gz

{sycommon_python_lib-0.2.6a3 → sycommon_python_lib-0.2.6a4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sycommon-python-lib
-Version: 0.2.6a3
+Version: 0.2.6a4
 Summary: Add your description here
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown

{sycommon_python_lib-0.2.6a3 → sycommon_python_lib-0.2.6a4}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sycommon-python-lib"
-version = "0.2.6a3"
+version = "0.2.6a4"
 description = "Add your description here"
 readme = "README.md"
 requires-python = ">=3.11"

{sycommon_python_lib-0.2.6a3 → sycommon_python_lib-0.2.6a4}/src/sycommon/agent/deep_agent.py

@@ -982,7 +982,7 @@ class DeepAgent:
         不在此处重试整个 graph。
         """
         import httpx
-        from openai import APIConnectionError, APIError, APITimeoutError
+        from openai import APIConnectionError, APIError, APITimeoutError, BadRequestError
 
         last_error = None
         sandbox_retries = 0
@@ -1010,6 +1010,18 @@ class DeepAgent:
                         print(
                             f"[DeepAgent] _astream_with_retry: stream ended normally, total_chunks={chunk_count}", flush=True)
                         return
+                    except BadRequestError as e:
+                        # BadRequestError（含 ContextOverflowError）不重试：
+                        # 上下文溢出由 SummarizationMiddleware 在模型调用层处理，
+                        # 重试只会浪费时间且无法恢复。
+                        from langchain_core.exceptions import ContextOverflowError as _COE
+                        if isinstance(e, _COE):
+                            SYLogger.warning(
+                                f"[DeepAgent] ContextOverflowError: {e}, 不重试，交由 SummarizationMiddleware 处理")
+                        else:
+                            SYLogger.warning(
+                                f"[DeepAgent] BadRequestError: {e}, 不重试")
+                        raise
                     except (APIConnectionError, APIError, APITimeoutError, ConnectionError, httpx.RemoteProtocolError) as e:
                         last_error = e
                         print(

{sycommon_python_lib-0.2.6a3 → sycommon_python_lib-0.2.6a4}/src/sycommon/agent/summarization_utils.py

@@ -25,6 +25,68 @@ if TYPE_CHECKING:
 
 logger = logging.getLogger(__name__)
 
+
+# ==================== ContextOverflowError 识别补丁 ====================
+# MiniMax2.5 等兼容 OpenAI API 的模型返回上下文溢出错误时，
+# 错误格式为 "This model's maximum context length is N tokens..."，
+# 而 langchain_openai 的 _handle_openai_bad_request 只识别 OpenAI 原生格式：
+#   - "context_length_exceeded"
+#   - "Input tokens exceed the configured limit"
+#   - "prompt is too long"
+# 导致 MiniMax 的 BadRequestError 不被转换为 ContextOverflowError，
+# SummarizationMiddleware 无法捕获并触发压缩。
+# 此补丁扩展错误模式匹配，使 MiniMax/vLLM 等兼容 API 的溢出错误也被正确识别。
+try:
+    import langchain_openai.chat_models.base as _lc_oai_base
+    import openai
+    from langchain_core.exceptions import ContextOverflowError
+
+    _orig_handle_bad_request = _lc_oai_base._handle_openai_bad_request
+
+    def _patched_handle_openai_bad_request(e: openai.BadRequestError) -> None:
+        """扩展版 _handle_openai_bad_request，识别更多上下文溢出错误格式。"""
+        error_str = str(e)
+        error_msg = getattr(e, 'message', '') or ''
+
+        # 新增模式：覆盖 MiniMax/vLLM/其他兼容 API 的错误格式
+        if (
+            "maximum context length is" in error_str
+            or "Please reduce the length of the input prompt" in error_msg
+            or "maximum context length" in error_msg
+        ):
+            raise _lc_oai_base.OpenAIContextOverflowError(
+                message=e.message, response=e.response, body=e.body
+            ) from e
+
+        # 回退到原始处理逻辑
+        _orig_handle_bad_request(e)
+
+    _lc_oai_base._handle_openai_bad_request = _patched_handle_openai_bad_request
+
+    # 同步 patch _handle_openai_api_error
+    _orig_handle_api_error = _lc_oai_base._handle_openai_api_error
+
+    def _patched_handle_openai_api_error(e: openai.APIError) -> None:
+        """扩展版 _handle_openai_api_error，识别更多上下文溢出错误格式。"""
+        error_str = str(e)
+        if "maximum context length is" in error_str:
+            raise _lc_oai_base.OpenAIAPIContextOverflowError(
+                message=e.message, request=e.request, body=e.body
+            ) from e
+        _orig_handle_api_error(e)
+
+    _lc_oai_base._handle_openai_api_error = _patched_handle_openai_api_error
+
+    logger.info(
+        "[ContextOverflowPatch] 已扩展 _handle_openai_bad_request/_handle_openai_api_error "
+        "以识别 MiniMax/vLLM 上下文溢出错误格式"
+    )
+except ImportError:
+    logger.debug("[ContextOverflowPatch] langchain_openai 不可用，跳过补丁")
+except Exception as _patch_err:
+    logger.warning(f"[ContextOverflowPatch] 补丁应用失败: {_patch_err}")
+
+
 # 中文摘要提示词，替代 langchain DEFAULT_SUMMARY_PROMPT
 # 使用纯自然段落格式（无标题/小标题/列表标记），避免结构化标题被模型回显到用户输出
 CUSTOM_SUMMARY_PROMPT = """<role>

{sycommon_python_lib-0.2.6a3 → sycommon_python_lib-0.2.6a4}/src/sycommon/llm/token_utils.py

@@ -78,6 +78,11 @@ def clamp_max_tokens(
     将 model.max_tokens = min(max_output_tokens, context_window - input - buffer)，
     最低保底 min_output。
 
+    安全措施：
+    - context_window 乘以 0.95 作为有效上限，防止模型实际上下文窗口小于配置值
+      （如 MiniMax2.5 配置 200000 但实际为 196608）
+    - tiktoken 计数可能与模型内部 tokenizer 存在偏差，额外余量可以覆盖
+
     Args:
         model: 拥有 ``max_tokens`` 属性的 ChatModel 实例。
         messages: 即将发送给模型的消息列表。
@@ -89,8 +94,10 @@ def clamp_max_tokens(
     Returns:
         调整后的 max_tokens 值。
     """
+    # 有效上下文窗口 = 配置值 * 0.95，防止配置值大于模型实际上下文
+    effective_context = int(context_window * 0.95)
     input_est = count_messages_tokens(messages)
-    available = context_window - input_est - buffer
+    available = effective_context - input_est - buffer
     dynamic_max = min(max_output_tokens, max(available, min_output))
     model.max_tokens = dynamic_max
     return dynamic_max

{sycommon_python_lib-0.2.6a3 → sycommon_python_lib-0.2.6a4}/src/sycommon_python_lib.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sycommon-python-lib
-Version: 0.2.6a3
+Version: 0.2.6a4
 Summary: Add your description here
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown