sycommon-python-lib 0.2.3a6__tar.gz → 0.2.3a8__tar.gz

{sycommon_python_lib-0.2.3a6 → sycommon_python_lib-0.2.3a8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sycommon-python-lib
-Version: 0.2.3a6
+Version: 0.2.3a8
 Summary: Add your description here
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
@@ -9,16 +9,16 @@ Requires-Dist: aiohttp>=3.13.5
 Requires-Dist: aiomysql>=0.3.2
 Requires-Dist: anyio>=4.12.1
 Requires-Dist: decorator>=5.2.1
-Requires-Dist: deepagents>=0.5.7
+Requires-Dist: deepagents>=0.6.1
 Requires-Dist: elasticsearch>=9.4.0
 Requires-Dist: fastapi>=0.136.1
 Requires-Dist: jinja2>=3.1.6
 Requires-Dist: kafka-python>=2.3.1
-Requires-Dist: langchain>=1.2.17
-Requires-Dist: langchain-core>=1.3.3
+Requires-Dist: langchain>=1.3.0
+Requires-Dist: langchain-core>=1.4.0
 Requires-Dist: langchain-openai>=1.2.1
-Requires-Dist: langfuse>=4.5.1
-Requires-Dist: langgraph>=1.1.10
+Requires-Dist: langfuse>=4.6.1
+Requires-Dist: langgraph>=1.2.0
 Requires-Dist: langgraph-checkpoint-redis>=0.4.1
 Requires-Dist: ldap3>=2.9.1
 Requires-Dist: loguru>=0.7.3
@@ -28,10 +28,10 @@ Requires-Dist: psutil>=7.2.2
 Requires-Dist: pyxxl>=0.4.6
 Requires-Dist: pydantic>=2.13.4
 Requires-Dist: python-dotenv>=1.2.2
-Requires-Dist: python-multipart>=0.0.27
+Requires-Dist: python-multipart>=0.0.28
 Requires-Dist: pyyaml>=6.0.3
 Requires-Dist: redis>=7.3.0
-Requires-Dist: sentry-sdk[fastapi]>=2.59.0
+Requires-Dist: sentry-sdk[fastapi]>=2.60.0
 Requires-Dist: sqlalchemy[asyncio]>=2.0.48
 Requires-Dist: starlette[full]>=1.0.0
 Requires-Dist: tiktoken>=0.12.0

{sycommon_python_lib-0.2.3a6 → sycommon_python_lib-0.2.3a8}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sycommon-python-lib"
-version = "0.2.3a6"
+version = "0.2.3a8"
 description = "Add your description here"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -10,16 +10,16 @@ dependencies = [
     "aiomysql>=0.3.2",
     "anyio>=4.12.1",
     "decorator>=5.2.1",
-    "deepagents>=0.5.7",
+    "deepagents>=0.6.1",
     "elasticsearch>=9.4.0",
     "fastapi>=0.136.1",
     "jinja2>=3.1.6",
     "kafka-python>=2.3.1",
-    "langchain>=1.2.17",
-    "langchain-core>=1.3.3",
+    "langchain>=1.3.0",
+    "langchain-core>=1.4.0",
     "langchain-openai>=1.2.1",
-    "langfuse>=4.5.1",
-    "langgraph>=1.1.10",
+    "langfuse>=4.6.1",
+    "langgraph>=1.2.0",
     "langgraph-checkpoint-redis>=0.4.1",
     "ldap3>=2.9.1",
     "loguru>=0.7.3",
@@ -29,10 +29,10 @@ dependencies = [
     "pyxxl>=0.4.6",
     "pydantic>=2.13.4",
     "python-dotenv>=1.2.2",
-    "python-multipart>=0.0.27",
+    "python-multipart>=0.0.28",
     "pyyaml>=6.0.3",
     "redis>=7.3.0",
-    "sentry-sdk[fastapi]>=2.59.0",
+    "sentry-sdk[fastapi]>=2.60.0",
     "sqlalchemy[asyncio]>=2.0.48",
     "starlette[full]>=1.0.0",
     "tiktoken>=0.12.0",

sycommon_python_lib-0.2.3a8/src/sycommon/agent/summarization_utils.py

@@ -0,0 +1,144 @@
+"""上下文压缩 middleware 构建工具。
+
+根据 nacos 中配置的模型 maxTokens，用绝对 token 数设置压缩阈值。
+中文场景下 count_tokens_approximately 默认 chars_per_token=4.0 会低估约 2.1 倍，
+因此传入自定义 token_counter 修正偏差，并按真实 token 数设定阈值。
+"""
+
+from __future__ import annotations
+
+import functools
+import logging
+from typing import TYPE_CHECKING
+
+from deepagents.middleware.summarization import (
+    SummarizationMiddleware,
+    SummarizationToolMiddleware,
+)
+from langchain_core.messages.utils import count_tokens_approximately
+
+if TYPE_CHECKING:
+    from langchain_core.language_models import BaseChatModel
+    from deepagents.backends.protocol import BACKEND_TYPES
+
+logger = logging.getLogger(__name__)
+
+# 中文场景修正：chars_per_token 从 4.0 改为 2.0，使近似计数接近真实 token 数
+_chinese_token_counter = functools.partial(count_tokens_approximately, chars_per_token=2.0)
+
+
+def _patched_compute_summarization_defaults(model):
+    """覆盖 deepagents 默认值，返回中文场景修正后的绝对 token 阈值。
+
+    不走 fraction 路径（fraction 依赖 model.profile.max_input_tokens，
+    但内置 middleware 用默认 chars_per_token=4.0 计数，fraction 基于的基数不准）。
+    直接返回修正后的绝对 token 数。
+    """
+    try:
+        from sycommon.config.Config import Config
+        model_name = getattr(model, 'model_name', None) or getattr(model, 'model', None)
+        if model_name:
+            llm_cfg = Config().get_llm_config(model_name)
+            max_tokens = llm_cfg.get("maxTokens", 72000)
+        else:
+            max_tokens = 72000
+    except Exception:
+        max_tokens = 72000
+
+    trigger = int(max_tokens * 0.70)
+    keep = int(max_tokens * 0.10)
+    return {
+        "trigger": ("tokens", trigger),
+        "keep": ("tokens", keep),
+        "truncate_args_settings": {
+            "trigger": ("tokens", trigger),
+            "keep": ("tokens", keep),
+        },
+    }
+
+
+# monkey-patch：替换 deepagents 的默认计算函数
+import deepagents.middleware.summarization as _summ_mod
+_summ_mod.compute_summarization_defaults = _patched_compute_summarization_defaults
+
+# monkey-patch：在内置 middleware 的 awrap_model_call 中注入日志
+_OrigDeepAgentsSumm = _summ_mod._DeepAgentsSummarizationMiddleware
+_orig_awrap_model_call = _OrigDeepAgentsSumm.awrap_model_call
+
+
+async def _patched_awrap_model_call(self, request, handler):
+    effective_messages = self._get_effective_messages(request)
+    truncated_messages, _ = self._truncate_args(
+        effective_messages, request.system_message, request.tools,
+    )
+    counted_messages = [request.system_message, *truncated_messages] if request.system_message is not None else truncated_messages
+    try:
+        total_tokens = self.token_counter(counted_messages, tools=request.tools)
+    except TypeError:
+        total_tokens = self.token_counter(counted_messages)
+    should = self._should_summarize(truncated_messages, total_tokens)
+    print(f"[TokenCount] tokens={total_tokens} msgs={len(counted_messages)} should_summarize={should}")
+    return await _orig_awrap_model_call(self, request, handler)
+
+
+_OrigDeepAgentsSumm.awrap_model_call = _patched_awrap_model_call
+
+
+def build_summarization_middleware(
+    model: BaseChatModel,
+    model_name: str,
+    backend: "BACKEND_TYPES",
+    *,
+    trigger_fraction: float = 0.70,
+    keep_fraction: float = 0.10,
+    default_max_tokens: int = 200000,
+) -> SummarizationToolMiddleware:
+    """根据模型上下文窗口大小构建 compact_conversation 工具 middleware。
+
+    自动压缩由 deepagents 内置的 SummarizationMiddleware 处理，
+    它通过 model.profile.max_input_tokens（由 LLMWithTokenTracking 从 nacos 配置注入）
+    计算压缩阈值。
+
+    此处创建的 SummarizationToolMiddleware 仅提供 compact_conversation 手动压缩工具，
+    使用 nacos 配置的 maxTokens 作为其内部 SummarizationMiddleware 的阈值参数。
+
+    Args:
+        model: LLM 实例。
+        model_name: 模型名称（用于从 nacos 读取配置）。
+        backend: 后端实例。
+        trigger_fraction: 触发压缩的上下文窗口比例，默认 70%。
+        keep_fraction: 压缩后保留的上下文窗口比例，默认 10%。
+        default_max_tokens: 无法从配置读取时的默认上下文窗口大小。
+
+    Returns:
+        SummarizationToolMiddleware 实例（提供 compact_conversation 工具）。
+    """
+    try:
+        from sycommon.config.Config import Config
+
+        llm_cfg = Config().get_llm_config(model_name)
+        max_tokens = llm_cfg.get("maxTokens", default_max_tokens)
+    except Exception:
+        max_tokens = default_max_tokens
+
+    trigger_tokens = int(max_tokens * trigger_fraction)
+    keep_tokens = int(max_tokens * keep_fraction)
+
+    summ = SummarizationMiddleware(
+        model=model,
+        backend=backend,
+        trigger=("tokens", trigger_tokens),
+        keep=("tokens", keep_tokens),
+        token_counter=_chinese_token_counter,
+        trim_tokens_to_summarize=None,
+        truncate_args_settings={
+            "trigger": ("tokens", trigger_tokens),
+            "keep": ("tokens", keep_tokens),
+            "max_length": 2000,
+        },
+    )
+
+    print(f"[Summarization] compact_conversation 工具配置: model={model_name}, max_tokens={max_tokens}, "
+          f"trigger={trigger_tokens} tokens ({trigger_fraction:.0%}), "
+          f"keep={keep_tokens} tokens ({keep_fraction:.0%})")
+    return SummarizationToolMiddleware(summ)

{sycommon_python_lib-0.2.3a6 → sycommon_python_lib-0.2.3a8}/src/sycommon_python_lib.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sycommon-python-lib
-Version: 0.2.3a6
+Version: 0.2.3a8
 Summary: Add your description here
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
@@ -9,16 +9,16 @@ Requires-Dist: aiohttp>=3.13.5
 Requires-Dist: aiomysql>=0.3.2
 Requires-Dist: anyio>=4.12.1
 Requires-Dist: decorator>=5.2.1
-Requires-Dist: deepagents>=0.5.7
+Requires-Dist: deepagents>=0.6.1
 Requires-Dist: elasticsearch>=9.4.0
 Requires-Dist: fastapi>=0.136.1
 Requires-Dist: jinja2>=3.1.6
 Requires-Dist: kafka-python>=2.3.1
-Requires-Dist: langchain>=1.2.17
-Requires-Dist: langchain-core>=1.3.3
+Requires-Dist: langchain>=1.3.0
+Requires-Dist: langchain-core>=1.4.0
 Requires-Dist: langchain-openai>=1.2.1
-Requires-Dist: langfuse>=4.5.1
-Requires-Dist: langgraph>=1.1.10
+Requires-Dist: langfuse>=4.6.1
+Requires-Dist: langgraph>=1.2.0
 Requires-Dist: langgraph-checkpoint-redis>=0.4.1
 Requires-Dist: ldap3>=2.9.1
 Requires-Dist: loguru>=0.7.3
@@ -28,10 +28,10 @@ Requires-Dist: psutil>=7.2.2
 Requires-Dist: pyxxl>=0.4.6
 Requires-Dist: pydantic>=2.13.4
 Requires-Dist: python-dotenv>=1.2.2
-Requires-Dist: python-multipart>=0.0.27
+Requires-Dist: python-multipart>=0.0.28
 Requires-Dist: pyyaml>=6.0.3
 Requires-Dist: redis>=7.3.0
-Requires-Dist: sentry-sdk[fastapi]>=2.59.0
+Requires-Dist: sentry-sdk[fastapi]>=2.60.0
 Requires-Dist: sqlalchemy[asyncio]>=2.0.48
 Requires-Dist: starlette[full]>=1.0.0
 Requires-Dist: tiktoken>=0.12.0

{sycommon_python_lib-0.2.3a6 → sycommon_python_lib-0.2.3a8}/src/sycommon_python_lib.egg-info/requires.txt

@@ -3,16 +3,16 @@ aiohttp>=3.13.5
 aiomysql>=0.3.2
 anyio>=4.12.1
 decorator>=5.2.1
-deepagents>=0.5.7
+deepagents>=0.6.1
 elasticsearch>=9.4.0
 fastapi>=0.136.1
 jinja2>=3.1.6
 kafka-python>=2.3.1
-langchain>=1.2.17
-langchain-core>=1.3.3
+langchain>=1.3.0
+langchain-core>=1.4.0
 langchain-openai>=1.2.1
-langfuse>=4.5.1
-langgraph>=1.1.10
+langfuse>=4.6.1
+langgraph>=1.2.0
 langgraph-checkpoint-redis>=0.4.1
 ldap3>=2.9.1
 loguru>=0.7.3
@@ -22,10 +22,10 @@ psutil>=7.2.2
 pyxxl>=0.4.6
 pydantic>=2.13.4
 python-dotenv>=1.2.2
-python-multipart>=0.0.27
+python-multipart>=0.0.28
 pyyaml>=6.0.3
 redis>=7.3.0
-sentry-sdk[fastapi]>=2.59.0
+sentry-sdk[fastapi]>=2.60.0
 sqlalchemy[asyncio]>=2.0.48
 starlette[full]>=1.0.0
 tiktoken>=0.12.0

sycommon_python_lib-0.2.3a6/src/sycommon/agent/summarization_utils.py

@@ -1,77 +0,0 @@
-"""上下文压缩 middleware 构建工具。
-
-根据 nacos 中配置的模型 maxTokens，用绝对 token 数设置压缩阈值。
-仅创建 SummarizationToolMiddleware（compact_conversation 工具），
-自动压缩由 deepagents 内置的 SummarizationMiddleware 处理（通过 model.profile 获取阈值）。
-"""
-
-from __future__ import annotations
-
-from typing import TYPE_CHECKING
-
-from deepagents.middleware.summarization import (
-    SummarizationMiddleware,
-    SummarizationToolMiddleware,
-)
-
-if TYPE_CHECKING:
-    from langchain_core.language_models import BaseChatModel
-    from deepagents.backends.protocol import BACKEND_TYPES
-
-
-def build_summarization_middleware(
-    model: BaseChatModel,
-    model_name: str,
-    backend: "BACKEND_TYPES",
-    *,
-    trigger_fraction: float = 0.70,
-    keep_fraction: float = 0.10,
-    default_max_tokens: int = 200000,
-) -> SummarizationToolMiddleware:
-    """根据模型上下文窗口大小构建 compact_conversation 工具 middleware。
-
-    自动压缩由 deepagents 内置的 SummarizationMiddleware 处理，
-    它通过 model.profile.max_input_tokens（由 LLMWithTokenTracking 从 nacos 配置注入）
-    计算压缩阈值。
-
-    此处创建的 SummarizationToolMiddleware 仅提供 compact_conversation 手动压缩工具，
-    使用 nacos 配置的 maxTokens 作为其内部 SummarizationMiddleware 的阈值参数。
-
-    Args:
-        model: LLM 实例。
-        model_name: 模型名称（用于从 nacos 读取配置）。
-        backend: 后端实例。
-        trigger_fraction: 触发压缩的上下文窗口比例，默认 70%。
-        keep_fraction: 压缩后保留的上下文窗口比例，默认 10%。
-        default_max_tokens: 无法从配置读取时的默认上下文窗口大小。
-
-    Returns:
-        SummarizationToolMiddleware 实例（提供 compact_conversation 工具）。
-    """
-    try:
-        from sycommon.config.Config import Config
-
-        llm_cfg = Config().get_llm_config(model_name)
-        max_tokens = llm_cfg.get("maxTokens", default_max_tokens)
-    except Exception:
-        max_tokens = default_max_tokens
-
-    trigger_tokens = int(max_tokens * trigger_fraction)
-    keep_tokens = int(max_tokens * keep_fraction)
-
-    summ = SummarizationMiddleware(
-        model=model,
-        backend=backend,
-        trigger=("tokens", trigger_tokens),
-        keep=("tokens", keep_tokens),
-        trim_tokens_to_summarize=None,
-        truncate_args_settings={
-            "trigger": ("tokens", trigger_tokens),
-            "keep": ("tokens", keep_tokens),
-            "max_length": 2000,
-        },
-    )
-    print(f"[Summarization] compact_conversation 工具配置: model={model_name}, max_tokens={max_tokens}, "
-          f"trigger={trigger_tokens} tokens ({trigger_fraction:.0%}), "
-          f"keep={keep_tokens} tokens ({keep_fraction:.0%})")
-    return SummarizationToolMiddleware(summ)