sycommon-python-lib 0.2.5a2__py3-none-any.whl → 0.2.5a3__py3-none-any.whl

sycommon/agent/deep_agent.py

@@ -484,6 +484,9 @@ class DeepAgent:
                     SYLogger.warning(f"[DeepAgent] Token 兜底统计失败: {te}")
 
             event = ChatEventBuilder.done()
+            if total_input_tokens > 0 or total_output_tokens > 0:
+                event.data["input_tokens"] = total_input_tokens
+                event.data["output_tokens"] = total_output_tokens
             if on_event:
                 await on_event(event)
             yield event

sycommon/tests/test_minimax_reasoning.py

@@ -0,0 +1,290 @@
+"""
+MiniMax2.5 模型推理内容(reasoning_content)测试
+
+测试 MiniMax2.5 模型的流式输出是否正确返回 reasoning_content，
+以及 deep_agent 是否正确将其包装为 AI_CHUNK 事件。
+
+使用方式:
+    cd sycommon-python-lib
+    PYTHONPATH=src python -m src.sycommon.tests.test_minimax_reasoning
+"""
+
+import asyncio
+import json
+from langchain.chat_models import init_chat_model
+from langchain_core.messages import HumanMessage, SystemMessage
+
+# MiniMax2.5 模型配置
+MODEL = "MiniMax2.5"
+PROVIDER = "openai"
+BASE_URL = "http://221.130.59.59:8888/8vllm/v1"
+API_KEY = "-"
+MAX_TOKENS = 128000
+
+
+def create_llm(streaming: bool = False, thinking: bool = True):
+    """创建 MiniMax2.5 LLM 实例"""
+    extra_body = {
+        "top_k": 20,
+        "min_p": 0.0,
+        "repetition_penalty": 1.0,
+    }
+
+    if thinking:
+        extra_body["chat_template_kwargs"] = {"enable_thinking": True}
+        temperature = 0.6
+        top_p = 0.95
+        presence_penalty = 0.0
+    else:
+        extra_body["chat_template_kwargs"] = {"enable_thinking": False}
+        temperature = 0.7
+        top_p = 0.8
+        presence_penalty = 1.5
+
+    if streaming:
+        extra_body["stream_options"] = {"include_usage": True}
+
+    return init_chat_model(
+        model_provider=PROVIDER,
+        model=MODEL,
+        base_url=BASE_URL,
+        api_key=API_KEY,
+        max_tokens=MAX_TOKENS,
+        streaming=streaming,
+        temperature=temperature,
+        top_p=top_p,
+        presence_penalty=presence_penalty,
+        extra_body=extra_body,
+        timeout=60,
+    )
+
+
+async def test_non_stream_basic():
+    """测试非流式基本调用"""
+    print("\n" + "=" * 60)
+    print("测试1: 非流式基本调用（thinking=True）")
+    print("=" * 60)
+
+    llm = create_llm(streaming=False, thinking=True)
+    messages = [
+        SystemMessage(content="你是一个有用的助手，使用中文回答。"),
+        HumanMessage(content="1+1等于几？请简短回答。"),
+    ]
+
+    response = await llm.ainvoke(messages)
+
+    print(f"类型: {type(response).__name__}")
+    print(f"内容: {response.content[:500]}")
+
+    # 检查 reasoning_content
+    reasoning = getattr(response, "reasoning_content", None)
+    kwargs_reasoning = getattr(response, "additional_kwargs", {}).get(
+        "reasoning_content", None
+    )
+    print(f"reasoning_content 属性: {repr(reasoning)[:200] if reasoning else 'None'}")
+    print(f"additional_kwargs.reasoning_content: {repr(kwargs_reasoning)[:200] if kwargs_reasoning else 'None'}")
+
+    # 检查 usage
+    usage = getattr(response, "usage_metadata", None)
+    print(f"usage_metadata: {usage}")
+
+    assert response.content, "响应内容为空"
+    print("[PASS] 非流式调用成功\n")
+
+
+async def test_non_stream_no_thinking():
+    """测试非流式无思考模式"""
+    print("=" * 60)
+    print("测试2: 非流式调用（thinking=False）")
+    print("=" * 60)
+
+    llm = create_llm(streaming=False, thinking=False)
+    messages = [HumanMessage(content="今天星期几？简短回答。")]
+
+    response = await llm.ainvoke(messages)
+
+    print(f"内容: {response.content[:500]}")
+
+    reasoning = getattr(response, "reasoning_content", None)
+    kwargs_reasoning = getattr(response, "additional_kwargs", {}).get(
+        "reasoning_content", None
+    )
+    print(f"reasoning_content: {repr(reasoning)[:200] if reasoning else 'None'}")
+    print(f"additional_kwargs.reasoning_content: {repr(kwargs_reasoning)[:200] if kwargs_reasoning else 'None'}")
+
+    print("[PASS] 无思考模式调用成功\n")
+
+
+async def test_stream_with_reasoning():
+    """测试流式输出 + reasoning_content"""
+    print("=" * 60)
+    print("测试3: 流式调用（thinking=True）- 检查 reasoning_content")
+    print("=" * 60)
+
+    llm = create_llm(streaming=True, thinking=True)
+    messages = [
+        SystemMessage(content="你是一个有用的助手，使用中文回答。"),
+        HumanMessage(content="请思考一下：一个笼子里有鸡和兔，共10个头，26条腿，鸡和兔各几只？"),
+    ]
+
+    reasoning_chunks = []
+    content_chunks = []
+    chunk_types = {}
+
+    async for chunk in llm.astream(messages):
+        chunk_type = type(chunk).__name__
+
+        # 统计 chunk 类型
+        chunk_types[chunk_type] = chunk_types.get(chunk_type, 0) + 1
+
+        # 提取 reasoning_content
+        reasoning = getattr(chunk, "reasoning_content", None)
+        kwargs_reasoning = getattr(chunk, "additional_kwargs", {}).get(
+            "reasoning_content", None
+        )
+
+        if reasoning:
+            reasoning_chunks.append(reasoning)
+            # 只打印前几条
+            if len(reasoning_chunks) <= 5:
+                print(f"  [reasoning] {repr(reasoning[:100])}")
+
+        if kwargs_reasoning:
+            reasoning_chunks.append(kwargs_reasoning)
+            if len(reasoning_chunks) <= 5:
+                print(f"  [kwargs.reasoning] {repr(kwargs_reasoning[:100])}")
+
+        # 提取内容
+        if chunk.content:
+            content_chunks.append(chunk.content)
+
+    print(f"\n--- 流式统计 ---")
+    print(f"chunk 类型分布: {chunk_types}")
+    print(f"reasoning chunks 数量: {len(reasoning_chunks)}")
+    print(f"content chunks 数量: {len(content_chunks)}")
+
+    if reasoning_chunks:
+        full_reasoning = "".join(reasoning_chunks)
+        print(f"完整 reasoning (前500字): {full_reasoning[:500]}")
+    else:
+        print("WARNING: 未收到 reasoning_content")
+
+    full_content = "".join(content_chunks)
+    print(f"完整 content (前500字): {full_content[:500]}")
+
+    print("[PASS] 流式调用完成\n")
+
+
+async def test_stream_no_thinking():
+    """测试流式输出无思考模式"""
+    print("=" * 60)
+    print("测试4: 流式调用（thinking=False）")
+    print("=" * 60)
+
+    llm = create_llm(streaming=True, thinking=False)
+    messages = [HumanMessage(content="用一句话介绍北京。")]
+
+    reasoning_chunks = []
+    content_chunks = []
+
+    async for chunk in llm.astream(messages):
+        reasoning = getattr(chunk, "reasoning_content", None) or getattr(
+            chunk, "additional_kwargs", {}
+        ).get("reasoning_content", None)
+
+        if reasoning:
+            reasoning_chunks.append(reasoning)
+        if chunk.content:
+            content_chunks.append(chunk.content)
+
+    full_content = "".join(content_chunks)
+    print(f"content: {full_content[:500]}")
+    print(f"reasoning chunks: {len(reasoning_chunks)} (应为 0)")
+
+    assert content_chunks, "没有收到内容"
+    print("[PASS] 无思考模式流式调用成功\n")
+
+
+async def test_reasoning_to_chat_event():
+    """测试将 reasoning_content 转换为 ChatEvent（模拟 deep_agent 逻辑）"""
+    print("=" * 60)
+    print("测试5: reasoning_content -> ChatEvent 转换")
+    print("=" * 60)
+
+    from sycommon.agent.chat_events import EventType, ChatEvent
+
+    llm = create_llm(streaming=True, thinking=True)
+    messages = [HumanMessage(content="9.11 和 9.9 哪个大？请仔细思考。")]
+
+    events = []
+
+    async for msg in llm.astream(messages):
+        msg_type = msg.__class__.__name__
+        content = msg.content or ""
+
+        if msg_type in ("AIMessage", "AIMessageChunk"):
+            # 模拟 deep_agent 的 reasoning_content 提取逻辑
+            reasoning_content = (
+                getattr(msg, "reasoning_content", None)
+                or getattr(msg, "additional_kwargs", {}).get(
+                    "reasoning_content", None
+                )
+            )
+
+            if reasoning_content:
+                event = ChatEvent(
+                    EventType.AI_CHUNK,
+                    {"content": reasoning_content, "reasoning": True},
+                )
+                events.append(event)
+
+            if content:
+                event = ChatEvent(
+                    EventType.AI_CHUNK,
+                    {"content": content, "reasoning": False},
+                )
+                events.append(event)
+
+    reasoning_events = [e for e in events if e.data.get("reasoning")]
+    content_events = [e for e in events if not e.data.get("reasoning")]
+
+    print(f"总事件数: {len(events)}")
+    print(f"reasoning 事件: {len(reasoning_events)}")
+    print(f"content 事件: {len(content_events)}")
+
+    if reasoning_events:
+        sample = reasoning_events[0]
+        print(f"reasoning 事件示例: type={sample.type}, data={repr(sample.data)[:200]}")
+    if content_events:
+        sample = content_events[0]
+        print(f"content 事件示例: type={sample.type}, data={repr(sample.data)[:200]}")
+
+    # 验证事件可以序列化
+    if events:
+        serialized = events[0].to_dict()
+        print(f"序列化示例: {json.dumps(serialized, ensure_ascii=False)[:300]}")
+
+    assert events, "没有生成任何事件"
+    print("[PASS] ChatEvent 转换成功\n")
+
+
+async def run_all():
+    """运行所有测试"""
+    print("=" * 60)
+    print(f"MiniMax2.5 推理内容测试")
+    print(f"模型: {MODEL} | 地址: {BASE_URL}")
+    print("=" * 60)
+
+    await test_non_stream_basic()
+    await test_non_stream_no_thinking()
+    await test_stream_with_reasoning()
+    await test_stream_no_thinking()
+    await test_reasoning_to_chat_event()
+
+    print("=" * 60)
+    print("全部测试完成!")
+    print("=" * 60)
+
+
+if __name__ == "__main__":
+    asyncio.run(run_all())

{sycommon_python_lib-0.2.5a2.dist-info → sycommon_python_lib-0.2.5a3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sycommon-python-lib
-Version: 0.2.5a2
+Version: 0.2.5a3
 Summary: Add your description here
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown

{sycommon_python_lib-0.2.5a2.dist-info → sycommon_python_lib-0.2.5a3.dist-info}/RECORD

@@ -129,7 +129,7 @@ sycommon/services.py,sha256=UtWn36dPmB9g1CTKkNgzbU8M28iOkZd2BtdhvDZLfbk,22214
 sycommon/agent/__init__.py,sha256=mxceAeUifQ-DKvWp7ZEJIFlmOCb5wpYHPGQw3rwEN8I,4378
 sycommon/agent/agent_manager.py,sha256=UhhaekEumT7g4v_Z1UB4jTp13X0n8M8erYaQdkGGWkA,13620
 sycommon/agent/chat_events.py,sha256=bWAMWYIZ2L_yqUcn5jq9ius_lQxLHEv4zQLEqX6UaeM,13190
-sycommon/agent/deep_agent.py,sha256=WTGKWx0MxmiIbbrKPTbQl6on2w4LQRWtH0hdiViA-dE,31709
+sycommon/agent/deep_agent.py,sha256=n2LCUF8RDAqwjpY-PP9-L_URkqXuxBSRHO4ec07x108,31905
 sycommon/agent/multi_agent_team.py,sha256=NHmsUNwe3huguUUzbeoiOjgo9wh4-eXH0AjPDvm1dP4,26781
 sycommon/agent/summarization_utils.py,sha256=PRCIFtYBrH0bbSxsIc-qpC4iEXJzk72UuR7u5mQTt2w,7360
 sycommon/agent/mcp/__init__.py,sha256=iKrdDhIrFsNIkqG_kgcwNe-nOiM6uVfolKv44LfQ-FQ,636
@@ -254,6 +254,7 @@ sycommon/synacos/param.py,sha256=KcfSkxnXOa0TGmCjY8hdzU9pzUsA8-4PeyBKWI2-568,176
 sycommon/tests/deep_agent_server.py,sha256=t7snT2J6KWZrjJsYVi4TU5jpfvXNz4iuzMk4cc-kQGM,16696
 sycommon/tests/test_deep_agent.py,sha256=34gP2KL8SSmtqqhaA9OV97OAl_I0T4TfEutZrR_0srM,5874
 sycommon/tests/test_email.py,sha256=-oPtYVGQzJ3Cv-op3ZNRMeyYOF-UNidGJC35CHRgjGQ,5442
+sycommon/tests/test_minimax_reasoning.py,sha256=IKrPEf_Ybxg_vz4AITStkb-GKo5NooaLiLK3RbmW_Cw,9354
 sycommon/tests/test_mq.py,sha256=Gpr9Eep-osRkcnlwGeshROEf83Ai3qYbAMHwpoML68o,4366
 sycommon/tests/test_oa_login.py,sha256=5psNnmUst20x-LdjPa_liunhMLGky2uTDVpQzefBnQE,6239
 sycommon/tests/test_real_summarization.py,sha256=7B89es7-UwULk-kq9xUiWH1ylXUO3QDJm4oZWzJNPk0,6193
@@ -269,8 +270,8 @@ sycommon/tools/syemail.py,sha256=BDFhgf7WDOQeTcjxJEQdu0dQhnHFPO_p3eI0-Ni3LhQ,561
 sycommon/tools/timing.py,sha256=OiiE7P07lRoMzX9kzb8sZU9cDb0zNnqIlY5pWqHcnkY,2064
 sycommon/xxljob/__init__.py,sha256=7eoBlQxv-B39IfRSCY2bkqdGYs1QRe1umAWd88VMEEM,86
 sycommon/xxljob/xxljob_service.py,sha256=JIEJaGXhqrTLcyxlyynSrsHg9bBnDNzX-D4qIWLRPUE,6815
-sycommon_python_lib-0.2.5a2.dist-info/METADATA,sha256=Al8QcJtx8mGbEePjfphPJ827JvyX_shWGbxwIKY1taU,7879
-sycommon_python_lib-0.2.5a2.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
-sycommon_python_lib-0.2.5a2.dist-info/entry_points.txt,sha256=gsR4SssKxDWjRU8ggidzNcdMXDPRSKRS7UaGyNP84Qg,92
-sycommon_python_lib-0.2.5a2.dist-info/top_level.txt,sha256=RgphKrg7nJyZ7irJqbxFr-5H2LUYTvI7ivoWZH2hcD0,29
-sycommon_python_lib-0.2.5a2.dist-info/RECORD,,
+sycommon_python_lib-0.2.5a3.dist-info/METADATA,sha256=qlZaDnNw8273-F4OgL7RIsJHkhow7kRWmpf2mHfN5is,7879
+sycommon_python_lib-0.2.5a3.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
+sycommon_python_lib-0.2.5a3.dist-info/entry_points.txt,sha256=gsR4SssKxDWjRU8ggidzNcdMXDPRSKRS7UaGyNP84Qg,92
+sycommon_python_lib-0.2.5a3.dist-info/top_level.txt,sha256=RgphKrg7nJyZ7irJqbxFr-5H2LUYTvI7ivoWZH2hcD0,29
+sycommon_python_lib-0.2.5a3.dist-info/RECORD,,