feat: Enhance OpenAI compatibility and introduce unified LLM service data models

- Refactored model fetching logic to support various authentication methods for OpenAI-compatible APIs. - Introduced new data models for LLM service requests and responses to standardize interactions across layers. - Added an adapter base class for unified request execution across different providers. - Implemented utility functions for building OpenAI-compatible client configurations and request overrides.
2026-03-26 16:15:42 +08:00
parent 6e7daae55d
commit 777d4cb0d2
48 changed files with 5443 additions and 2945 deletions
--- a/src/chat/brain_chat/PFC/action_planner.py
+++ b/src/chat/brain_chat/PFC/action_planner.py
@@ -2,8 +2,8 @@ import time
 from typing import Tuple, Optional  # 增加了 Optional

 from src.common.logger import get_logger
-from src.llm_models.utils_model import LLMRequest
-from src.config.config import global_config, model_config
+from src.services.llm_service import LLMServiceClient
+from src.config.config import global_config
 import random
 from .chat_observer import ChatObserver
 from .pfc_utils import get_items_from_json
@@ -109,8 +109,8 @@ class ActionPlanner:
    """行动规划器"""

    def __init__(self, stream_id: str, private_name: str):
-        self.llm = LLMRequest(
-            model_set=model_config.model_task_config.planner,
+        self.llm = LLMServiceClient(
+            task_name="planner",
            request_type="action_planning",
        )
        self.personality_info = self._get_personality_prompt()
@@ -398,7 +398,8 @@ class ActionPlanner:

        logger.debug(f"[私聊][{self.private_name}]发送到LLM的最终提示词:\n------\n{prompt}\n------")
        try:
-            content, _ = await self.llm.generate_response_async(prompt)
+            generation_result = await self.llm.generate_response(prompt)
+            content = generation_result.response
            logger.debug(f"[私聊][{self.private_name}]LLM (行动规划) 原始返回内容: {content}")

            # --- 初始行动规划解析 ---
@@ -427,7 +428,8 @@ class ActionPlanner:
                    f"[私聊][{self.private_name}]发送到LLM的结束决策提示词:\n------\n{end_decision_prompt}\n------"
                )
                try:
-                    end_content, _ = await self.llm.generate_response_async(end_decision_prompt)  # 再次调用LLM
+                    end_generation_result = await self.llm.generate_response(end_decision_prompt)
+                    end_content = end_generation_result.response  # 再次调用LLM
                    logger.debug(f"[私聊][{self.private_name}]LLM (结束决策) 原始返回内容: {end_content}")

                    # 解析结束决策的JSON
--- a/src/chat/brain_chat/PFC/pfc.py
+++ b/src/chat/brain_chat/PFC/pfc.py
@@ -1,7 +1,7 @@
 from typing import List, Tuple, TYPE_CHECKING
 from src.common.logger import get_logger
-from src.llm_models.utils_model import LLMRequest
-from src.config.config import global_config, model_config
+from src.services.llm_service import LLMServiceClient
+from src.config.config import global_config
 import random
 from .chat_observer import ChatObserver
 from .pfc_utils import get_items_from_json
@@ -43,7 +43,9 @@ class GoalAnalyzer:
    """对话目标分析器"""

    def __init__(self, stream_id: str, private_name: str):
-        self.llm = LLMRequest(model_set=model_config.model_task_config.planner, request_type="conversation_goal")
+        self.llm = LLMServiceClient(
+            task_name="planner", request_type="conversation_goal"
+        )

        self.personality_info = self._get_personality_prompt()
        self.name = global_config.bot.nickname
@@ -157,7 +159,8 @@ class GoalAnalyzer:

        logger.debug(f"[私聊][{self.private_name}]发送到LLM的提示词: {prompt}")
        try:
-            content, _ = await self.llm.generate_response_async(prompt)
+            generation_result = await self.llm.generate_response(prompt)
+            content = generation_result.response
            logger.debug(f"[私聊][{self.private_name}]LLM原始返回内容: {content}")
        except Exception as e:
            logger.error(f"[私聊][{self.private_name}]分析对话目标时出错: {str(e)}")
@@ -271,7 +274,8 @@ class GoalAnalyzer:
 }}"""

        try:
-            content, _ = await self.llm.generate_response_async(prompt)
+            generation_result = await self.llm.generate_response(prompt)
+            content = generation_result.response
            logger.debug(f"[私聊][{self.private_name}]LLM原始返回内容: {content}")

            # 尝试解析JSON
--- a/src/chat/brain_chat/PFC/pfc_KnowledgeFetcher.py
+++ b/src/chat/brain_chat/PFC/pfc_KnowledgeFetcher.py
@@ -3,8 +3,7 @@ from src.common.logger import get_logger

 # NOTE: HippocampusManager doesn't exist in v0.12.2 - memory system was redesigned
 # from src.plugins.memory_system.Hippocampus import HippocampusManager
-from src.llm_models.utils_model import LLMRequest
-from src.config.config import model_config
+from src.services.llm_service import LLMServiceClient
 from src.chat.knowledge import qa_manager

 logger = get_logger("knowledge_fetcher")
@@ -14,7 +13,7 @@ class KnowledgeFetcher:
    """知识调取器"""

    def __init__(self, private_name: str):
-        self.llm = LLMRequest(model_set=model_config.model_task_config.utils)
+        self.llm = LLMServiceClient(task_name="utils")
        self.private_name = private_name

    def _lpmm_get_knowledge(self, query: str) -> str:
--- a/src/chat/brain_chat/PFC/reply_checker.py
+++ b/src/chat/brain_chat/PFC/reply_checker.py
@@ -2,8 +2,8 @@ import json
 import random
 from typing import Tuple, List, Dict, Any
 from src.common.logger import get_logger
-from src.llm_models.utils_model import LLMRequest
-from src.config.config import global_config, model_config
+from src.services.llm_service import LLMServiceClient
+from src.config.config import global_config
 from .chat_observer import ChatObserver
 from maim_message import UserInfo

@@ -14,7 +14,7 @@ class ReplyChecker:
    """回复检查器"""

    def __init__(self, stream_id: str, private_name: str):
-        self.llm = LLMRequest(model_set=model_config.model_task_config.utils, request_type="reply_check")
+        self.llm = LLMServiceClient(task_name="utils", request_type="reply_check")
        self.personality_info = self._get_personality_prompt()
        self.name = global_config.bot.nickname
        self.private_name = private_name
@@ -137,7 +137,8 @@ class ReplyChecker:
 注意：请严格按照JSON格式输出，不要包含任何其他内容。"""

        try:
-            content, _ = await self.llm.generate_response_async(prompt)
+            generation_result = await self.llm.generate_response(prompt)
+            content = generation_result.response
            logger.debug(f"[私聊][{self.private_name}]检查回复的原始返回: {content}")

            # 清理内容，尝试提取JSON部分
--- a/src/chat/brain_chat/PFC/reply_generator.py
+++ b/src/chat/brain_chat/PFC/reply_generator.py
@@ -1,7 +1,7 @@
 from typing import Tuple, List, Dict, Any
 from src.common.logger import get_logger
-from src.llm_models.utils_model import LLMRequest
-from src.config.config import global_config, model_config
+from src.services.llm_service import LLMServiceClient
+from src.config.config import global_config
 import random
 from .chat_observer import ChatObserver
 from .reply_checker import ReplyChecker
@@ -87,8 +87,8 @@ class ReplyGenerator:
    """回复生成器"""

    def __init__(self, stream_id: str, private_name: str):
-        self.llm = LLMRequest(
-            model_set=model_config.model_task_config.replyer,
+        self.llm = LLMServiceClient(
+            task_name="replyer",
            request_type="reply_generation",
        )
        self.personality_info = self._get_personality_prompt()
@@ -223,7 +223,8 @@ class ReplyGenerator:
        # --- 调用 LLM 生成 ---
        logger.debug(f"[私聊][{self.private_name}]发送到LLM的生成提示词:\n------\n{prompt}\n------")
        try:
-            content, _ = await self.llm.generate_response_async(prompt)
+            generation_result = await self.llm.generate_response(prompt)
+            content = generation_result.response
            logger.debug(f"[私聊][{self.private_name}]生成的回复: {content}")
            # 移除旧的检查新消息逻辑，这应该由 conversation 控制流处理
            return content
--- a/src/chat/brain_chat/brain_planner.py
+++ b/src/chat/brain_chat/brain_planner.py
@@ -17,9 +17,9 @@ from src.chat.utils.utils import get_chat_type_and_target_info
 from src.common.data_models.info_data_model import ActionPlannerInfo
 from src.common.logger import get_logger
 from src.common.utils.utils_action import ActionUtils
-from src.config.config import global_config, model_config
+from src.config.config import global_config
 from src.core.types import ActionActivationType, ActionInfo, ComponentType
-from src.llm_models.utils_model import LLMRequest
+from src.services.llm_service import LLMServiceClient
 from src.plugin_runtime.component_query import component_query_service
 from src.prompt.prompt_manager import prompt_manager
 from src.services.message_service import (
@@ -43,8 +43,8 @@ class BrainPlanner:
        self.log_prefix = f"[{_chat_manager.get_session_name(chat_id) or chat_id}]"
        self.action_manager = action_manager
        # LLM规划器配置
-        self.planner_llm = LLMRequest(
-            model_set=model_config.model_task_config.planner, request_type="planner"
+        self.planner_llm = LLMServiceClient(
+            task_name="planner", request_type="planner"
        )  # 用于动作规划

        self.last_obs_time_mark = 0.0
@@ -412,7 +412,9 @@ class BrainPlanner:
        try:
            # 调用LLM
            llm_start = time.perf_counter()
-            llm_content, (reasoning_content, _, _) = await self.planner_llm.generate_response_async(prompt=prompt)
+            generation_result = await self.planner_llm.generate_response(prompt=prompt)
+            llm_content = generation_result.response
+            reasoning_content = generation_result.reasoning
            llm_duration_ms = (time.perf_counter() - llm_start) * 1000
            llm_reasoning = reasoning_content