AstrBot 4.9.2__py3-none-any.whl → 4.10.0__py3-none-any.whl

astrbot/core/provider/manager.py

@@ -1,4 +1,5 @@
 import asyncio
+import copy
 import traceback
 from typing import Protocol, runtime_checkable
 
@@ -32,10 +33,12 @@ class ProviderManager:
         persona_mgr: PersonaManager,
     ):
         self.reload_lock = asyncio.Lock()
+        self.resource_lock = asyncio.Lock()
         self.persona_mgr = persona_mgr
         self.acm = acm
         config = acm.confs["default"]
         self.providers_config: list = config["provider"]
+        self.provider_sources_config: list = config.get("provider_sources", [])
         self.provider_settings: dict = config["provider_settings"]
         self.provider_stt_settings: dict = config.get("provider_stt_settings", {})
         self.provider_tts_settings: dict = config.get("provider_tts_settings", {})
@@ -148,6 +151,7 @@ class ProviderManager:
 
         """
         provider = None
+        provider_id = None
         if umo:
             provider_id = sp.get(
                 f"provider_perf_{provider_type.value}",
@@ -185,6 +189,12 @@ class ProviderManager:
                     )
             else:
                 raise ValueError(f"Unknown provider type: {provider_type}")
+
+        if not provider and provider_id:
+            logger.warning(
+                f"没有找到 ID 为 {provider_id} 的提供商，这可能是由于您修改了提供商（模型）ID 导致的。"
+            )
+
         return provider
 
     async def initialize(self):
@@ -251,7 +261,136 @@ class ProviderManager:
         # 初始化 MCP Client 连接
         asyncio.create_task(self.llm_tools.init_mcp_clients(), name="init_mcp_clients")
 
+    def dynamic_import_provider(self, type: str):
+        """动态导入提供商适配器模块
+
+        Args:
+            type (str): 提供商请求类型。
+
+        Raises:
+            ImportError: 如果提供商类型未知或无法导入对应模块，则抛出异常。
+        """
+        match type:
+            case "openai_chat_completion":
+                from .sources.openai_source import (
+                    ProviderOpenAIOfficial as ProviderOpenAIOfficial,
+                )
+            case "zhipu_chat_completion":
+                from .sources.zhipu_source import ProviderZhipu as ProviderZhipu
+            case "groq_chat_completion":
+                from .sources.groq_source import ProviderGroq as ProviderGroq
+            case "anthropic_chat_completion":
+                from .sources.anthropic_source import (
+                    ProviderAnthropic as ProviderAnthropic,
+                )
+            case "googlegenai_chat_completion":
+                from .sources.gemini_source import (
+                    ProviderGoogleGenAI as ProviderGoogleGenAI,
+                )
+            case "sensevoice_stt_selfhost":
+                from .sources.sensevoice_selfhosted_source import (
+                    ProviderSenseVoiceSTTSelfHost as ProviderSenseVoiceSTTSelfHost,
+                )
+            case "openai_whisper_api":
+                from .sources.whisper_api_source import (
+                    ProviderOpenAIWhisperAPI as ProviderOpenAIWhisperAPI,
+                )
+            case "openai_whisper_selfhost":
+                from .sources.whisper_selfhosted_source import (
+                    ProviderOpenAIWhisperSelfHost as ProviderOpenAIWhisperSelfHost,
+                )
+            case "xinference_stt":
+                from .sources.xinference_stt_provider import (
+                    ProviderXinferenceSTT as ProviderXinferenceSTT,
+                )
+            case "openai_tts_api":
+                from .sources.openai_tts_api_source import (
+                    ProviderOpenAITTSAPI as ProviderOpenAITTSAPI,
+                )
+            case "edge_tts":
+                from .sources.edge_tts_source import (
+                    ProviderEdgeTTS as ProviderEdgeTTS,
+                )
+            case "gsv_tts_selfhost":
+                from .sources.gsv_selfhosted_source import (
+                    ProviderGSVTTS as ProviderGSVTTS,
+                )
+            case "gsvi_tts_api":
+                from .sources.gsvi_tts_source import (
+                    ProviderGSVITTS as ProviderGSVITTS,
+                )
+            case "fishaudio_tts_api":
+                from .sources.fishaudio_tts_api_source import (
+                    ProviderFishAudioTTSAPI as ProviderFishAudioTTSAPI,
+                )
+            case "dashscope_tts":
+                from .sources.dashscope_tts import (
+                    ProviderDashscopeTTSAPI as ProviderDashscopeTTSAPI,
+                )
+            case "azure_tts":
+                from .sources.azure_tts_source import (
+                    AzureTTSProvider as AzureTTSProvider,
+                )
+            case "minimax_tts_api":
+                from .sources.minimax_tts_api_source import (
+                    ProviderMiniMaxTTSAPI as ProviderMiniMaxTTSAPI,
+                )
+            case "volcengine_tts":
+                from .sources.volcengine_tts import (
+                    ProviderVolcengineTTS as ProviderVolcengineTTS,
+                )
+            case "gemini_tts":
+                from .sources.gemini_tts_source import (
+                    ProviderGeminiTTSAPI as ProviderGeminiTTSAPI,
+                )
+            case "openai_embedding":
+                from .sources.openai_embedding_source import (
+                    OpenAIEmbeddingProvider as OpenAIEmbeddingProvider,
+                )
+            case "gemini_embedding":
+                from .sources.gemini_embedding_source import (
+                    GeminiEmbeddingProvider as GeminiEmbeddingProvider,
+                )
+            case "vllm_rerank":
+                from .sources.vllm_rerank_source import (
+                    VLLMRerankProvider as VLLMRerankProvider,
+                )
+            case "xinference_rerank":
+                from .sources.xinference_rerank_source import (
+                    XinferenceRerankProvider as XinferenceRerankProvider,
+                )
+            case "bailian_rerank":
+                from .sources.bailian_rerank_source import (
+                    BailianRerankProvider as BailianRerankProvider,
+                )
+
+    def get_merged_provider_config(self, provider_config: dict) -> dict:
+        """获取 provider 配置和 provider_source 配置合并后的结果
+
+        Returns:
+            dict: 合并后的 provider 配置，key 为 provider id，value 为合并后的配置字典
+        """
+        pc = copy.deepcopy(provider_config)
+        provider_source_id = pc.get("provider_source_id", "")
+        if provider_source_id:
+            provider_source = None
+            for ps in self.provider_sources_config:
+                if ps.get("id") == provider_source_id:
+                    provider_source = ps
+                    break
+
+            if provider_source:
+                # 合并配置，provider 的配置优先级更高
+                merged_config = {**provider_source, **pc}
+                # 保持 id 为 provider 的 id，而不是 source 的 id
+                merged_config["id"] = pc["id"]
+                pc = merged_config
+        return pc
+
     async def load_provider(self, provider_config: dict):
+        # 如果 provider_source_id 存在且不为空，则从 provider_sources 中找到对应的配置并合并
+        provider_config = self.get_merged_provider_config(provider_config)
+
         if not provider_config["enable"]:
             logger.info(f"Provider {provider_config['id']} is disabled, skipping")
             return
@@ -264,99 +403,7 @@ class ProviderManager:
 
         # 动态导入
         try:
-            match provider_config["type"]:
-                case "openai_chat_completion":
-                    from .sources.openai_source import (
-                        ProviderOpenAIOfficial as ProviderOpenAIOfficial,
-                    )
-                case "zhipu_chat_completion":
-                    from .sources.zhipu_source import ProviderZhipu as ProviderZhipu
-                case "groq_chat_completion":
-                    from .sources.groq_source import ProviderGroq as ProviderGroq
-                case "anthropic_chat_completion":
-                    from .sources.anthropic_source import (
-                        ProviderAnthropic as ProviderAnthropic,
-                    )
-                case "googlegenai_chat_completion":
-                    from .sources.gemini_source import (
-                        ProviderGoogleGenAI as ProviderGoogleGenAI,
-                    )
-                case "sensevoice_stt_selfhost":
-                    from .sources.sensevoice_selfhosted_source import (
-                        ProviderSenseVoiceSTTSelfHost as ProviderSenseVoiceSTTSelfHost,
-                    )
-                case "openai_whisper_api":
-                    from .sources.whisper_api_source import (
-                        ProviderOpenAIWhisperAPI as ProviderOpenAIWhisperAPI,
-                    )
-                case "openai_whisper_selfhost":
-                    from .sources.whisper_selfhosted_source import (
-                        ProviderOpenAIWhisperSelfHost as ProviderOpenAIWhisperSelfHost,
-                    )
-                case "xinference_stt":
-                    from .sources.xinference_stt_provider import (
-                        ProviderXinferenceSTT as ProviderXinferenceSTT,
-                    )
-                case "openai_tts_api":
-                    from .sources.openai_tts_api_source import (
-                        ProviderOpenAITTSAPI as ProviderOpenAITTSAPI,
-                    )
-                case "edge_tts":
-                    from .sources.edge_tts_source import (
-                        ProviderEdgeTTS as ProviderEdgeTTS,
-                    )
-                case "gsv_tts_selfhost":
-                    from .sources.gsv_selfhosted_source import (
-                        ProviderGSVTTS as ProviderGSVTTS,
-                    )
-                case "gsvi_tts_api":
-                    from .sources.gsvi_tts_source import (
-                        ProviderGSVITTS as ProviderGSVITTS,
-                    )
-                case "fishaudio_tts_api":
-                    from .sources.fishaudio_tts_api_source import (
-                        ProviderFishAudioTTSAPI as ProviderFishAudioTTSAPI,
-                    )
-                case "dashscope_tts":
-                    from .sources.dashscope_tts import (
-                        ProviderDashscopeTTSAPI as ProviderDashscopeTTSAPI,
-                    )
-                case "azure_tts":
-                    from .sources.azure_tts_source import (
-                        AzureTTSProvider as AzureTTSProvider,
-                    )
-                case "minimax_tts_api":
-                    from .sources.minimax_tts_api_source import (
-                        ProviderMiniMaxTTSAPI as ProviderMiniMaxTTSAPI,
-                    )
-                case "volcengine_tts":
-                    from .sources.volcengine_tts import (
-                        ProviderVolcengineTTS as ProviderVolcengineTTS,
-                    )
-                case "gemini_tts":
-                    from .sources.gemini_tts_source import (
-                        ProviderGeminiTTSAPI as ProviderGeminiTTSAPI,
-                    )
-                case "openai_embedding":
-                    from .sources.openai_embedding_source import (
-                        OpenAIEmbeddingProvider as OpenAIEmbeddingProvider,
-                    )
-                case "gemini_embedding":
-                    from .sources.gemini_embedding_source import (
-                        GeminiEmbeddingProvider as GeminiEmbeddingProvider,
-                    )
-                case "vllm_rerank":
-                    from .sources.vllm_rerank_source import (
-                        VLLMRerankProvider as VLLMRerankProvider,
-                    )
-                case "xinference_rerank":
-                    from .sources.xinference_rerank_source import (
-                        XinferenceRerankProvider as XinferenceRerankProvider,
-                    )
-                case "bailian_rerank":
-                    from .sources.bailian_rerank_source import (
-                        BailianRerankProvider as BailianRerankProvider,
-                    )
+            self.dynamic_import_provider(provider_config["type"])
         except (ImportError, ModuleNotFoundError) as e:
             logger.critical(
                 f"加载 {provider_config['type']}({provider_config['id']}) 提供商适配器失败：{e}。可能是因为有未安装的依赖。",
@@ -499,6 +546,7 @@ class ProviderManager:
 
             # 和配置文件保持同步
             self.providers_config = astrbot_config["provider"]
+            self.provider_sources_config = astrbot_config.get("provider_sources", [])
             config_ids = [provider["id"] for provider in self.providers_config]
             logger.info(f"providers in user's config: {config_ids}")
             for key in list(self.inst_map.keys()):
@@ -570,6 +618,68 @@ class ProviderManager:
             )
             del self.inst_map[provider_id]
 
+    async def delete_provider(
+        self, provider_id: str | None = None, provider_source_id: str | None = None
+    ):
+        """Delete provider and/or provider source from config and terminate the instances. Config will be saved after deletion."""
+        async with self.resource_lock:
+            # delete from config
+            target_prov_ids = []
+            if provider_id:
+                target_prov_ids.append(provider_id)
+            else:
+                for prov in self.providers_config:
+                    if prov.get("provider_source_id") == provider_source_id:
+                        target_prov_ids.append(prov.get("id"))
+            config = self.acm.default_conf
+            for tpid in target_prov_ids:
+                await self.terminate_provider(tpid)
+                config["provider"] = [
+                    prov for prov in config["provider"] if prov.get("id") != tpid
+                ]
+            config.save_config()
+            logger.info(f"Provider {target_prov_ids} 已从配置中删除。")
+
+    async def update_provider(self, origin_provider_id: str, new_config: dict):
+        """Update provider config and reload the instance. Config will be saved after update."""
+        async with self.resource_lock:
+            npid = new_config.get("id", None)
+            if not npid:
+                raise ValueError("New provider config must have an 'id' field")
+            config = self.acm.default_conf
+            for provider in config["provider"]:
+                if (
+                    provider.get("id", None) == npid
+                    and provider.get("id", None) != origin_provider_id
+                ):
+                    raise ValueError(f"Provider ID {npid} already exists")
+            # update config
+            for idx, provider in enumerate(config["provider"]):
+                if provider.get("id", None) == origin_provider_id:
+                    config["provider"][idx] = new_config
+                    break
+            else:
+                raise ValueError(f"Provider ID {origin_provider_id} not found")
+            config.save_config()
+            # reload instance
+            await self.reload(new_config)
+
+    async def create_provider(self, new_config: dict):
+        """Add new provider config and load the instance. Config will be saved after addition."""
+        async with self.resource_lock:
+            npid = new_config.get("id", None)
+            if not npid:
+                raise ValueError("New provider config must have an 'id' field")
+            config = self.acm.default_conf
+            for provider in config["provider"]:
+                if provider.get("id", None) == npid:
+                    raise ValueError(f"Provider ID {npid} already exists")
+            # add to config
+            config["provider"].append(new_config)
+            config.save_config()
+            # load instance
+            await self.load_provider(new_config)
+
     async def terminate(self):
         for provider_inst in self.provider_insts:
             if hasattr(provider_inst, "terminate"):

astrbot/core/provider/sources/anthropic_source.py

@@ -6,10 +6,12 @@ from mimetypes import guess_type
 import anthropic
 from anthropic import AsyncAnthropic
 from anthropic.types import Message
+from anthropic.types.message_delta_usage import MessageDeltaUsage
+from anthropic.types.usage import Usage
 
 from astrbot import logger
 from astrbot.api.provider import Provider
-from astrbot.core.provider.entities import LLMResponse
+from astrbot.core.provider.entities import LLMResponse, TokenUsage
 from astrbot.core.provider.func_tool_manager import ToolSet
 from astrbot.core.utils.io import download_image_by_url
 
@@ -45,7 +47,7 @@ class ProviderAnthropic(Provider):
             base_url=self.base_url,
         )
 
-        self.set_model(provider_config["model_config"]["model"])
+        self.set_model(provider_config.get("model", "unknown"))
 
     def _prepare_payload(self, messages: list[dict]):
         """准备 Anthropic API 的请求 payload
@@ -107,12 +109,32 @@ class ProviderAnthropic(Provider):
 
         return system_prompt, new_messages
 
+    def _extract_usage(self, usage: Usage) -> TokenUsage:
+        # https://docs.claude.com/en/docs/build-with-claude/prompt-caching#tracking-cache-performance
+        return TokenUsage(
+            input_other=usage.input_tokens or 0,
+            input_cached=usage.cache_read_input_tokens or 0,
+            output=usage.output_tokens,
+        )
+
+    def _update_usage(self, token_usage: TokenUsage, usage: MessageDeltaUsage) -> None:
+        if usage.input_tokens is not None:
+            token_usage.input_other = usage.input_tokens
+        if usage.cache_read_input_tokens is not None:
+            token_usage.input_cached = usage.cache_read_input_tokens
+        if usage.output_tokens is not None:
+            token_usage.output = usage.output_tokens
+
     async def _query(self, payloads: dict, tools: ToolSet | None) -> LLMResponse:
         if tools:
             if tool_list := tools.get_func_desc_anthropic_style():
                 payloads["tools"] = tool_list
 
-        completion = await self.client.messages.create(**payloads, stream=False)
+        extra_body = self.provider_config.get("custom_extra_body", {})
+
+        completion = await self.client.messages.create(
+            **payloads, stream=False, extra_body=extra_body
+        )
 
         assert isinstance(completion, Message)
         logger.debug(f"completion: {completion}")
@@ -131,6 +153,10 @@ class ProviderAnthropic(Provider):
                 llm_response.tools_call_args.append(content_block.input)
                 llm_response.tools_call_name.append(content_block.name)
                 llm_response.tools_call_ids.append(content_block.id)
+
+        llm_response.id = completion.id
+        llm_response.usage = self._extract_usage(completion.usage)
+
         # TODO(Soulter): 处理 end_turn 情况
         if not llm_response.completion_text and not llm_response.tools_call_args:
             raise Exception(f"Anthropic API 返回的 completion 无法解析：{completion}。")
@@ -151,10 +177,19 @@ class ProviderAnthropic(Provider):
         # 用于累积最终结果
         final_text = ""
         final_tool_calls = []
+        id = None
+        usage = TokenUsage()
+        extra_body = self.provider_config.get("custom_extra_body", {})
 
-        async with self.client.messages.stream(**payloads) as stream:
+        async with self.client.messages.stream(
+            **payloads, extra_body=extra_body
+        ) as stream:
             assert isinstance(stream, anthropic.AsyncMessageStream)
             async for event in stream:
+                if event.type == "message_start":
+                    # the usage contains input token usage
+                    id = event.message.id
+                    usage = self._extract_usage(event.message.usage)
                 if event.type == "content_block_start":
                     if event.content_block.type == "text":
                         # 文本块开始
@@ -162,6 +197,8 @@ class ProviderAnthropic(Provider):
                             role="assistant",
                             completion_text="",
                             is_chunk=True,
+                            usage=usage,
+                            id=id,
                         )
                     elif event.content_block.type == "tool_use":
                         # 工具使用块开始，初始化缓冲区
@@ -179,6 +216,8 @@ class ProviderAnthropic(Provider):
                             role="assistant",
                             completion_text=event.delta.text,
                             is_chunk=True,
+                            usage=usage,
+                            id=id,
                         )
                     elif event.delta.type == "input_json_delta":
                         # 工具调用参数增量
@@ -215,6 +254,8 @@ class ProviderAnthropic(Provider):
                                 tools_call_name=[tool_info["name"]],
                                 tools_call_ids=[tool_info["id"]],
                                 is_chunk=True,
+                                usage=usage,
+                                id=id,
                             )
                         except json.JSONDecodeError:
                             # JSON 解析失败，跳过这个工具调用
@@ -223,11 +264,17 @@ class ProviderAnthropic(Provider):
                         # 清理缓冲区
                         del tool_use_buffer[event.index]
 
+                elif event.type == "message_delta":
+                    if event.usage:
+                        self._update_usage(usage, event.usage)
+
         # 返回最终的完整结果
         final_response = LLMResponse(
             role="assistant",
             completion_text=final_text,
             is_chunk=False,
+            usage=usage,
+            id=id,
         )
 
         if final_tool_calls:
@@ -277,10 +324,9 @@ class ProviderAnthropic(Provider):
 
         system_prompt, new_messages = self._prepare_payload(context_query)
 
-        model_config = self.provider_config.get("model_config", {})
-        model_config["model"] = model or self.get_model()
+        model = model or self.get_model()
 
-        payloads = {"messages": new_messages, **model_config}
+        payloads = {"messages": new_messages, "model": model}
 
         # Anthropic has a different way of handling system prompts
         if system_prompt:
@@ -290,7 +336,6 @@ class ProviderAnthropic(Provider):
         try:
             llm_response = await self._query(payloads, func_tool)
         except Exception as e:
-            # logger.error(f"发生了错误。Provider 配置如下: {model_config}")
             raise e
 
         return llm_response
@@ -332,10 +377,9 @@ class ProviderAnthropic(Provider):
 
         system_prompt, new_messages = self._prepare_payload(context_query)
 
-        model_config = self.provider_config.get("model_config", {})
-        model_config["model"] = model or self.get_model()
+        model = model or self.get_model()
 
-        payloads = {"messages": new_messages, **model_config}
+        payloads = {"messages": new_messages, "model": model}
 
         # Anthropic has a different way of handling system prompts
         if system_prompt: