feat(llm): LLM module protocol optimization

openjiuwen/core/foundation/llm/__init__.py

@@ -7,7 +7,15 @@
 from openjiuwen.core.foundation.llm.output_parsers.output_parser import BaseOutputParser
 
 # Configuration
-from openjiuwen.core.foundation.llm.schema.config import ModelRequestConfig, ModelClientConfig, ProviderType
+from openjiuwen.core.foundation.llm.schema.config import (
+    LLMApiMode,
+    LLMAuthMode,
+    LLMExtensionsConfig,
+    KVCacheExtensionConfig,
+    ModelRequestConfig,
+    ModelClientConfig,
+    ProviderType,
+)
 from openjiuwen.core.foundation.llm.schema.mode_info import BaseModelInfo, ModelConfig
 # Messages
 from openjiuwen.core.foundation.llm.schema.message import (
@@ -26,8 +34,8 @@
 from openjiuwen.core.foundation.llm.schema.tool_call import ToolCall
 
 # Built-in implementations
+from openjiuwen.core.foundation.llm.model_clients.anthropic_model_client import AnthropicModelClient
 from openjiuwen.core.foundation.llm.model_clients.openai_model_client import OpenAIModelClient
-from openjiuwen.core.foundation.llm.model_clients.ascend_affinity_model_client import AscendAffinityModelClient
 from openjiuwen.core.foundation.llm.output_parsers.json_output_parser import JsonOutputParser
 from openjiuwen.core.foundation.llm.output_parsers.markdown_output_parser import MarkdownOutputParser
 
@@ -47,6 +55,10 @@
     "ModelRequestConfig",
     "ModelClientConfig",
     "ProviderType",
+    "LLMApiMode",
+    "LLMAuthMode",
+    "LLMExtensionsConfig",
+    "KVCacheExtensionConfig",
     "BaseModelInfo",
     "ModelConfig"
 ]
@@ -73,8 +85,8 @@
 
 # Built-in ModelClient implementations
 _PREBUILT_MODEL_CLIENTS = [
+    "AnthropicModelClient",
     "OpenAIModelClient",
-    "AscendAffinityModelClient",
 ]
 
 # Built-in OutputParser implementations

openjiuwen/core/foundation/llm/inference_affinity_model.py

@@ -9,14 +9,14 @@
 from openjiuwen.core.foundation.tool import ToolInfo
 from openjiuwen.core.foundation.llm.schema.config import ModelRequestConfig, ModelClientConfig
 from openjiuwen.core.foundation.llm.output_parsers.output_parser import BaseOutputParser
-from openjiuwen.core.foundation.llm.model_clients.inference_affinity_model_client import InferenceAffinityModelClient
+from openjiuwen.core.foundation.llm.model_clients import create_model_client
 
 
 class InferenceAffinityModel:
     """InferenceAffinity (vLLM) model unified invocation entry point
 
     Responsibilities:
-    1. Manage InferenceAffinityModelClient instances
+    1. Manage OpenAI-compatible clients configured for KV cache release
     2. Provide unified asynchronous interfaces (ainvoke, astream)
     3. Support release functionality
 
@@ -38,10 +38,10 @@ def __init__(
         """
         self.model_config = model_config
         self.model_client_config = model_client_config
-        self._client: Optional[InferenceAffinityModelClient] = None
+        self._client = None
 
         if model_client_config is not None:
-            self._client = InferenceAffinityModelClient(model_config, model_client_config)
+            self._client = create_model_client(model_client_config, model_config)
         else:
             raise build_error(StatusCode.MODEL_SERVICE_CONFIG_ERROR,
                               error_msg="model client config is none")
@@ -187,4 +187,4 @@ def build_kv_cache_invoke_kwargs(
             extra["session_id"] = session.get_session_id()
         if enable_kv_cache_release:
             extra["enable_cache_sharing"] = True
-        return extra
+        return extra

openjiuwen/core/foundation/llm/model.py

@@ -20,7 +20,6 @@
     VideoGenerationResponse
 )
 from openjiuwen.core.foundation.llm.model_clients.base_model_client import BaseModelClient
-from openjiuwen.core.foundation.llm.model_clients.inference_affinity_model_client import InferenceAffinityModelClient
 from openjiuwen.core.runner.callback import trigger
 
 
@@ -303,15 +302,13 @@ def build_kv_cache_invoke_kwargs(
           - session_id: use session.get_session_id() if provided
           - enable_cache_sharing: follow enable_kv_cache_release
         """
-        if not isinstance(self._client, InferenceAffinityModelClient):
+        build_fn = getattr(self._client, "build_kv_cache_invoke_kwargs", None)
+        if not callable(build_fn):
             return {}
-
-        extra: dict = {}
-        if session is not None and hasattr(session, "get_session_id"):
-            extra["session_id"] = session.get_session_id()
-        if enable_kv_cache_release:
-            extra["enable_cache_sharing"] = True
-        return extra
+        return build_fn(
+            session=session,
+            enable_kv_cache_release=enable_kv_cache_release,
+        )
 
     def build_kv_cache_affinity_invoke_kwargs(
             self,

openjiuwen/core/foundation/llm/model_clients/__init__.py

@@ -4,7 +4,28 @@
 from openjiuwen.core.common.exception.codes import StatusCode
 from openjiuwen.core.common.exception.errors import build_error
 from openjiuwen.core.foundation.llm.model_clients.base_model_client import BaseModelClient
-from openjiuwen.core.foundation.llm.schema.config import ModelRequestConfig, ModelClientConfig, ProviderType
+from openjiuwen.core.foundation.llm.schema.config import (
+    LLMAuthMode,
+    ModelRequestConfig,
+    ModelClientConfig,
+    ProviderType,
+)
+from openjiuwen.core.foundation.llm.utils.endpoint_profiles import normalize_model_client_config
+
+
+def _value(value):
+    return value.value if hasattr(value, "value") else value
+
+
+def _implementation_provider(client_config: ModelClientConfig) -> str:
+    provider = _value(client_config.client_provider)
+    if provider != ProviderType.OpenAI.value:
+        return provider
+
+    auth_mode = _value(getattr(client_config, "auth_mode", LLMAuthMode.ApiKey.value))
+    if auth_mode == LLMAuthMode.OpenAIAccountOAuth.value:
+        return ProviderType.OpenAIAccount.value
+    return ProviderType.OpenAI.value
 
 
 def _builtin_model_client(provider, client_config: ModelClientConfig, model_config: ModelRequestConfig):
@@ -18,39 +39,10 @@ def _builtin_model_client(provider, client_config: ModelClientConfig, model_conf
         from openjiuwen.core.foundation.llm.model_clients.openai_account_model_client import OpenAIAccountModelClient
         return OpenAIAccountModelClient(model_config=model_config, model_client_config=client_config)
 
-    if provider == ProviderType.OpenRouter.value:
-        from openjiuwen.core.foundation.llm.model_clients.openrouter_model_client import OpenRouterModelClient
-        return OpenRouterModelClient(model_config=model_config, model_client_config=client_config)
-
     if provider == ProviderType.Anthropic.value:
         from openjiuwen.core.foundation.llm.model_clients.anthropic_model_client import AnthropicModelClient
         return AnthropicModelClient(model_config=model_config, model_client_config=client_config)
 
-    if provider == ProviderType.SiliconFlow.value:
-        from openjiuwen.core.foundation.llm.model_clients.siliconflow_model_client import \
-            SiliconFlowModelClient
-        return SiliconFlowModelClient(model_config=model_config, model_client_config=client_config)
-
-    if provider == ProviderType.DashScope.value:
-        from openjiuwen.core.foundation.llm.model_clients.dashscope_model_client import \
-            DashScopeModelClient
-        return DashScopeModelClient(model_config=model_config, model_client_config=client_config)
-
-    if provider == ProviderType.InferenceAffinity.value:
-        from openjiuwen.core.foundation.llm.model_clients.inference_affinity_model_client import \
-            InferenceAffinityModelClient
-        return InferenceAffinityModelClient(model_config=model_config, model_client_config=client_config)
-
-    if provider == ProviderType.AscendAffinity.value:
-        from openjiuwen.core.foundation.llm.model_clients.ascend_affinity_model_client import \
-            AscendAffinityModelClient
-        return AscendAffinityModelClient(model_config=model_config, model_client_config=client_config)
-
-    if provider == ProviderType.DeepSeek.value:
-        from openjiuwen.core.foundation.llm.model_clients.deepseek_model_client import \
-            DeepSeekModelClient
-        return DeepSeekModelClient(model_config=model_config, model_client_config=client_config)
-
     if provider == ProviderType.IntelliRouter.value:
         from openjiuwen.core.foundation.llm.model_clients.intelli_router_model_client import \
             IntelliRouterModelClient
@@ -78,7 +70,10 @@ def create_model_client(client_config: ModelClientConfig, model_config: ModelReq
                           error_msg="model client config client_id is none")
     provider = client_config.client_provider.value if isinstance(client_config.client_provider, ProviderType)\
         else client_config.client_provider
-    client = _builtin_model_client(provider, client_config, model_config)
+    normalized_config = normalize_model_client_config(client_config)
+    dispatch_provider = _implementation_provider(normalized_config)
+    dispatch_config = normalized_config
+    client = _builtin_model_client(dispatch_provider, dispatch_config, model_config)
     if client is not None:
         return client
     try: