Merge remote-tracking branch 'origin/gcr-custom' into xuhan/gcr-custom

source/infrastructure/lib/chat/chat-stack.ts

@@ -144,7 +144,7 @@ export class ChatStack extends NestedStack implements ChatStackOutputs {
         BEDROCK_AWS_SECRET_ACCESS_KEY: props.config.chat.bedrockSk || ""
       },
       // layers: [apiLambdaOnlineSourceLayer, modelLayer],
-      layers: [apiLambdaOnlineSourceLayer],
+      layers: [apiLambdaOnlineSourceLayer]
     });
     this.lambdaOnlineMain = lambdaOnlineMain.function;
 

source/infrastructure/lib/ui/ui-stack.ts

@@ -17,7 +17,7 @@ import { join } from "path";
 
 import { SystemConfig } from "../shared/types";
 import { PortalConstruct } from "../ui/ui-portal";
-import { UserConstruct } from "../user/user-construct";
+
 
 
 
@@ -28,14 +28,12 @@ interface UIStackProps extends StackProps {
 export interface UIStackOutputs {
   readonly mainPortalConstruct: PortalConstruct;
   readonly clientPortalConstruct: PortalConstruct;
-  readonly userConstruct: UserConstruct;
 }
 
 export class UIStack extends Stack implements UIStackOutputs {
 
   public mainPortalConstruct: PortalConstruct;
   public clientPortalConstruct: PortalConstruct;
-  public userConstruct: UserConstruct;
 
   constructor(scope: Construct, id: string, props: UIStackProps) {
     super(scope, id, props);
@@ -48,44 +46,8 @@ export class UIStack extends Stack implements UIStackOutputs {
       responseHeadersPolicyName: `SecHdr${Aws.REGION}${Aws.STACK_NAME}-client`
     });
 
-    const userConstruct = new UserConstruct(this, "User", {
-      deployRegion: props.config.deployRegion,
-      adminEmail: props.config.email,
-      callbackUrls: [
-        `https://${clientPortalConstruct.portalUrl}/signin`,
-        `https://${mainPortalConstruct.portalUrl}/signin`
-      ],
-      logoutUrls: [
-        `https://${clientPortalConstruct.portalUrl}`,
-        `https://${mainPortalConstruct.portalUrl}`
-      ],
-      // userPoolName: `${Constants.SOLUTION_NAME}-workspace_UserPool`,
-      // domainPrefix: `${Constants.SOLUTION_NAME.toLowerCase()}-workspace-${Aws.ACCOUNT_ID}`,
-    });
     this.mainPortalConstruct = mainPortalConstruct;
     this.clientPortalConstruct = clientPortalConstruct;
-    this.userConstruct = userConstruct;
-
-    // Add CfnOutputs to export values
-    new CfnOutput(this, 'UserPoolId', {
-      value: userConstruct.userPoolId,
-      exportName: `${id}-user-pool-id`
-    });
-
-    new CfnOutput(this, 'OidcClientId', {
-      value: userConstruct.oidcClientId,
-      exportName: `${id}-oidc-client-id`
-    });
-
-    new CfnOutput(this, 'OidcIssuer', {
-      value: userConstruct.oidcIssuer,
-      exportName: `${id}-oidc-issuer`
-    });
-
-    new CfnOutput(this, 'OidcLogoutUrl', {
-      value: userConstruct.oidcLogoutUrl,
-      exportName: `${id}-oidc-logout-url`
-    });
 
     new CfnOutput(this, 'PortalBucketName', {
       value: mainPortalConstruct.portalBucket.bucketName,

source/lambda/online/common_logic/common_utils/constant.py

@@ -133,6 +133,8 @@ class ModelProvider(ConstantBase):
     BRCONNECTOR_BEDROCK = "Bedrock API"
     OPENAI = "OpenAI API"
     SAGEMAKER = "SageMaker"
+    SILICONFLOW = "siliconflow"
+
 
 
 class LLMModelType(ConstantBase):
@@ -161,7 +163,13 @@ class LLMModelType(ConstantBase):
     QWEN25_INSTRUCT_72B_AWQ = "Qwen2.5-72B-Instruct-AWQ"
     DEEPSEEK_R1_DISTILL_LLAMA_8B = "DeepSeek-R1-Distill-Llama-8B"
     DEEPSEEK_R1_DISTILL_LLAMA_70B = "DeepSeek-R1-Distill-Llama-70B"
+    DEEPSEEK_R1_DISTILL_QWEN_1d5B = "DeepSeek-R1-Distill-Qwen-1.5B"
+    DEEPSEEK_R1_DISTILL_QWEN_7B = "DeepSeek-R1-Distill-Qwen-7B"
+    DEEPSEEK_R1_DISTILL_QWEN_14B = "DeepSeek-R1-Distill-Qwen-14B"
     DEEPSEEK_R1_DISTILL_QWEN_32B = "DeepSeek-R1-Distill-Qwen-32B"
+    DEEPSEEK_R1 = "DeepSeek-R1"
+    DEEPSEEK_V3 = "DeepSeek-R3"
+
     QWEN15INSTRUCT32B = "qwen1_5-32B-instruct"
     LLAMA3_1_70B_INSTRUCT = "meta.llama3-1-70b-instruct-v1:0"
     LLAMA3_2_90B_INSTRUCT = "us.meta.llama3-2-90b-instruct-v1:0"

source/lambda/online/common_logic/langchain_integration/models/chat_models/sagemaker_models.py

@@ -7,7 +7,7 @@
     ModelProvider,
 )
 from dmaa.integrations.langchain_clients import (
-    SageMakerVllmChatModel as _SageMakerVllmChatModel,
+    SageMakerVllmChatModel as _SageMakerVllmChatModel
 )
 
 from . import Model

source/lambda/online/common_logic/langchain_integration/models/chat_models/siliconflow_models.py

@@ -0,0 +1,45 @@
+import os
+import boto3
+from common_logic.common_utils.constant import (
+    ModelProvider
+)
+from common_logic.common_utils.logger_utils import get_logger, llm_messages_print_decorator
+from . import Model
+from langchain_deepseek import ChatDeepSeek as _ChatDeepSeek
+from ..model_config import SILICONFLOW_DEEPSEEK_MODEL_CONFIGS
+
+logger = get_logger("siliconflow_models")
+
+class ChatDeepSeek(_ChatDeepSeek):
+    enable_any_tool_choice: bool = False
+    any_tool_choice_value: str = 'required'
+    enable_prefill: bool = False
+
+
+class DeepSeekBaseModel(Model):
+    default_model_kwargs = {"max_tokens": 2000,
+                            "temperature": 0.6, "top_p": 0.9}
+    enable_any_tool_choice: bool = False
+    enable_prefill: bool = False
+    model_provider = ModelProvider.SILICONFLOW
+
+    @classmethod
+    def create_model(cls, model_kwargs=None, **kwargs):
+        model_kwargs = model_kwargs or {}
+        model_kwargs = {**cls.default_model_kwargs, **model_kwargs}
+        # base_url = kwargs.get("base_url", None) or os.environ.get(
+        #     "OPENAI_BASE_URL", None)
+        api_key = kwargs.get('siliconflow_api_key', None) or os.environ.get(
+            "SILICONFLOW_API_KEY", None)
+        return ChatDeepSeek(
+            enable_any_tool_choice=cls.enable_any_tool_choice,
+            enable_prefill=cls.enable_prefill,
+            base_url="https://api.siliconflow.cn/v1",
+            api_key=api_key,
+            model=cls.model_id,
+            **model_kwargs
+        )
+
+
+DeepSeekBaseModel.create_for_models(SILICONFLOW_DEEPSEEK_MODEL_CONFIGS)
+

source/lambda/online/common_logic/langchain_integration/models/model_config.py

@@ -6,11 +6,13 @@
 @dataclass
 class ModelConfig:
     model_id: str
+    model: Union[str,None] = None
     default_model_kwargs: Dict[str, Any] = None
     enable_any_tool_choice: bool = True
     any_tool_choice_value: str = "any"
     enable_prefill: bool = True
 
+
     def __post_init__(self):
         if self.default_model_kwargs is None:
             self.default_model_kwargs = {}
@@ -184,6 +186,58 @@ def __post_init__(self):
     )
 ]
 
+SILICONFLOW_DEEPSEEK_MODEL_CONFIGS = [
+    ModelConfig(
+        model_id=LLMModelType.DEEPSEEK_R1_DISTILL_LLAMA_70B,
+        model="deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
+        default_model_kwargs=BASE_CONFIG.default_model_kwargs.copy(),
+        enable_any_tool_choice=False,
+        enable_prefill=False
+    ),
+    ModelConfig(
+        model_id=LLMModelType.DEEPSEEK_R1_DISTILL_LLAMA_8B,
+        model="deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
+        default_model_kwargs=BASE_CONFIG.default_model_kwargs.copy(),
+        enable_any_tool_choice=False,
+        enable_prefill=False
+    ),
+    ModelConfig(
+        model_id=LLMModelType.DEEPSEEK_R1_DISTILL_QWEN_7B,
+        model="deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
+        default_model_kwargs=BASE_CONFIG.default_model_kwargs.copy(),
+        enable_any_tool_choice=False,
+        enable_prefill=False
+    ),
+    ModelConfig(
+        model_id=LLMModelType.DEEPSEEK_R1_DISTILL_QWEN_14B,
+        model="deepseek-ai/DeepSeek-R1-Distill-Qwen-14B",
+        default_model_kwargs=BASE_CONFIG.default_model_kwargs.copy(),
+        enable_any_tool_choice=False,
+        enable_prefill=False
+    ),
+    ModelConfig(
+        model_id=LLMModelType.DEEPSEEK_R1_DISTILL_QWEN_1d5B,
+        model="deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
+        default_model_kwargs=BASE_CONFIG.default_model_kwargs.copy(),
+        enable_any_tool_choice=False,
+        enable_prefill=False
+    ),
+    ModelConfig(
+        model_id=LLMModelType.DEEPSEEK_R1_DISTILL_QWEN_32B,
+        model="deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
+        default_model_kwargs=BASE_CONFIG.default_model_kwargs.copy(),
+        enable_any_tool_choice=False,
+        enable_prefill=False
+    ),
+    ModelConfig(
+        model_id=LLMModelType.DEEPSEEK_R1,
+        model="deepseek-ai/DeepSeek-R1",
+        default_model_kwargs=BASE_CONFIG.default_model_kwargs.copy(),
+        enable_any_tool_choice=False,
+        enable_prefill=False
+    )
+]
+
 
 OPENAI_MODEL_CONFIGS = [
     ModelConfig(

source/lambda/online/common_logic/langchain_integration/tools/common_tools/rag.py

@@ -29,7 +29,7 @@ def filter_response(res: Iterable, state: dict):
     references = []
     tag_start = "<reference>"
     tag_end = "</reference>"
-
+    
     for char in res:
         char_strip = char.strip()
         if not buffer and char_strip not in ["<", "<reference"]:
@@ -138,6 +138,8 @@ def rag_tool(retriever_config: dict, query=None):
     )
     output = chain.invoke(llm_input)
 
+
+
     filtered_output = filter_response(output, state)
 
     return filtered_output, filtered_output

source/lambda/online/requirements.txt

@@ -5,4 +5,5 @@ opensearch-py==2.2.0
 requests_aws4auth==1.2.2
 pydantic==2.9.2
 datamodel-code-generator==0.26.2
-langchain-openai==0.2.6
+langchain-openai==0.3.7
+langchain-deepseek==0.1.2