chore: merge from dev

.viperlightignore

@@ -15,6 +15,7 @@ source/portal/src/utils/const.ts
 source/lambda/online/lambda_main/test/main_local_test_retail.py
 source/lambda/online/lambda_main/test/main_local_test_common.py
 source/lambda/online/functions/retail_tools/lambda_product_information_search/product_information_search.py
+source/lambda/online/common_logic/common_utils/prompt_utils.py
 source/lambda/job/test/prepare_data.py
 README.md
 README_zh-cn.md

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

@@ -146,6 +146,8 @@ class LLMModelType(ConstantBase):
     LLAMA3_2_90B_INSTRUCT = "us.meta.llama3-2-90b-instruct-v1:0"
     MISTRAL_LARGE_2407 = "mistral.mistral-large-2407-v1:0"
     COHERE_COMMAND_R_PLUS = "cohere.command-r-plus-v1:0"
+    # NOVA_PRO = "amazon.nova-pro-v1:0"
+    NOVA_PRO = "us.amazon.nova-pro-v1:0"
 
 
 class EmbeddingModelType(ConstantBase):

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

@@ -21,7 +21,8 @@
     LLMModelType.CLAUDE_3_5_SONNET_V2,
     LLMModelType.LLAMA3_1_70B_INSTRUCT,
     LLMModelType.MISTRAL_LARGE_2407,
-    LLMModelType.COHERE_COMMAND_R_PLUS
+    LLMModelType.COHERE_COMMAND_R_PLUS,
+    LLMModelType.NOVA_PRO,
 ]
 
 EXPORT_SCENES = [
@@ -126,18 +127,45 @@ def prompt_template_render(self, prompt_template: dict):
 
 #### rag template #######
 
-CLAUDE_RAG_SYSTEM_PROMPT = """You are a customer service agent, and answering user's query. You ALWAYS follow these response rules when writing your response:
+CLAUDE_RAG_SYSTEM_PROMPT = """
+You are a customer service agent responding to user queries. ALWAYS adhere to these response rules:
+
 <response_rules>
-- 如果<docs> </docs>里面的内容包含markdown格式的图片，如 ![image](https://www.demo.com/demo.png)，请保留这个markdown格式的图片，并将他原封不动的输出到回答内容的最后，注意：不要修改这个markdown格式的图片.
-- NERVER say "根据搜索结果/大家好/谢谢/根据这个文档...".
-- 回答简单明了
-- 如果问题与<docs> </docs>里面的内容不相关，直接回答 "根据内部知识库，找不到相关内容。"
+1. Image Handling:
+   - If <docs></docs> contains markdown-formatted images, append them unaltered to the end of your response.
+   - Only process markdown-formatted images within <docs></docs>.
+   - IMPORTANT: If no markdown-formatted images are present in <docs></docs>, do not add any image references to your response.
+
+2. Language and Tone:
+   - Never use phrases like "According to search results," "Hello everyone," "Thank you," or "According to this document..."
+   - Provide concise and clear answers.
+   - Maintain a professional and helpful tone throughout the response.
+
+3. Relevance:
+   - If the query is unrelated to the content in <docs></docs>, respond with: "根据内部知识库，找不到相关内容。"
+
+4. Language Adaptation:
+   - Respond in the same language as the user's query.
+   - If the query is in a language other than English, adapt your response accordingly.
+
+5. Confidentiality:
+   - Do not disclose any information not present in the provided documents.
+   - If asked about topics outside your knowledge base, politely state that you don't have that information.
+
+6. Formatting:
+   - Use appropriate formatting (bold, italics, bullet points) to enhance readability when necessary.
+
+7. Completeness:
+   - Ensure your response addresses all aspects of the user's query.
+   - If multiple relevant documents are provided, synthesize the information coherently.
+
 </response_rules>
 
-Here are some documents for you to reference for your query.
+Reference the following documents to answer the query:
 <docs>
 {context}
-</docs>"""
+</docs>
+"""
 
 register_prompt_templates(
     model_ids=[
@@ -154,6 +182,7 @@ def prompt_template_render(self, prompt_template: dict):
         LLMModelType.LLAMA3_2_90B_INSTRUCT,
         LLMModelType.MISTRAL_LARGE_2407,
         LLMModelType.COHERE_COMMAND_R_PLUS,
+        LLMModelType.NOVA_PRO,
     ],
     task_type=LLMTaskType.RAG,
     prompt_template=CLAUDE_RAG_SYSTEM_PROMPT,
@@ -274,7 +303,7 @@ def prompt_template_render(self, prompt_template: dict):
         LLMModelType.LLAMA3_2_90B_INSTRUCT,
         LLMModelType.MISTRAL_LARGE_2407,
         LLMModelType.COHERE_COMMAND_R_PLUS,
-
+        LLMModelType.NOVA_PRO,
     ],
     task_type=LLMTaskType.CONVERSATION_SUMMARY_TYPE,
     prompt_template=CQR_SYSTEM_PROMPT,
@@ -299,6 +328,7 @@ def prompt_template_render(self, prompt_template: dict):
         LLMModelType.LLAMA3_2_90B_INSTRUCT,
         LLMModelType.MISTRAL_LARGE_2407,
         LLMModelType.COHERE_COMMAND_R_PLUS,
+        LLMModelType.NOVA_PRO,
     ],
     task_type=LLMTaskType.CONVERSATION_SUMMARY_TYPE,
     prompt_template=CQR_USER_PROMPT_TEMPLATE,
@@ -324,6 +354,7 @@ def prompt_template_render(self, prompt_template: dict):
         LLMModelType.LLAMA3_2_90B_INSTRUCT,
         LLMModelType.MISTRAL_LARGE_2407,
         LLMModelType.COHERE_COMMAND_R_PLUS,
+        LLMModelType.NOVA_PRO,
     ],
     task_type=LLMTaskType.CONVERSATION_SUMMARY_TYPE,
     prompt_template=json.dumps(CQR_FEW_SHOTS, ensure_ascii=False, indent=2),
@@ -382,15 +413,15 @@ def prompt_template_render(self, prompt_template: dict):
 AGENT_SYSTEM_PROMPT = """\
 You are a helpful and honest AI assistant. Today is {date},{weekday}. 
 Here are some guidelines for you:
-<guidlines>
+<guidelines>
 - Here are steps for you to decide to use which tool:
     1. Determine whether the current context is sufficient to answer the user's question.
     2. If the current context is sufficient to answer the user's question, call the `give_final_response` tool.
     3. If the current context is not sufficient to answer the user's question, you can consider calling one of the provided tools.
     4. If any of required parameters of the tool you want to call do not appears in context, call the `give_rhetorical_question` tool to ask the user for more information. 
 - Always output with the same language as the content from user. If the content is English, use English to output. If the content is Chinese, use Chinese to output.
 - Always call one tool at a time.
-</guidlines>
+</guidelines>
 Here's some context for reference:
 <context>
 {context}
@@ -441,6 +472,7 @@ def prompt_template_render(self, prompt_template: dict):
         LLMModelType.CLAUDE_3_5_SONNET,
         LLMModelType.CLAUDE_3_5_SONNET_V2,
         LLMModelType.CLAUDE_3_5_HAIKU,
+        LLMModelType.NOVA_PRO,
         # LLMModelType.LLAMA3_1_70B_INSTRUCT,
         # LLMModelType.LLAMA3_2_90B_INSTRUCT,
         # LLMModelType.MISTRAL_LARGE_2407,
@@ -458,7 +490,7 @@ def prompt_template_render(self, prompt_template: dict):
 Here's some context for reference:
 {context}
 
-## Guidlines
+## Guidelines
 Here are some guidelines for you:
 - Here are strategies for you to decide to use which tool:
     1. Determine whether the current context is sufficient to answer the user's question.
@@ -498,6 +530,7 @@ def prompt_template_render(self, prompt_template: dict):
         LLMModelType.LLAMA3_2_90B_INSTRUCT,
         LLMModelType.MISTRAL_LARGE_2407,
         LLMModelType.COHERE_COMMAND_R_PLUS,
+        LLMModelType.NOVA_PRO,
     ],
     task_type=LLMTaskType.TOOL_CALLING_API,
     prompt_template=TOOL_FEWSHOT_PROMPT,

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

@@ -72,7 +72,6 @@ def stream_response(event_body: dict, response: dict):
     ws_connection_id = event_body["ws_connection_id"]
     custom_message_id = event_body["custom_message_id"]
     answer = response["answer"]
-    figure = response.get("ddb_additional_kwargs", {}).get("figure")
     if isinstance(answer, str):
         answer = iter([answer])
 
@@ -133,10 +132,13 @@ def stream_response(event_body: dict, response: dict):
                 "message_type": StreamMessageType.CONTEXT,
                 "message_id": f"ai_{message_id}",
                 "custom_message_id": custom_message_id,
+                "ddb_additional_kwargs": {},
                 **response["extra_response"]
             }
-            if figure and len(figure) > 1:
-                context_msg["figure"] = figure
+
+            figure = response.get("extra_response").get("ref_figures", [])
+            if figure and figure[0]:
+                context_msg["ddb_additional_kwargs"]["figure"] = figure[0][:2]
             send_to_ws_client(
                 message=context_msg,
                 ws_connection_id=ws_connection_id

source/lambda/online/common_logic/langchain_integration/chains/__init__.py

@@ -39,6 +39,7 @@ def _import_chat_chain():
     Baichuan2Chat13B4BitsChatChain,
     Claude3HaikuChatChain,
     Claude3SonnetChatChain,
+    NovaProChatChain
 )
 
 def _import_conversation_summary_chain():
@@ -48,7 +49,8 @@ def _import_conversation_summary_chain():
     Claude21ConversationSummaryChain,
     Claude3HaikuConversationSummaryChain,
     Claude3SonnetConversationSummaryChain,
-    Internlm2Chat20BConversationSummaryChain
+    Internlm2Chat20BConversationSummaryChain,
+    NovaProConversationSummaryChain
 )
 
 def _import_intention_chain():
@@ -71,7 +73,8 @@ def _import_rag_chain():
     ClaudeInstanceRAGLLMChain,
     Claude3HaikuRAGLLMChain,
     Claude3SonnetRAGLLMChain,
-    Baichuan2Chat13B4BitsKnowledgeQaChain
+    Baichuan2Chat13B4BitsKnowledgeQaChain,
+    NovaProRAGLLMChain
 )
 
 
@@ -117,7 +120,8 @@ def _import_hyde_chain():
     Claude3SonnetHydeChain,
     ClaudeInstanceHydeChain,
     Internlm2Chat20BHydeChain,
-    Internlm2Chat7BHydeChain
+    Internlm2Chat7BHydeChain,
+    NovaProHydeChain
 )
 
 def _import_query_rewrite_chain():
@@ -128,7 +132,8 @@ def _import_query_rewrite_chain():
     Claude3HaikuQueryRewriteChain,
     Claude3SonnetQueryRewriteChain,
     Internlm2Chat20BQueryRewriteChain,
-    Internlm2Chat7BQueryRewriteChain
+    Internlm2Chat7BQueryRewriteChain,
+    NovaProQueryRewriteChain
 )
 
 
@@ -138,7 +143,8 @@ def _import_tool_calling_chain_claude_xml():
     Claude3HaikuToolCallingChain,
     Claude2ToolCallingChain,
     Claude3SonnetToolCallingChain,
-    ClaudeInstanceToolCallingChain
+    ClaudeInstanceToolCallingChain,
+    NovaProToolCallingChain
 )
 
 def _import_retail_conversation_summary_chain():
@@ -169,7 +175,8 @@ def _import_tool_calling_chain_api():
         Claude3SonnetToolCallingChain,
         Llama31Instruct70BToolCallingChain,
         CohereCommandRPlusToolCallingChain,
-        MistraLlarge2407ToolCallingChain
+        MistraLlarge2407ToolCallingChain,
+        NovaProToolCallingChain,
     )
 
 

source/lambda/online/common_logic/langchain_integration/chains/chat_chain.py

@@ -372,3 +372,7 @@ class ChatGPT4ChatChain(ChatGPT35ChatChain):
 
 class ChatGPT4oChatChain(ChatGPT35ChatChain):
     model_id = LLMModelType.CHATGPT_4O
+
+
+class NovaProChatChain(Claude2ChatChain):
+    model_id = LLMModelType.NOVA_PRO

source/lambda/online/common_logic/langchain_integration/chains/conversation_summary_chain.py

@@ -254,3 +254,7 @@ class Qwen2Instruct7BConversationSummaryChain(Claude2ConversationSummaryChain):
 
 class GLM4Chat9BConversationSummaryChain(Claude2ConversationSummaryChain):
     model_id = LLMModelType.GLM_4_9B_CHAT
+
+
+class NovaProConversationSummaryChain(Claude2ConversationSummaryChain):
+    model_id = LLMModelType.NOVA_PRO

source/lambda/online/common_logic/langchain_integration/chains/hyde_chain.py

@@ -101,3 +101,7 @@ def create_prompt(cls, x):
 class Internlm2Chat20BHydeChain(Internlm2Chat7BHydeChain):
     model_id = LLMModelType.INTERNLM2_CHAT_20B
     intent_type = HYDE_TYPE
+
+
+class NovaProHydeChain(Claude2HydeChain):
+    model_id = LLMModelType.NOVA_PRO

source/lambda/online/common_logic/langchain_integration/chains/query_rewrite_chain.py

@@ -87,7 +87,7 @@ class Claude3SonnetQueryRewriteChain(Claude2QueryRewriteChain):
 
 
 class Claude35SonnetQueryRewriteChain(Claude2QueryRewriteChain):
-    mdoel_id = LLMModelType.CLAUDE_3_5_SONNET
+    model_id = LLMModelType.CLAUDE_3_5_SONNET
 
 
 class Claude35SonnetV2QueryRewriteChain(Claude2QueryRewriteChain):
@@ -151,3 +151,8 @@ def create_chain(cls, model_kwargs=None, **kwargs):
 class Internlm2Chat20BQueryRewriteChain(Internlm2Chat7BQueryRewriteChain):
     model_id = LLMModelType.INTERNLM2_CHAT_20B
     intent_type = QUERY_REWRITE_TYPE
+
+
+class NovaProQueryRewriteChain(Claude2QueryRewriteChain):
+    model_id = LLMModelType.NOVA_PRO
+

source/lambda/online/common_logic/langchain_integration/chains/rag_chain.py

@@ -186,3 +186,8 @@ def add_system_prompt(x):
         )
         llm_chain = chat_history_chain | llm_chain
         return llm_chain
+
+
+class NovaProRAGLLMChain(Claude2RagLLMChain):
+    model_id = LLMModelType.NOVA_PRO
+

source/lambda/online/common_logic/langchain_integration/chains/tool_calling_chain_api.py

@@ -178,3 +178,7 @@ class MistraLlarge2407ToolCallingChain(Claude2ToolCallingChain):
 
 class CohereCommandRPlusToolCallingChain(Claude2ToolCallingChain):
     model_id = LLMModelType.COHERE_COMMAND_R_PLUS
+
+
+class NovaProToolCallingChain(Claude2ToolCallingChain):
+    model_id = LLMModelType.NOVA_PRO

source/lambda/online/common_logic/langchain_integration/chains/tool_calling_chain_claude_xml.py

@@ -321,3 +321,8 @@ class Claude3HaikuToolCallingChain(Claude2ToolCallingChain):
 
 class Claude35SonnetToolCallingChain(Claude2ToolCallingChain):
     model_id = "anthropic.claude-3-5-sonnet-20240620-v1:0"
+
+
+class NovaProToolCallingChain(Claude2ToolCallingChain):
+    model_id = LLMModelType.NOVA_PRO
+

source/lambda/online/common_logic/langchain_integration/chat_models/__init__.py

@@ -63,7 +63,8 @@ def _import_bedrock_models():
         Claude35SonnetV2,
         MistralLarge2407,
         Llama3d1Instruct70B,
-        CohereCommandRPlus
+        CohereCommandRPlus,
+        NovaPro
     )
 
 def _import_openai_models():
@@ -95,6 +96,7 @@ def _load_module(model_id):
     LLMModelType.COHERE_COMMAND_R_PLUS:_import_bedrock_models,
     LLMModelType.CLAUDE_3_5_SONNET_V2:_import_bedrock_models,
     LLMModelType.CLAUDE_3_5_HAIKU:_import_bedrock_models,
+    LLMModelType.NOVA_PRO:_import_bedrock_models,
 }
 
 

source/lambda/online/common_logic/langchain_integration/chat_models/bedrock_models.py

@@ -39,12 +39,14 @@ def create_model(cls, model_kwargs=None, **kwargs):
             or None
         )
         br_aws_access_key_id = os.environ.get("BEDROCK_AWS_ACCESS_KEY_ID", "")
-        br_aws_secret_access_key = os.environ.get("BEDROCK_AWS_SECRET_ACCESS_KEY", "")
+        br_aws_secret_access_key = os.environ.get(
+            "BEDROCK_AWS_SECRET_ACCESS_KEY", "")
 
         if br_aws_access_key_id != "" and br_aws_secret_access_key != "":
-            logger.info(f"Bedrock Using AWS AKSK from environment variables. Key ID: {br_aws_access_key_id}")
+            logger.info(
+                f"Bedrock Using AWS AKSK from environment variables. Key ID: {br_aws_access_key_id}")
 
-            client = boto3.client("bedrock-runtime", region_name=region_name, 
+            client = boto3.client("bedrock-runtime", region_name=region_name,
                                   aws_access_key_id=br_aws_access_key_id, aws_secret_access_key=br_aws_secret_access_key)
 
             llm = ChatBedrockConverse(
@@ -119,3 +121,9 @@ class CohereCommandRPlus(Claude2):
     model_id = LLMModelType.COHERE_COMMAND_R_PLUS
     enable_auto_tool_choice = False
     enable_prefill = False
+
+
+class NovaPro(Claude2):
+    model_id = LLMModelType.NOVA_PRO
+    enable_auto_tool_choice = False
+    enable_prefill = False