V1.5.0

application/.env.template

@@ -21,7 +21,7 @@ RDS_REGION_NAME=us-west-2
 AWS_DEFAULT_REGION=us-west-2
 DYNAMODB_AWS_REGION=us-west-2
 
-EMBEDDING_DIMENSION=1536
+EMBEDDING_DIMENSION=1024
 BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v1
 
 # If you need to use ak/sk to access bedrock, please configure bedrock's ak/sk to Secrets Manager, Examples are as follows
@@ -32,4 +32,4 @@ BEDROCK_SECRETS_AK_SK=
 OPENSEARCH_SECRETS_URL_HOST=opensearch-host-url
 OPENSEARCH_SECRETS_USERNAME_PASSWORD=opensearch-master-user
 
-# SAGEMAKER_ENDPOINT_EMBEDDING=
+SAGEMAKER_ENDPOINT_EMBEDDING=bge-zh-15-2024-08-17-03-56-58-281-endpoint

application/config_files/stauth_config.yaml

@@ -5,7 +5,7 @@ credentials:
       failed_login_attempts: 0 # Will be managed automatically
       logged_in: False # Will be managed automatically
       name: AWS
-      password: $2b$12$NDQv5NLaWiVlNuzQYHwAo.tv.f.TuX1nbdoUZi44/Y3xv4I4QAfjy # Set the password following instructions in README
+      password: $2b$12$pP4Vi1ovItxf/22zYn1UFeYrA2IM/D7glGNoAd3TrY0Gr4QzdZSNC # Set the password following instructions in README
 cookie:
   expiry_days: 2
   key: some_signature_key # Must be string

application/nlq/business/vector_store.py

@@ -76,39 +76,42 @@ def get_all_agent_cot_samples(cls, profile_name):
     def add_sample(cls, profile_name, question, answer):
         logger.info(f'add sample question: {question} to profile {profile_name}')
         if SAGEMAKER_ENDPOINT_EMBEDDING is not None and SAGEMAKER_ENDPOINT_EMBEDDING != "":
-            embedding = cls.create_vector_embedding_with_sagemaker(question)
+            embedding = cls.create_vector_embedding_with_sagemaker(SAGEMAKER_ENDPOINT_EMBEDDING,question,opensearch_info['sql_index'])
         else:
             embedding = cls.create_vector_embedding_with_bedrock(question)
-        has_same_sample = cls.search_same_query(profile_name, 1, opensearch_info['sql_index'], embedding)
+        #has_same_sample = cls.search_same_query(profile_name, 1, opensearch_info['sql_index'], embedding)
+        has_same_sample = False
         if has_same_sample:
             logger.info(f'delete sample sample entity: {question} to profile {profile_name}')
-        if cls.opensearch_dao.add_sample(opensearch_info['sql_index'], profile_name, question, answer, embedding):
+        if cls.opensearch_dao.add_sample(opensearch_info['sql_index'], profile_name, question, answer, embedding['vector_field']):
             logger.info('Sample added')
 
     @classmethod
     def add_entity_sample(cls, profile_name, entity, comment):
         logger.info(f'add sample entity: {entity} to profile {profile_name}')
         if SAGEMAKER_ENDPOINT_EMBEDDING is not None and SAGEMAKER_ENDPOINT_EMBEDDING != "":
-            embedding = cls.create_vector_embedding_with_sagemaker(entity)
+            embedding = cls.create_vector_embedding_with_sagemaker(SAGEMAKER_ENDPOINT_EMBEDDING,entity,opensearch_info['ner_index'])
         else:
             embedding = cls.create_vector_embedding_with_bedrock(entity)
-        has_same_sample = cls.search_same_query(profile_name, 1, opensearch_info['ner_index'], embedding)
+        #has_same_sample = cls.search_same_query(profile_name, 1, opensearch_info['ner_index'], embedding)
+        has_same_sample = False
         if has_same_sample:
             logger.info(f'delete sample sample entity: {entity} to profile {profile_name}')
-        if cls.opensearch_dao.add_entity_sample(opensearch_info['ner_index'], profile_name, entity, comment, embedding):
+        if cls.opensearch_dao.add_entity_sample(opensearch_info['ner_index'], profile_name, entity, comment, embedding['vector_field']):
             logger.info('Sample added')
 
     @classmethod
     def add_agent_cot_sample(cls, profile_name, entity, comment):
         logger.info(f'add agent sample query: {entity} to profile {profile_name}')
         if SAGEMAKER_ENDPOINT_EMBEDDING is not None and SAGEMAKER_ENDPOINT_EMBEDDING != "":
-            embedding = cls.create_vector_embedding_with_sagemaker(entity)
+            embedding = cls.create_vector_embedding_with_sagemaker(SAGEMAKER_ENDPOINT_EMBEDDING,entity,opensearch_info['agent_index'])
         else:
             embedding = cls.create_vector_embedding_with_bedrock(entity)
-        has_same_sample = cls.search_same_query(profile_name, 1, opensearch_info['agent_index'], embedding)
+        #has_same_sample = cls.search_same_query(profile_name, 1, opensearch_info['agent_index'], embedding)
+        has_same_sample = False
         if has_same_sample:
             logger.info(f'delete agent sample sample query: {entity} to profile {profile_name}')
-        if cls.opensearch_dao.add_agent_cot_sample(opensearch_info['agent_index'], profile_name, entity, comment, embedding):
+        if cls.opensearch_dao.add_agent_cot_sample(opensearch_info['agent_index'], profile_name, entity, comment, embedding['vector_field']):
             logger.info('Sample added')
 
     @classmethod
@@ -129,19 +132,19 @@ def create_vector_embedding_with_bedrock(cls, text):
         return embedding
 
     @classmethod
-    def create_vector_embedding_with_sagemaker(cls, text):
-        try:
-            body = json.dumps(
-                {
-                    "inputs": text,
-                    "is_query": True
-                }
-            )
-            response = invoke_model_sagemaker_endpoint(SAGEMAKER_ENDPOINT_EMBEDDING, body, model_type="embedding")
-            embeddings = response[0]
-            return embeddings
-        except Exception as e:
-            logger.error(f'create_vector_embedding_with_sagemaker is error {e}')
+    def create_vector_embedding_with_sagemaker(cls,endpoint_name, text, index_name):
+        body=json.dumps(
+        {
+            "inputs": text,
+            "is_query": True,
+            "instruction" :  "Represent this sentence for searching relevant passages:"
+        }
+        )
+        response = invoke_model_sagemaker_endpoint(endpoint_name, body, "embeddings")
+        embeddings = response['sentence_embeddings'][0]
+        logger.info("embeddings to ingestion")
+        logger.info(embeddings[:10])
+        return {"_index": index_name, "text": text, "vector_field": embeddings}
 
     @classmethod
     def delete_sample(cls, profile_name, doc_id):

application/nlq/data_access/opensearch.py

@@ -3,8 +3,9 @@
 from opensearchpy import OpenSearch
 from opensearchpy.helpers import bulk
 
-from utils.llm import create_vector_embedding_with_bedrock
-
+from utils.llm import create_vector_embedding_with_bedrock,create_vector_embedding_with_sagemaker
+from utils.env_var import BEDROCK_REGION, AOS_HOST, AOS_PORT, AOS_USER, AOS_PASSWORD, opensearch_info, \
+    SAGEMAKER_ENDPOINT_EMBEDDING
 logger = logging.getLogger(__name__)
 
 def put_bulk_in_opensearch(list, client):
@@ -190,8 +191,14 @@ def delete_sample(self, index_name, profile_name, doc_id):
         return self.opensearch_client.delete(index=index_name, id=doc_id)
 
     def search_sample(self, profile_name, top_k, index_name, query):
-        records_with_embedding = create_vector_embedding_with_bedrock(query, index_name=index_name)
-        return self.search_sample_with_embedding(profile_name, top_k, index_name,  records_with_embedding['vector_field'])
+        if SAGEMAKER_ENDPOINT_EMBEDDING is not None and SAGEMAKER_ENDPOINT_EMBEDDING != "":
+            records_with_embedding = create_vector_embedding_with_sagemaker(SAGEMAKER_ENDPOINT_EMBEDDING, query, index_name)
+            return self.search_sample_with_embedding(profile_name, top_k, index_name,  records_with_embedding['vector_field'])
+        else:
+            records_with_embedding = create_vector_embedding_with_bedrock(query, index_name=index_name)
+            return self.search_sample_with_embedding(profile_name, top_k, index_name,  records_with_embedding['vector_field'])
+
+
 
 
     def search_sample_with_embedding(self, profile_name, top_k, index_name, query_embedding):

application/utils/llm.py

@@ -232,7 +232,7 @@ def invoke_model_sagemaker_endpoint(endpoint_name, body, model_type="LLM", with_
                 Body=body,
                 ContentType="application/json",
             )
-            response_body = json.loads(response.get('Body').read())
+            response_body = json.loads(response.get('Body').read().decode("utf8"))
             return response_body
 
 
@@ -570,14 +570,17 @@ def create_vector_embedding_with_bedrock(text, index_name):
 
 
 def create_vector_embedding_with_sagemaker(endpoint_name, text, index_name):
-    body = json.dumps(
+    body=json.dumps(
         {
             "inputs": text,
-            "is_query": True
+            "is_query": True,
+            "instruction" :  "Represent this sentence for searching relevant passages:"
         }
     )
-    response = invoke_model_sagemaker_endpoint(endpoint_name, body, model_type="embedding")
-    embeddings = response[0]
+    response = invoke_model_sagemaker_endpoint(endpoint_name, body, "embeddings")
+    embeddings = response['sentence_embeddings'][0]
+    logger.info("embeddings in llm.py")
+    logger.info(embeddings[:10])
     return {"_index": index_name, "text": text, "vector_field": embeddings}