From 8054bcac39121ac3297608ae4d68f07fe78ce879 Mon Sep 17 00:00:00 2001
From: Dai Sugimori <daixque@gmail.com>
Date: Sat, 23 Nov 2024 17:06:51 +0900
Subject: [PATCH 1/2] Add support for semantic search with the dense vector

---
 example-apps/chatbot-rag-app/api/chat.py | 38 +++++++++++++++++-------
 example-apps/chatbot-rag-app/env.example |  9 ++++++
 2 files changed, 37 insertions(+), 10 deletions(-)

diff --git a/example-apps/chatbot-rag-app/api/chat.py b/example-apps/chatbot-rag-app/api/chat.py
index 54380010..c03b00a4 100644
--- a/example-apps/chatbot-rag-app/api/chat.py
+++ b/example-apps/chatbot-rag-app/api/chat.py
@@ -1,4 +1,4 @@
-from langchain_elasticsearch import ElasticsearchStore, SparseVectorStrategy
+from langchain_elasticsearch import ElasticsearchStore, DenseVectorStrategy, SparseVectorStrategy
 from llm_integrations import get_llm
 from elasticsearch_client import (
     elasticsearch_client,
@@ -12,16 +12,33 @@
 INDEX_CHAT_HISTORY = os.getenv(
     "ES_INDEX_CHAT_HISTORY", "workplace-app-docs-chat-history"
 )
-ELSER_MODEL = os.getenv("ELSER_MODEL", ".elser_model_2")
+MODEL_ID = os.getenv("ES_MODEL_ID", ".elser_model_2")
+STRATEGY_TYPE = os.getenv("ES_STRATEGY_TYPE", "sparse")
+VECTOR_FIELD = os.getenv("ES_VECTOR_FIELD", "vector")
+QUERY_FIELD = os.getenv("ES_QUERY_FIELD", "text")
+
 SESSION_ID_TAG = "[SESSION_ID]"
 SOURCE_TAG = "[SOURCE]"
 DONE_TAG = "[DONE]"
 
-store = ElasticsearchStore(
-    es_connection=elasticsearch_client,
-    index_name=INDEX,
-    strategy=SparseVectorStrategy(model_id=ELSER_MODEL),
-)
+if STRATEGY_TYPE == "sparse":
+    strategy = SparseVectorStrategy(model_id=MODEL_ID)
+    store = ElasticsearchStore(
+        es_connection=elasticsearch_client,
+        index_name=INDEX,
+        strategy=strategy,
+    )
+elif STRATEGY_TYPE == "dense":
+    strategy = DenseVectorStrategy(model_id=MODEL_ID, hybrid=True)
+    store = ElasticsearchStore(
+        es_connection=elasticsearch_client,
+        index_name=INDEX,
+        vector_query_field=VECTOR_FIELD,
+        query_field=QUERY_FIELD,
+        strategy=strategy,
+    )
+else:
+    raise ValueError(f"Invalid strategy type: {STRATEGY_TYPE}")
 
 
 @stream_with_context
@@ -50,9 +67,10 @@ def ask_question(question, session_id):
     docs = store.as_retriever().invoke(condensed_question)
     for doc in docs:
         doc_source = {**doc.metadata, "page_content": doc.page_content}
-        current_app.logger.debug(
-            "Retrieved document passage from: %s", doc.metadata["name"]
-        )
+        if "name" in doc.metadata:
+            current_app.logger.debug(
+                "Retrieved document passage from: %s", doc.metadata["name"]
+            )
         yield f"data: {SOURCE_TAG} {json.dumps(doc_source)}\n\n"
 
     qa_prompt = render_template(
diff --git a/example-apps/chatbot-rag-app/env.example b/example-apps/chatbot-rag-app/env.example
index 2dc651cc..5c6c5b0e 100644
--- a/example-apps/chatbot-rag-app/env.example
+++ b/example-apps/chatbot-rag-app/env.example
@@ -11,6 +11,15 @@ ELASTIC_API_KEY=
 ES_INDEX=workplace-app-docs
 ES_INDEX_CHAT_HISTORY=workplace-app-docs-chat-history
 
+# For sparse vector (e.g. ELSER)
+ES_STRATEGY_TYPE=sparse
+ES_MODEL_ID=.elser_model_2
+# For dense vector (e.g. E5)
+# ES_STRATEGY_TYPE=dense
+# ES_MODEL_ID=.multilingual-e5-small_linux-x86_64
+# ES_VECTOR_FIELD=text_semantic.inference.chunks.embeddings
+# ES_QUERY_FIELD=text
+
 # Uncomment and complete if you want to use OpenAI
 # LLM_TYPE=openai
 # OPENAI_API_KEY=

From a205ec91c5e93cc5c042004f9ddb93f906d5ae44 Mon Sep 17 00:00:00 2001
From: Dai Sugimori <daixque@gmail.com>
Date: Sat, 23 Nov 2024 17:22:12 +0900
Subject: [PATCH 2/2] apply formatter

---
 example-apps/chatbot-rag-app/api/chat.py | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/example-apps/chatbot-rag-app/api/chat.py b/example-apps/chatbot-rag-app/api/chat.py
index c03b00a4..c7923b0d 100644
--- a/example-apps/chatbot-rag-app/api/chat.py
+++ b/example-apps/chatbot-rag-app/api/chat.py
@@ -1,4 +1,8 @@
-from langchain_elasticsearch import ElasticsearchStore, DenseVectorStrategy, SparseVectorStrategy
+from langchain_elasticsearch import (
+    ElasticsearchStore,
+    DenseVectorStrategy,
+    SparseVectorStrategy,
+)
 from llm_integrations import get_llm
 from elasticsearch_client import (
     elasticsearch_client,