Merge pull request #43 from bukosabino/feature/tavily-and-gpt35-endpoint

Adding endpoints using the Tavily API and GPT3.5 to perform some comparisons

doc/cron_etl_daily_public.sh

@@ -11,4 +11,4 @@ export QDRANT_API_URL="<your_supabase_api_url>"
 cd ia-boe/
 source venv3.9/bin/activate
 pip install -r requirements.txt
-python -m src.etls.etl_daily
+python -m src.etls.boe.load.daily

requirements.txt

@@ -18,6 +18,7 @@ supabase==1.0.2
 pinecone-client==2.2.2
 sentence_transformers==2.2.2
 openai==1.2.3
+tavily-python==0.2.6
 
 sendgrid==6.10.0
 

src/initialize.py

@@ -18,6 +18,7 @@
 from qdrant_client import QdrantClient
 from qdrant_client.models import Distance, VectorParams
 from supabase.client import Client, create_client
+from tavily import TavilyClient
 
 from src.utils import StandardSupabaseVectorStore
 
@@ -44,12 +45,13 @@ def initialize_app():
     config_loader = _init_config()
     vector_store = _init_vector_store(config_loader)
     openai_client = _init_openai_client()
+    tavily_client = TavilyClient(api_key=os.environ['TAVILY_API_KEY'])
     # retrieval_qa = _init_retrieval_qa_llm(vector_store, config_loader)
     logger.info("Initialized application")
     init_objects = collections.namedtuple(
-        "init_objects", ["config_loader", "vector_store", "openai_client"]
+        "init_objects", ["config_loader", "vector_store", "openai_client", "tavily_client"]
     )
-    return init_objects(config_loader, vector_store, openai_client)
+    return init_objects(config_loader, vector_store, openai_client, tavily_client)
 
 
 def _init_config():

src/service/main.py

@@ -41,6 +41,24 @@ async def semantic_search(input_query: str = DEFAULT_INPUT_QUERY):
     return docs
 
 
+@APP.get("/semantic_search_tavily")
+@timeit
+async def semantic_search_tavily(input_query: str = DEFAULT_INPUT_QUERY):
+    logger = lg.getLogger(semantic_search_tavily.__name__)
+    logger.info(input_query)
+    docs = INIT_OBJECTS.tavily_client.search(
+        query=input_query,
+        search_depth="advanced",
+        include_domains=["https://www.boe.es/"],
+        max_results=10,
+        topic="general",
+        include_raw_content=False,
+        include_answer=False
+    )
+    logger.info(docs)
+    return docs
+
+
 async def a_request_get(url):
     """Requests for sync/async load tests"""
     async with httpx.AsyncClient(timeout=10.0) as client:
@@ -63,6 +81,62 @@ async def qa(input_query: str = DEFAULT_INPUT_QUERY):
     context_preprocessed = [
         {"context": doc[0].page_content, "score": doc[1]} for doc in docs
     ]
+    messages = [
+        {"role": "system", "content": INIT_OBJECTS.config_loader["prompt_system"]},
+        {
+            "role": "system",
+            "content": INIT_OBJECTS.config_loader["prompt_system_context"],
+        },
+        {"role": "system", "content": "A continuación se proporciona el contexto:"},
+        {"role": "system", "content": str(context_preprocessed)},
+        {
+            "role": "system",
+            "content": "A continuación se proporciona la pregunta del usuario:",
+        },
+        {"role": "user", "content": input_query},
+    ]
+    # logger.info(messages)
+    response = await INIT_OBJECTS.openai_client.chat.completions.create(
+        model=INIT_OBJECTS.config_loader["llm_model_name"],
+        messages=messages,
+        temperature=INIT_OBJECTS.config_loader["temperature"],
+        seed=INIT_OBJECTS.config_loader["seed"],
+        max_tokens=INIT_OBJECTS.config_loader["max_tokens"],
+    )
+    answer = response.choices[0].message.content
+    logger.info(answer)
+    logger.info(response.usage)
+
+    response_payload = dict(
+        scoring_id=str(uuid.uuid4()),
+        context=docs,
+        answer=answer,
+    )
+    return response_payload
+
+
+@APP.get("/qa_tavily")
+@timeit
+async def qa_tavily(input_query: str = DEFAULT_INPUT_QUERY):
+    logger = lg.getLogger(qa_tavily.__name__)
+    logger.info(input_query)
+
+    # Getting context from internet browser (Tavily)
+    docs = INIT_OBJECTS.tavily_client.search(
+        query=input_query,
+        search_depth="advanced",
+        include_domains=["https://www.boe.es/"],
+        max_results=10,
+        topic="general",
+        include_raw_content=False,
+        include_answer=False
+    )
+
+    # Generate response using a LLM (OpenAI)
+    context_preprocessed = [
+        {"context": doc['content'], "score": doc['score']} for doc in docs['results']
+    ]
+
     response = await INIT_OBJECTS.openai_client.chat.completions.create(
         model=INIT_OBJECTS.config_loader["llm_model_name"],
         messages=[
@@ -95,6 +169,55 @@ async def qa(input_query: str = DEFAULT_INPUT_QUERY):
     return response_payload
 
 
+@APP.get("/qa_35")
+@timeit
+async def qa_35(input_query: str = DEFAULT_INPUT_QUERY):
+    logger = lg.getLogger(qa_35.__name__)
+    logger.info(input_query)
+
+    # Getting context from embedding database (Qdrant)
+    docs = await INIT_OBJECTS.vector_store.asimilarity_search_with_score(
+        query=input_query, k=INIT_OBJECTS.config_loader["top_k_results"]
+    )
+
+    # Generate response using a LLM (OpenAI)
+    context_preprocessed = [
+        {"context": doc[0].page_content, "score": doc[1]} for doc in docs
+    ]
+    messages = [
+        {"role": "system", "content": INIT_OBJECTS.config_loader["prompt_system"]},
+        {
+            "role": "system",
+            "content": INIT_OBJECTS.config_loader["prompt_system_context"],
+        },
+        {"role": "system", "content": "A continuación se proporciona el contexto:"},
+        {"role": "system", "content": str(context_preprocessed)},
+        {
+            "role": "system",
+            "content": "A continuación se proporciona la pregunta del usuario:",
+        },
+        {"role": "user", "content": input_query},
+    ]
+    # logger.info(messages)
+    response = await INIT_OBJECTS.openai_client.chat.completions.create(
+        model='gpt-3.5-turbo',
+        messages=messages,
+        temperature=INIT_OBJECTS.config_loader["temperature"],
+        seed=INIT_OBJECTS.config_loader["seed"],
+        max_tokens=INIT_OBJECTS.config_loader["max_tokens"],
+    )
+    answer = response.choices[0].message.content
+    logger.info(answer)
+    logger.info(response.usage)
+
+    response_payload = dict(
+        scoring_id=str(uuid.uuid4()),
+        context=docs,
+        answer=answer,
+    )
+    return response_payload
+
+
 @APP.get("/sleep")
 @timeit
 async def sleep():