Update app.py

app.py

@@ -5,7 +5,7 @@ import time
 from langchain import PromptTemplate
 from langchain.llms import OpenAI
 from langchain.chat_models import ChatOpenAI
-from langchain.embeddings import HuggingFaceEmbeddings
+from langchain.embeddings import HuggingFaceEmbeddings, HuggingFaceInstructEmbeddings
 from langchain.vectorstores import Pinecone
 from langchain.chains import LLMChain
 from langchain.chains.question_answering import load_qa_chain
@@ -27,10 +27,11 @@ PINECONE_INDEX = os.environ.get("PINECONE_INDEX", '3gpp-r16')
 PINECONE_LINK  = "[Pinecone](https://www.pinecone.io)"
 LANGCHAIN_LINK = "[LangChain](https://python.langchain.com/en/latest/index.html)"
 
-EMBEDDING_MODEL = os.environ.get("PINECONE_INDEX", "sentence-transformers/all-mpnet-base-v2")
+EMBEDDING_MODEL = os.environ.get("EMBEDDING_MODEL", "hkunlp/instructor-large")
+EMBEDDING_LOADER = HuggingFaceInstructEmbeddings
 
 # return top-k text chunks from vector store
-TOP_K_DEFAULT = 15
+TOP_K_DEFAULT = 7
 TOP_K_MAX = 30
 SCORE_DEFAULT = 0.3
 
@@ -122,11 +123,11 @@ Question:
 {question}
 
 Optinal:
-Don't use standalone clause/figure name in the answer, expand it with corresponding metadata TS name
+Expand clause/figure name with corresponding metadata TS name in the answer.
 
 Desired format:
 Clause/figure name: <dot_separated_numbers>
-TS name: [\w\.]
+TS name: ^[\w\.]$
 
 Answer:"""
 )
@@ -171,7 +172,7 @@ def init_model(api_key, emb_name, db_api_key, db_env, db_index):
         if not (emb_name and db_api_key and db_env and db_index):
             return api_key,MODEL_DONE+DOCS_NULL,llm_dict,None,None,None
             
-        embeddings = HuggingFaceEmbeddings(model_name=emb_name)
+        embeddings = EMBEDDING_LOADER(model_name=emb_name)
 
         pinecone.init(api_key     = db_api_key,
                       environment = db_env)