cleanup

app/VectorStore/index/id_to_uuid_3c194f90-478a-4f8e-a5ac-67776218c783.pkl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:30970b5f5b747f17da7471033fc01cda622aeb236966628f64285f507247c30e
-size 7594
+oid sha256:b3fd923d38dbc7773fa8ddd035a3a12b35b36c0596120795d5441fa2631aa500
+size 7657

app/VectorStore/index/index_3c194f90-478a-4f8e-a5ac-67776218c783.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c220ae8125b8343264f932baba1c2abb62767acb593271c3745d43e50094eefb
-size 773296
+oid sha256:e8012c468a836e45dec5264f07e79a82dd9b0cfbd57b7db82ab3e5f87659e004
+size 779728

app/VectorStore/index/index_metadata_3c194f90-478a-4f8e-a5ac-67776218c783.pkl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:70331079a9c91229b14c8d52b03b769cc42a7753a3e28272be2dfc41323d1af7
+oid sha256:fe883ac5dc1e9c3d5b56fe942e1fef13b990df4e9b32e59c5eb7b12bba00e7c0
 size 73

app/VectorStore/index/uuid_to_id_3c194f90-478a-4f8e-a5ac-67776218c783.pkl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0e3ec8fd73d4a0139cfbc5a75be1534a545f7d6885200397eb3397dc792a60e
-size 8896
+oid sha256:d94d83b22ad6a388ffd24e1151e31ff2b22aaee250d0a8e442f0744bc00cffda
+size 8970

app/app.py

@@ -21,7 +21,7 @@ else:
 
     model_type = st.selectbox(
         'Select the Documents to be used to answer your question',
-        ('OpenAI', 'local_model') ) 
+        ('OpenAI', 'Load local model') ) 
 
     if model_type=='OpenAI':
         if 'openai_key' not in st.session_state:
@@ -33,6 +33,7 @@ else:
             os.environ["OPENAI_API_KEY"] = st.session_state.openai_key
         llm= load_model.load_openai_model()
     else:
+        # Add more models here
         llm = load_model.load_gpu_model("decapoda-research/llama-7b-hf")
 
 
@@ -43,7 +44,7 @@ else:
 
     st.write('You selected:', option)
 
-    chain = load_model.create_chain(llm, collection=option)
+    chain = load_model.create_chain(llm, collection=option, model_name="hkunlp/instructor-large ")
     try:
         query = st.text_area('Ask a question:', 'Hallo how are you today?')
         result = chain({"query": query})

app/load_model.py

@@ -44,7 +44,7 @@ def load_cpu_model():
     return llm
 
 @st.cache_resource(max_entries =1)
-def load_gpu_model(used_model = "chavinlo/gpt4-x-alpaca"):
+def load_gpu_model(used_model):
     torch.cuda.empty_cache()
     tokenizer = LlamaTokenizer.from_pretrained(used_model)
 
@@ -113,10 +113,7 @@ def load_vectorstore(model_name, collection):
         )
         return vectorstore
 
-def add_document_to_vectorstore(vectorstore, docs):
-    pass
-
-def create_chain(_llm, collection, model_name = "hkunlp/instructor-large"):
+def create_chain(_llm, collection, model_name):
     vectorstore = load_vectorstore(model_name, collection)
     retriever = vectorstore.as_retriever(search_kwargs={"k": 4})
     chain = RetrievalQA.from_chain_type(llm=_llm, chain_type="stuff", retriever=retriever, return_source_documents=True)

app/load_vectors.py

@@ -12,7 +12,7 @@ from hashlib import sha256
 import cloudpickle
 import logging
 import os
-from load_model import load_embedding
+from load_model import load_embedding, load_vectorstore
 import torch
 import re
 import pathlib
@@ -42,34 +42,20 @@ def create_collection(collection_name, model_name, client):
     return True
 
 def create_and_add(collection_name, sub_docs, model_name):
-    client_settings = chromadb.config.Settings(
-        chroma_db_impl="duckdb+parquet",
-        persist_directory=persist_directory,
-        anonymized_telemetry=False
-    )
-
-    client = chromadb.Client(client_settings)
-    collection_name = collection_name # + "_" + re.sub('[^A-Za-z0-9]+', '', model_name)
-
-    embeddings = load_embedding(model_name) 
     logging.info(f"Adding documents to {collection_name}")
-    vectorstore = Chroma(
-        collection_name=collection_name,
-        embedding_function=embeddings,
-        client_settings=client_settings,
-        persist_directory=persist_directory,
-    )
+    embeddings = load_embedding(model_name) 
+    vectorstore = load_vectorstore(model_name, collection_name)  
     vectorstore.add_documents(documents=sub_docs, embedding=embeddings)
     vectorstore.persist()
 
     # Test Vectorstore
-    vectorstore2 = Chroma(
-    collection_name=collection_name,
-    embedding_function=embeddings,
-    client_settings=client_settings,
-    persist_directory=persist_directory,
-    )
-    print( vectorstore2.similarity_search_with_score(query="What are AXAs green Goals?", k=4) )
+    #vectorstore2 = Chroma(
+    #collection_name=collection_name,
+    #embedding_function=embeddings,
+    #client_settings=client_settings,
+    #persist_directory=persist_directory,
+    #)
+    #print( vectorstore2.similarity_search_with_score(query="What are AXAs green Goals?", k=4) )
 
     return vectorstore
 
@@ -113,7 +99,7 @@ def load_from_web(urls, cache=True):
     #update metadata
     i=0
     for doc in docs:
-        doc.metadata = {'source': docs_list[i], 'url': docs_list[i], 'company':'AXA'}
+        doc.metadata = {'source': docs_list[i], 'url': docs_list[i], 'owner':'Heiko Wagner'}
         i=i+1
     return docs
         

app/run.py

@@ -12,6 +12,6 @@ import cloudpickle
 llm= load_model.load_openai_model()
 
 # %%
-chain = load_model.create_chain(llm, collection="heikospaper")
+chain = load_model.create_chain(llm, collection="heikospaper", model_name="hkunlp/instructor-large")
 result = chain({"query": "What are AXAs green Goals?"})
 print(result)