instructions & tests

Dockerfile

@@ -11,7 +11,6 @@ COPY ./app /app
 WORKDIR /app
 RUN mkdir /data
 
-
 RUN pip install --no-cache-dir --upgrade -r requirements.txt
 # ^ no caching of the packages to save space
 
@@ -22,5 +21,7 @@ RUN pip install --no-cache-dir --upgrade -r requirements.txt
 RUN chmod -R 777 /usr/local/lib/python3.10/site-packages/llama_index/legacy/_static/nltk_cache
 
 ENV TRANSFORMERS_CACHE=/usr/local/lib/python3.10/site-packages/llama_index/legacy/_static/nltk_cache
+# ^ not elegant but it works
+# HF warning says that TRANSFORMERS_CACHE will be deprecated in transformers v5, and advise to use HF_HOME
 
 CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]

README.md

@@ -9,3 +9,13 @@ license: mit
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+
+### How to use the endpoint
+
+Please see the notebook 'app/notebooks/upload_index.ipynb' for examples of how to upload docs, index them, delete the data, erase the vector store, do a vector search or a full RAG.  
+
+One can upload as many documents as he wants, and decide when to index them, and then continue uploading documents, and, again, index them at any time.  
+
+The code works locally with uvicorn and here on Huggingface.  
+
+In 'tests/test_main.py', one can find a few ideas about how to test the code.  It is of course far from being exhaustive, but I included simple unit tests and also some that test the overall capability of the code, ie answering a question with a LLM, fed by the results of a hybrid search on a Weaviate database.  

app/tests/test_main.py

@@ -0,0 +1,36 @@
+import os, sys
+sys.path.append("../")
+from main import app
+
+from fastapi.testclient import TestClient
+
+from settings import datadir
+
+client = TestClient(app)
+
+def test_read_root():
+    response = client.get("/ping/")
+    assert response.status_code == 200
+    assert int(response.json()['answer']) < 100
+    
+def test_list_files():
+    response = client.get("/list_files/")
+    files = os.listdir(datadir)
+    assert response.status_code == 200
+    assert len(response.json()['files']) == len(files)
+    for f in response.json()['files']:
+        assert f in files
+        
+def test_vector_search():
+    question_data = {"question": "Does ATT have postpaid phone customers?"}
+    response = client.post("/ask/", json=question_data)
+    assert response.status_code == 200
+    assert len(response.json()['answer']) > 0 # we assume vector store works if it returns something
+    assert any(['postpaid' in a.lower() for a in response.json()['answer']])
+    
+
+def test_ragit():
+    question_data = {"question": "Does ATT have postpaid phone customers?"}
+    response = client.post("/ragit/", json=question_data)
+    assert response.status_code == 200
+    assert 'yes' in response.json()['answer'].lower()