Additional minimal UI changes, heavy refactoring

app.py

@@ -1,66 +1,26 @@
 import streamlit as st
-from backend_utils import (get_plain_pipeline, get_retrieval_augmented_pipeline,
-                           get_web_retrieval_augmented_pipeline, set_q1, set_q2, set_q3, set_q4, set_q5, QUERIES,
-                           PLAIN_GPT_ANS, GPT_WEB_RET_AUG_ANS, GPT_LOCAL_RET_AUG_ANS)
+from utils.backend import (get_plain_pipeline, get_retrieval_augmented_pipeline,
+                           get_web_retrieval_augmented_pipeline)
+from utils.ui import set_q1, set_q2, set_q3, set_q4, set_q5, left_sidebar, right_sidebar, main_column
+from utils.constants import QUERIES, PLAIN_GPT_ANS, GPT_WEB_RET_AUG_ANS, GPT_LOCAL_RET_AUG_ANS
 
 st.set_page_config(
     page_title="Retrieval Augmentation with Haystack",
+    layout="wide"
 )
+left_sidebar()
 
 st.markdown("<center> <h2> Reduce Hallucinations with Retrieval Augmentation </h2> </center>", unsafe_allow_html=True)
 
 st.markdown("Ask a question about the collapse of the Silicon Valley Bank (SVB).", unsafe_allow_html=True)
 
-# if not st.session_state.get('pipelines_loaded', False):
-#     with st.spinner('Loading pipelines... \n This may take a few mins and might also fail if OpenAI API server is down.'):
-#         p1, p2, p3 = app_init()
-#         st.success('Pipelines are loaded', icon="✅")
-#         st.session_state['pipelines_loaded'] = True
+col_1, col_2 = st.columns([4, 2], gap="small")
+with col_1:
+    run_pressed, placeholder_plain_gpt, placeholder_retrieval_augmented = main_column()
+    print(f"Run value: {st.session_state.get('run', 'not found')}")
 
-placeholder = st.empty()
-with placeholder:
-    search_bar, button = st.columns([3, 1])
-    with search_bar:
-        username = st.text_area(f" ", max_chars=200, key='query')
-
-    with button:
-        st.write(" ")
-        st.write(" ")
-        run_pressed = st.button("Run")
-
-st.markdown("<center> <h5> Example questions </h5> </center>", unsafe_allow_html=True)
-
-st.write(" ")
-st.write(" ")
-c1, c2, c3, c4, c5 = st.columns(5)
-with c1:
-    st.button(QUERIES[0], on_click=set_q1)
-with c2:
-    st.button(QUERIES[1], on_click=set_q2)
-with c3:
-    st.button(QUERIES[2], on_click=set_q3)
-with c4:
-    st.button(QUERIES[3], on_click=set_q4)
-with c5:
-    st.button(QUERIES[4], on_click=set_q5)
-
-st.write(" ")
-st.radio("Answer Type:", ("Retrieval Augmented (Static news dataset)", "Retrieval Augmented with Web Search"), key="query_type")
-
-# st.sidebar.selectbox(
-#      "Example Questions:",
-#      QUERIES,
-#      key='q_drop_down', on_change=set_question)
-
-st.markdown(f"<h5> {PLAIN_GPT_ANS} </h5>", unsafe_allow_html=True)
-placeholder_plain_gpt = st.empty()
-st.text(" ")
-st.text(" ")
-if st.session_state.get("query_type", "Retrieval Augmented (Static news dataset)") == "Retrieval Augmented (Static news dataset)":
-    st.markdown(f"<h5> {GPT_LOCAL_RET_AUG_ANS} </h5>", unsafe_allow_html=True)
-else:
-    st.markdown(f"<h5>{GPT_WEB_RET_AUG_ANS} </h5>", unsafe_allow_html=True)
-placeholder_retrieval_augmented = st.empty()
+with col_2:
+    right_sidebar()
 
 if st.session_state.get('query') and run_pressed:
     ip = st.session_state['query']

backend_utils.py

@@ -1,107 +0,0 @@
-import streamlit as st
-from haystack import Pipeline
-from haystack.document_stores import FAISSDocumentStore
-from haystack.nodes import Shaper, PromptNode, PromptTemplate, PromptModel, EmbeddingRetriever
-from haystack.nodes.retriever.web import WebRetriever
-
-
-QUERIES = [
-    "Did SVB collapse?",
-    "Why did SVB collapse?",
-    "What does SVB failure mean for our economy?",
-    "Who is responsible for SVC collapse?",
-    "When did SVB collapse?"
-]
-PLAIN_GPT_ANS = "Answer with plain GPT"
-GPT_LOCAL_RET_AUG_ANS = "Answer with Retrieval Augmented GPT (Static news dataset)"
-GPT_WEB_RET_AUG_ANS = "Answer with Retrieval Augmented GPT (Web Search)"
-
-
-@st.cache_resource(show_spinner=False)
-def get_plain_pipeline():
-    prompt_open_ai = PromptModel(model_name_or_path="text-davinci-003", api_key=st.secrets["OPENAI_API_KEY"])
-    # Now let make one PromptNode use the default model and the other one the OpenAI model:
-    plain_llm_template = PromptTemplate(name="plain_llm", prompt_text="Answer the following question: $query")
-    node_openai = PromptNode(prompt_open_ai, default_prompt_template=plain_llm_template, max_length=300)
-    pipeline = Pipeline()
-    pipeline.add_node(component=node_openai, name="prompt_node", inputs=["Query"])
-    return pipeline
-
-
-@st.cache_resource(show_spinner=False)
-def get_retrieval_augmented_pipeline():
-    ds = FAISSDocumentStore(faiss_index_path="data/my_faiss_index.faiss",
-                            faiss_config_path="data/my_faiss_index.json")
-
-    retriever = EmbeddingRetriever(
-        document_store=ds,
-        embedding_model="sentence-transformers/multi-qa-mpnet-base-dot-v1",
-        model_format="sentence_transformers",
-        top_k=2
-    )
-    shaper = Shaper(func="join_documents", inputs={"documents": "documents"}, outputs=["documents"])
-
-    default_template = PromptTemplate(
-        name="question-answering",
-        prompt_text="Given the context please answer the question. Context: $documents; Question: "
-                    "$query; Answer:",
-    )
-    # Let's initiate the PromptNode
-    node = PromptNode("text-davinci-003", default_prompt_template=default_template,
-                      api_key=st.secrets["OPENAI_API_KEY"], max_length=500)
-
-    # Let's create a pipeline with Shaper and PromptNode
-    pipeline = Pipeline()
-    pipeline.add_node(component=retriever, name='retriever', inputs=['Query'])
-    pipeline.add_node(component=shaper, name="shaper", inputs=["retriever"])
-    pipeline.add_node(component=node, name="prompt_node", inputs=["shaper"])
-    return pipeline
-
-
-@st.cache_resource(show_spinner=False)
-def get_web_retrieval_augmented_pipeline():
-    search_key = st.secrets["WEBRET_API_KEY"]
-    web_retriever = WebRetriever(api_key=search_key, search_engine_provider="SerperDev")
-    shaper = Shaper(func="join_documents", inputs={"documents": "documents"}, outputs=["documents"])
-    default_template = PromptTemplate(
-        name="question-answering",
-        prompt_text="Given the context please answer the question. Context: $documents; Question: "
-                    "$query; Answer:",
-    )
-    # Let's initiate the PromptNode
-    node = PromptNode("text-davinci-003", default_prompt_template=default_template,
-                      api_key=st.secrets["OPENAI_API_KEY"], max_length=500)
-    # Let's create a pipeline with Shaper and PromptNode
-    pipeline = Pipeline()
-    pipeline.add_node(component=web_retriever, name='retriever', inputs=['Query'])
-    pipeline.add_node(component=shaper, name="shaper", inputs=["retriever"])
-    pipeline.add_node(component=node, name="prompt_node", inputs=["shaper"])
-    return pipeline
-
-
-if 'query' not in st.session_state:
-    st.session_state['query'] = ""
-
-
-def set_question():
-    st.session_state['query'] = st.session_state['q_drop_down']
-
-
-def set_q1():
-    st.session_state['query'] = QUERIES[0]
-
-
-def set_q2():
-    st.session_state['query'] = QUERIES[1]
-
-
-def set_q3():
-    st.session_state['query'] = QUERIES[2]
-
-
-def set_q4():
-    st.session_state['query'] = QUERIES[3]
-
-
-def set_q5():
-    st.session_state['query'] = QUERIES[4]

utils/constants.py

@@ -0,0 +1,10 @@
+QUERIES = [
+    "Did SVB collapse?",
+    "Why did SVB collapse?",
+    "What does SVB failure mean for our economy?",
+    "Who is responsible for SVC collapse?",
+    "When did SVB collapse?"
+]
+PLAIN_GPT_ANS = "Answer with plain GPT"
+GPT_LOCAL_RET_AUG_ANS = "Answer with Retrieval Augmented GPT (Static news dataset)"
+GPT_WEB_RET_AUG_ANS = "Answer with Retrieval Augmented GPT (Web Search)"

utils/ui.py

@@ -0,0 +1,114 @@
+import streamlit as st
+from PIL import Image
+
+from .constants import QUERIES, PLAIN_GPT_ANS, GPT_WEB_RET_AUG_ANS, GPT_LOCAL_RET_AUG_ANS
+
+
+def set_question():
+    st.session_state['query'] = st.session_state['q_drop_down']
+
+
+def set_q1():
+    st.session_state['query'] = QUERIES[0]
+
+
+def set_q2():
+    st.session_state['query'] = QUERIES[1]
+
+
+def set_q3():
+    st.session_state['query'] = QUERIES[2]
+
+
+def set_q4():
+    st.session_state['query'] = QUERIES[3]
+
+
+def set_q5():
+    st.session_state['query'] = QUERIES[4]
+
+def main_column():
+    placeholder = st.empty()
+    with placeholder:
+        search_bar, button = st.columns([3, 1])
+        with search_bar:
+            username = st.text_area(f" ", max_chars=200, key='query')
+
+        with button:
+            st.write(" ")
+            st.write(" ")
+            run_pressed = st.button("Run", key="run")
+
+    st.write(" ")
+    st.radio("Answer Type:", ("Retrieval Augmented (Static news dataset)", "Retrieval Augmented with Web Search"), key="query_type")
+
+    # st.sidebar.selectbox(
+    #      "Example Questions:",
+    #      QUERIES,
+    #      key='q_drop_down', on_change=set_question)
+
+    st.markdown(f"<h5> {PLAIN_GPT_ANS} </h5>", unsafe_allow_html=True)
+    placeholder_plain_gpt = st.empty()
+    st.text(" ")
+    st.text(" ")
+    if st.session_state.get("query_type", "Retrieval Augmented (Static news dataset)") == "Retrieval Augmented (Static news dataset)":
+        st.markdown(f"<h5> {GPT_LOCAL_RET_AUG_ANS} </h5>", unsafe_allow_html=True)
+    else:
+        st.markdown(f"<h5>{GPT_WEB_RET_AUG_ANS} </h5>", unsafe_allow_html=True)
+    placeholder_retrieval_augmented = st.empty()
+
+    return run_pressed, placeholder_plain_gpt, placeholder_retrieval_augmented
+
+
+def right_sidebar():
+    st.markdown("<h5> Example questions </h5>", unsafe_allow_html=True)
+    # c1, c2, c3, c4, c5 = st.columns(5)
+    # with c1:
+    st.button(QUERIES[0], on_click=set_q1)
+    # with c2:
+    st.button(QUERIES[1], on_click=set_q2)
+    # with c3:
+    st.button(QUERIES[2], on_click=set_q3)
+    # with c4:
+    st.button(QUERIES[3], on_click=set_q4)
+    # with c5:
+    st.button(QUERIES[4], on_click=set_q5)
+
+
+def left_sidebar():
+    with st.sidebar:
+        image = Image.open('logo/haystack-logo-colored.png')
+        st.markdown("Thanks for coming to this 🤗 Space.\n\n"
+                    "This is an effort towards showcasing how can you use Haystack for Retrieval Augmented QA, "
+                    "with local document store as well as WebRetriever (coming soon!) \n\n"
+                    "For more on how this was built, instructions along with a Repository "
+                    "will be published soon and updated here.")
+
+        # st.markdown(
+        #     "## How to use\n"
+        #     "1. Enter your [OpenAI API key](https://platform.openai.com/account/api-keys) below\n"
+        #     "2. Enter a Serper Dev API key\n"
+        #     "3. Enjoy 🤗\n"
+        # )
+
+        # api_key_input = st.text_input(
+        #     "OpenAI API Key",
+        #     type="password",
+        #     placeholder="Paste your OpenAI API key here (sk-...)",
+        #     help="You can get your API key from https://platform.openai.com/account/api-keys.",
+        #     value=st.session_state.get("OPENAI_API_KEY", ""),
+        # )
+
+        # if api_key_input:
+        #     set_openai_api_key(api_key_input)
+
+        st.markdown("---")
+        st.markdown(
+            "## How this works\n"
+            "This app was built with [Haystack](https://haystack.deepset.ai) using the"
+            " [`PromptNode`](https://docs.haystack.deepset.ai/docs/prompt_node) and [`Retriever`](https://docs.haystack.deepset.ai/docs/retriever#embedding-retrieval-recommended).\n\n"
+            " You can find the source code in **Files and versions** tab."
+        )
+
+        st.markdown("---")
+        st.image(image, width=250)