Spaces:
Runtime error
Runtime error
Commit
Β·
bd2e0e7
1
Parent(s):
e09fe1d
Additional minimal UI changes, heavy refactoring
Browse files- app.py +12 -52
- backend_utils.py +0 -107
- utils/__init__.py +0 -0
- utils/constants.py +10 -0
- utils/ui.py +114 -0
app.py
CHANGED
|
@@ -1,66 +1,26 @@
|
|
| 1 |
import streamlit as st
|
| 2 |
-
from
|
| 3 |
-
get_web_retrieval_augmented_pipeline
|
| 4 |
-
|
|
|
|
| 5 |
|
| 6 |
st.set_page_config(
|
| 7 |
page_title="Retrieval Augmentation with Haystack",
|
|
|
|
| 8 |
)
|
|
|
|
| 9 |
|
| 10 |
st.markdown("<center> <h2> Reduce Hallucinations with Retrieval Augmentation </h2> </center>", unsafe_allow_html=True)
|
| 11 |
|
| 12 |
st.markdown("Ask a question about the collapse of the Silicon Valley Bank (SVB).", unsafe_allow_html=True)
|
| 13 |
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
# st.session_state['pipelines_loaded'] = True
|
| 19 |
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
search_bar, button = st.columns([3, 1])
|
| 23 |
-
with search_bar:
|
| 24 |
-
username = st.text_area(f" ", max_chars=200, key='query')
|
| 25 |
-
|
| 26 |
-
with button:
|
| 27 |
-
st.write(" ")
|
| 28 |
-
st.write(" ")
|
| 29 |
-
run_pressed = st.button("Run")
|
| 30 |
-
|
| 31 |
-
st.markdown("<center> <h5> Example questions </h5> </center>", unsafe_allow_html=True)
|
| 32 |
-
|
| 33 |
-
st.write(" ")
|
| 34 |
-
st.write(" ")
|
| 35 |
-
c1, c2, c3, c4, c5 = st.columns(5)
|
| 36 |
-
with c1:
|
| 37 |
-
st.button(QUERIES[0], on_click=set_q1)
|
| 38 |
-
with c2:
|
| 39 |
-
st.button(QUERIES[1], on_click=set_q2)
|
| 40 |
-
with c3:
|
| 41 |
-
st.button(QUERIES[2], on_click=set_q3)
|
| 42 |
-
with c4:
|
| 43 |
-
st.button(QUERIES[3], on_click=set_q4)
|
| 44 |
-
with c5:
|
| 45 |
-
st.button(QUERIES[4], on_click=set_q5)
|
| 46 |
-
|
| 47 |
-
st.write(" ")
|
| 48 |
-
st.radio("Answer Type:", ("Retrieval Augmented (Static news dataset)", "Retrieval Augmented with Web Search"), key="query_type")
|
| 49 |
-
|
| 50 |
-
# st.sidebar.selectbox(
|
| 51 |
-
# "Example Questions:",
|
| 52 |
-
# QUERIES,
|
| 53 |
-
# key='q_drop_down', on_change=set_question)
|
| 54 |
-
|
| 55 |
-
st.markdown(f"<h5> {PLAIN_GPT_ANS} </h5>", unsafe_allow_html=True)
|
| 56 |
-
placeholder_plain_gpt = st.empty()
|
| 57 |
-
st.text(" ")
|
| 58 |
-
st.text(" ")
|
| 59 |
-
if st.session_state.get("query_type", "Retrieval Augmented (Static news dataset)") == "Retrieval Augmented (Static news dataset)":
|
| 60 |
-
st.markdown(f"<h5> {GPT_LOCAL_RET_AUG_ANS} </h5>", unsafe_allow_html=True)
|
| 61 |
-
else:
|
| 62 |
-
st.markdown(f"<h5>{GPT_WEB_RET_AUG_ANS} </h5>", unsafe_allow_html=True)
|
| 63 |
-
placeholder_retrieval_augmented = st.empty()
|
| 64 |
|
| 65 |
if st.session_state.get('query') and run_pressed:
|
| 66 |
ip = st.session_state['query']
|
|
|
|
| 1 |
import streamlit as st
|
| 2 |
+
from utils.backend import (get_plain_pipeline, get_retrieval_augmented_pipeline,
|
| 3 |
+
get_web_retrieval_augmented_pipeline)
|
| 4 |
+
from utils.ui import set_q1, set_q2, set_q3, set_q4, set_q5, left_sidebar, right_sidebar, main_column
|
| 5 |
+
from utils.constants import QUERIES, PLAIN_GPT_ANS, GPT_WEB_RET_AUG_ANS, GPT_LOCAL_RET_AUG_ANS
|
| 6 |
|
| 7 |
st.set_page_config(
|
| 8 |
page_title="Retrieval Augmentation with Haystack",
|
| 9 |
+
layout="wide"
|
| 10 |
)
|
| 11 |
+
left_sidebar()
|
| 12 |
|
| 13 |
st.markdown("<center> <h2> Reduce Hallucinations with Retrieval Augmentation </h2> </center>", unsafe_allow_html=True)
|
| 14 |
|
| 15 |
st.markdown("Ask a question about the collapse of the Silicon Valley Bank (SVB).", unsafe_allow_html=True)
|
| 16 |
|
| 17 |
+
col_1, col_2 = st.columns([4, 2], gap="small")
|
| 18 |
+
with col_1:
|
| 19 |
+
run_pressed, placeholder_plain_gpt, placeholder_retrieval_augmented = main_column()
|
| 20 |
+
print(f"Run value: {st.session_state.get('run', 'not found')}")
|
|
|
|
| 21 |
|
| 22 |
+
with col_2:
|
| 23 |
+
right_sidebar()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
|
| 25 |
if st.session_state.get('query') and run_pressed:
|
| 26 |
ip = st.session_state['query']
|
backend_utils.py
DELETED
|
@@ -1,107 +0,0 @@
|
|
| 1 |
-
import streamlit as st
|
| 2 |
-
from haystack import Pipeline
|
| 3 |
-
from haystack.document_stores import FAISSDocumentStore
|
| 4 |
-
from haystack.nodes import Shaper, PromptNode, PromptTemplate, PromptModel, EmbeddingRetriever
|
| 5 |
-
from haystack.nodes.retriever.web import WebRetriever
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
QUERIES = [
|
| 9 |
-
"Did SVB collapse?",
|
| 10 |
-
"Why did SVB collapse?",
|
| 11 |
-
"What does SVB failure mean for our economy?",
|
| 12 |
-
"Who is responsible for SVC collapse?",
|
| 13 |
-
"When did SVB collapse?"
|
| 14 |
-
]
|
| 15 |
-
PLAIN_GPT_ANS = "Answer with plain GPT"
|
| 16 |
-
GPT_LOCAL_RET_AUG_ANS = "Answer with Retrieval Augmented GPT (Static news dataset)"
|
| 17 |
-
GPT_WEB_RET_AUG_ANS = "Answer with Retrieval Augmented GPT (Web Search)"
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
@st.cache_resource(show_spinner=False)
|
| 21 |
-
def get_plain_pipeline():
|
| 22 |
-
prompt_open_ai = PromptModel(model_name_or_path="text-davinci-003", api_key=st.secrets["OPENAI_API_KEY"])
|
| 23 |
-
# Now let make one PromptNode use the default model and the other one the OpenAI model:
|
| 24 |
-
plain_llm_template = PromptTemplate(name="plain_llm", prompt_text="Answer the following question: $query")
|
| 25 |
-
node_openai = PromptNode(prompt_open_ai, default_prompt_template=plain_llm_template, max_length=300)
|
| 26 |
-
pipeline = Pipeline()
|
| 27 |
-
pipeline.add_node(component=node_openai, name="prompt_node", inputs=["Query"])
|
| 28 |
-
return pipeline
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
@st.cache_resource(show_spinner=False)
|
| 32 |
-
def get_retrieval_augmented_pipeline():
|
| 33 |
-
ds = FAISSDocumentStore(faiss_index_path="data/my_faiss_index.faiss",
|
| 34 |
-
faiss_config_path="data/my_faiss_index.json")
|
| 35 |
-
|
| 36 |
-
retriever = EmbeddingRetriever(
|
| 37 |
-
document_store=ds,
|
| 38 |
-
embedding_model="sentence-transformers/multi-qa-mpnet-base-dot-v1",
|
| 39 |
-
model_format="sentence_transformers",
|
| 40 |
-
top_k=2
|
| 41 |
-
)
|
| 42 |
-
shaper = Shaper(func="join_documents", inputs={"documents": "documents"}, outputs=["documents"])
|
| 43 |
-
|
| 44 |
-
default_template = PromptTemplate(
|
| 45 |
-
name="question-answering",
|
| 46 |
-
prompt_text="Given the context please answer the question. Context: $documents; Question: "
|
| 47 |
-
"$query; Answer:",
|
| 48 |
-
)
|
| 49 |
-
# Let's initiate the PromptNode
|
| 50 |
-
node = PromptNode("text-davinci-003", default_prompt_template=default_template,
|
| 51 |
-
api_key=st.secrets["OPENAI_API_KEY"], max_length=500)
|
| 52 |
-
|
| 53 |
-
# Let's create a pipeline with Shaper and PromptNode
|
| 54 |
-
pipeline = Pipeline()
|
| 55 |
-
pipeline.add_node(component=retriever, name='retriever', inputs=['Query'])
|
| 56 |
-
pipeline.add_node(component=shaper, name="shaper", inputs=["retriever"])
|
| 57 |
-
pipeline.add_node(component=node, name="prompt_node", inputs=["shaper"])
|
| 58 |
-
return pipeline
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
@st.cache_resource(show_spinner=False)
|
| 62 |
-
def get_web_retrieval_augmented_pipeline():
|
| 63 |
-
search_key = st.secrets["WEBRET_API_KEY"]
|
| 64 |
-
web_retriever = WebRetriever(api_key=search_key, search_engine_provider="SerperDev")
|
| 65 |
-
shaper = Shaper(func="join_documents", inputs={"documents": "documents"}, outputs=["documents"])
|
| 66 |
-
default_template = PromptTemplate(
|
| 67 |
-
name="question-answering",
|
| 68 |
-
prompt_text="Given the context please answer the question. Context: $documents; Question: "
|
| 69 |
-
"$query; Answer:",
|
| 70 |
-
)
|
| 71 |
-
# Let's initiate the PromptNode
|
| 72 |
-
node = PromptNode("text-davinci-003", default_prompt_template=default_template,
|
| 73 |
-
api_key=st.secrets["OPENAI_API_KEY"], max_length=500)
|
| 74 |
-
# Let's create a pipeline with Shaper and PromptNode
|
| 75 |
-
pipeline = Pipeline()
|
| 76 |
-
pipeline.add_node(component=web_retriever, name='retriever', inputs=['Query'])
|
| 77 |
-
pipeline.add_node(component=shaper, name="shaper", inputs=["retriever"])
|
| 78 |
-
pipeline.add_node(component=node, name="prompt_node", inputs=["shaper"])
|
| 79 |
-
return pipeline
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
if 'query' not in st.session_state:
|
| 83 |
-
st.session_state['query'] = ""
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
def set_question():
|
| 87 |
-
st.session_state['query'] = st.session_state['q_drop_down']
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
def set_q1():
|
| 91 |
-
st.session_state['query'] = QUERIES[0]
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
def set_q2():
|
| 95 |
-
st.session_state['query'] = QUERIES[1]
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
def set_q3():
|
| 99 |
-
st.session_state['query'] = QUERIES[2]
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
def set_q4():
|
| 103 |
-
st.session_state['query'] = QUERIES[3]
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
def set_q5():
|
| 107 |
-
st.session_state['query'] = QUERIES[4]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
utils/__init__.py
ADDED
|
File without changes
|
utils/constants.py
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
QUERIES = [
|
| 2 |
+
"Did SVB collapse?",
|
| 3 |
+
"Why did SVB collapse?",
|
| 4 |
+
"What does SVB failure mean for our economy?",
|
| 5 |
+
"Who is responsible for SVC collapse?",
|
| 6 |
+
"When did SVB collapse?"
|
| 7 |
+
]
|
| 8 |
+
PLAIN_GPT_ANS = "Answer with plain GPT"
|
| 9 |
+
GPT_LOCAL_RET_AUG_ANS = "Answer with Retrieval Augmented GPT (Static news dataset)"
|
| 10 |
+
GPT_WEB_RET_AUG_ANS = "Answer with Retrieval Augmented GPT (Web Search)"
|
utils/ui.py
ADDED
|
@@ -0,0 +1,114 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import streamlit as st
|
| 2 |
+
from PIL import Image
|
| 3 |
+
|
| 4 |
+
from .constants import QUERIES, PLAIN_GPT_ANS, GPT_WEB_RET_AUG_ANS, GPT_LOCAL_RET_AUG_ANS
|
| 5 |
+
|
| 6 |
+
|
| 7 |
+
def set_question():
|
| 8 |
+
st.session_state['query'] = st.session_state['q_drop_down']
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
def set_q1():
|
| 12 |
+
st.session_state['query'] = QUERIES[0]
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
def set_q2():
|
| 16 |
+
st.session_state['query'] = QUERIES[1]
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
def set_q3():
|
| 20 |
+
st.session_state['query'] = QUERIES[2]
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
def set_q4():
|
| 24 |
+
st.session_state['query'] = QUERIES[3]
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
def set_q5():
|
| 28 |
+
st.session_state['query'] = QUERIES[4]
|
| 29 |
+
|
| 30 |
+
def main_column():
|
| 31 |
+
placeholder = st.empty()
|
| 32 |
+
with placeholder:
|
| 33 |
+
search_bar, button = st.columns([3, 1])
|
| 34 |
+
with search_bar:
|
| 35 |
+
username = st.text_area(f" ", max_chars=200, key='query')
|
| 36 |
+
|
| 37 |
+
with button:
|
| 38 |
+
st.write(" ")
|
| 39 |
+
st.write(" ")
|
| 40 |
+
run_pressed = st.button("Run", key="run")
|
| 41 |
+
|
| 42 |
+
st.write(" ")
|
| 43 |
+
st.radio("Answer Type:", ("Retrieval Augmented (Static news dataset)", "Retrieval Augmented with Web Search"), key="query_type")
|
| 44 |
+
|
| 45 |
+
# st.sidebar.selectbox(
|
| 46 |
+
# "Example Questions:",
|
| 47 |
+
# QUERIES,
|
| 48 |
+
# key='q_drop_down', on_change=set_question)
|
| 49 |
+
|
| 50 |
+
st.markdown(f"<h5> {PLAIN_GPT_ANS} </h5>", unsafe_allow_html=True)
|
| 51 |
+
placeholder_plain_gpt = st.empty()
|
| 52 |
+
st.text(" ")
|
| 53 |
+
st.text(" ")
|
| 54 |
+
if st.session_state.get("query_type", "Retrieval Augmented (Static news dataset)") == "Retrieval Augmented (Static news dataset)":
|
| 55 |
+
st.markdown(f"<h5> {GPT_LOCAL_RET_AUG_ANS} </h5>", unsafe_allow_html=True)
|
| 56 |
+
else:
|
| 57 |
+
st.markdown(f"<h5>{GPT_WEB_RET_AUG_ANS} </h5>", unsafe_allow_html=True)
|
| 58 |
+
placeholder_retrieval_augmented = st.empty()
|
| 59 |
+
|
| 60 |
+
return run_pressed, placeholder_plain_gpt, placeholder_retrieval_augmented
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
def right_sidebar():
|
| 64 |
+
st.markdown("<h5> Example questions </h5>", unsafe_allow_html=True)
|
| 65 |
+
# c1, c2, c3, c4, c5 = st.columns(5)
|
| 66 |
+
# with c1:
|
| 67 |
+
st.button(QUERIES[0], on_click=set_q1)
|
| 68 |
+
# with c2:
|
| 69 |
+
st.button(QUERIES[1], on_click=set_q2)
|
| 70 |
+
# with c3:
|
| 71 |
+
st.button(QUERIES[2], on_click=set_q3)
|
| 72 |
+
# with c4:
|
| 73 |
+
st.button(QUERIES[3], on_click=set_q4)
|
| 74 |
+
# with c5:
|
| 75 |
+
st.button(QUERIES[4], on_click=set_q5)
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
def left_sidebar():
|
| 79 |
+
with st.sidebar:
|
| 80 |
+
image = Image.open('logo/haystack-logo-colored.png')
|
| 81 |
+
st.markdown("Thanks for coming to this π€ Space.\n\n"
|
| 82 |
+
"This is an effort towards showcasing how can you use Haystack for Retrieval Augmented QA, "
|
| 83 |
+
"with local document store as well as WebRetriever (coming soon!) \n\n"
|
| 84 |
+
"For more on how this was built, instructions along with a Repository "
|
| 85 |
+
"will be published soon and updated here.")
|
| 86 |
+
|
| 87 |
+
# st.markdown(
|
| 88 |
+
# "## How to use\n"
|
| 89 |
+
# "1. Enter your [OpenAI API key](https://platform.openai.com/account/api-keys) below\n"
|
| 90 |
+
# "2. Enter a Serper Dev API key\n"
|
| 91 |
+
# "3. Enjoy π€\n"
|
| 92 |
+
# )
|
| 93 |
+
|
| 94 |
+
# api_key_input = st.text_input(
|
| 95 |
+
# "OpenAI API Key",
|
| 96 |
+
# type="password",
|
| 97 |
+
# placeholder="Paste your OpenAI API key here (sk-...)",
|
| 98 |
+
# help="You can get your API key from https://platform.openai.com/account/api-keys.",
|
| 99 |
+
# value=st.session_state.get("OPENAI_API_KEY", ""),
|
| 100 |
+
# )
|
| 101 |
+
|
| 102 |
+
# if api_key_input:
|
| 103 |
+
# set_openai_api_key(api_key_input)
|
| 104 |
+
|
| 105 |
+
st.markdown("---")
|
| 106 |
+
st.markdown(
|
| 107 |
+
"## How this works\n"
|
| 108 |
+
"This app was built with [Haystack](https://haystack.deepset.ai) using the"
|
| 109 |
+
" [`PromptNode`](https://docs.haystack.deepset.ai/docs/prompt_node) and [`Retriever`](https://docs.haystack.deepset.ai/docs/retriever#embedding-retrieval-recommended).\n\n"
|
| 110 |
+
" You can find the source code in **Files and versions** tab."
|
| 111 |
+
)
|
| 112 |
+
|
| 113 |
+
st.markdown("---")
|
| 114 |
+
st.image(image, width=250)
|