Spaces:
Sleeping
Sleeping
File size: 5,746 Bytes
4ff2d98 a7d8a51 4ff2d98 a7d8a51 4ff2d98 a7d8a51 4ff2d98 a7d8a51 4ff2d98 23e7cbd 4ff2d98 4b0268e 4ff2d98 23e7cbd 4ff2d98 a7d8a51 4ff2d98 23e7cbd a7d8a51 4ff2d98 23e7cbd 4ff2d98 a7d8a51 23e7cbd a7d8a51 23e7cbd |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 |
import streamlit as st
from langchain import memory as lc_memory
from langsmith import Client
from streamlit_feedback import streamlit_feedback
from utils import get_expression_chain, retriever, get_embeddings, create_qdrant_collection
from langchain_core.tracers.context import collect_runs
from qdrant_client import QdrantClient
from dotenv import load_dotenv
import os
load_dotenv()
client = Client()
qdrant_api=os.getenv("QDRANT_API_KEY")
qdrant_url=os.getenv("QDRANT_URL")
qdrant_client = QdrantClient(qdrant_url ,api_key=qdrant_api)
st.set_page_config(page_title = "SUP'ASSISTANT")
st.subheader("Hey there! How can I help you today!")
memory = lc_memory.ConversationBufferMemory(
chat_memory=lc_memory.StreamlitChatMessageHistory(key="langchain_messages"),
return_messages=True,
memory_key="chat_history",
)
st.sidebar.markdown("## Feedback Scale")
feedback_option = (
"thumbs" if st.sidebar.toggle(label="`Faces` ⇄ `Thumbs`", value=False) else "faces"
)
with st.sidebar:
model_name = st.selectbox("**Model**", options=["llama-3.1-70b-versatile","gemma2-9b-it","gemma-7b-it","llama-3.2-3b-preview", "llama3-70b-8192", "mixtral-8x7b-32768"])
temp = st.slider("**Temperature**", min_value=0.0, max_value=1.0, step=0.001)
n_docs = st.number_input("**Number of retireved documents**", min_value=0, max_value=10, value=5, step=1)
if st.sidebar.button("Clear message history"):
print("Clearing message history")
memory.clear()
retriever = retriever(n_docs=n_docs)
# Create Chain
chain = get_expression_chain(retriever,model_name,temp)
for msg in st.session_state.langchain_messages:
avatar = "🦜" if msg.type == "ai" else None
with st.chat_message(msg.type, avatar=avatar):
st.markdown(msg.content)
prompt = st.chat_input(placeholder="What do you need to know about SUP'COM ?")
if prompt :
with st.chat_message("user"):
st.write(prompt)
with st.chat_message("assistant", avatar="🦜"):
message_placeholder = st.empty()
full_response = ""
# Define the basic input structure for the chains
input_dict = {"input": prompt.lower()}
with collect_runs() as cb:
for chunk in chain.stream(input_dict, config={"tags": ["SUP'ASSISTANT"]}):
full_response += chunk.content
message_placeholder.markdown(full_response + "▌")
memory.save_context(input_dict, {"output": full_response})
st.session_state.run_id = cb.traced_runs[0].id
message_placeholder.markdown(full_response)
with st.spinner("Just a sec! Dont enter prompts while loading pelase!"):
run_id = st.session_state.run_id
question_embedding = get_embeddings(prompt)
answer_embedding = get_embeddings(full_response)
# Add question and answer to Qdrant
qdrant_client.upload_collection(
collection_name="chat-history",
payload=[
{"text": prompt, "type": "question", "question_ID": run_id},
{"text": full_response, "type": "answer", "question_ID": run_id}
],
vectors=[
question_embedding,
answer_embedding,
],
parallel=4,
max_retries=3,
)
if st.session_state.get("run_id"):
run_id = st.session_state.run_id
feedback = streamlit_feedback(
feedback_type=feedback_option,
optional_text_label="[Optional] Please provide an explanation",
key=f"feedback_{run_id}",
)
# Define score mappings for both "thumbs" and "faces" feedback systems
score_mappings = {
"thumbs": {"👍": 1, "👎": 0},
"faces": {"😀": 1, "🙂": 0.75, "😐": 0.5, "🙁": 0.25, "😞": 0},
}
# Get the score mapping based on the selected feedback option
scores = score_mappings[feedback_option]
if feedback:
# Get the score from the selected feedback option's score mapping
score = scores.get(feedback["score"])
if score is not None:
# Formulate feedback type string incorporating the feedback option
# and score value
feedback_type_str = f"{feedback_option} {feedback['score']}"
# Record the feedback with the formulated feedback type string
# and optional comment
with st.spinner("Just a sec! Dont enter prompts while loading pelase!"):
feedback_record = client.create_feedback(
run_id,
feedback_type_str,
score=score,
comment=feedback.get("text"),
)
st.session_state.feedback = {
"feedback_id": str(feedback_record.id),
"score": score,
}
else:
st.warning("Invalid feedback score.")
with st.spinner("Just a sec! Dont enter prompts while loading pelase!"):
if feedback.get("text"):
comment = feedback.get("text")
feedback_embedding = get_embeddings(comment)
else:
comment = "no comment"
feedback_embedding = get_embeddings(comment)
qdrant_client.upload_collection(
collection_name="chat-history",
payload=[
{"text": comment,"Score:":score, "type": "feedback", "question_ID": run_id}
],
vectors=[
feedback_embedding
],
parallel=4,
max_retries=3,
)
|