Spaces:
Running
Running
File size: 4,303 Bytes
4e00df7 6128070 4e00df7 df66e31 4e00df7 6128070 4e00df7 cae23e1 6128070 033cc04 4e00df7 033cc04 4e00df7 6128070 033cc04 6128070 033cc04 6128070 4e00df7 033cc04 6128070 033cc04 cae23e1 033cc04 cae23e1 6128070 |
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 |
import anthropic
import streamlit as st
from streamlit.logger import get_logger
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
from langchain.llms import OpenAI
from langchain.llms import HuggingFaceEndpoint
from langchain.chat_models import ChatAnthropic
from langchain.vectorstores import SupabaseVectorStore
from stats import add_usage
memory = ConversationBufferMemory(memory_key="chat_history", input_key='question', output_key='answer', return_messages=True)
openai_api_key = st.secrets.openai_api_key
anthropic_api_key = st.secrets.anthropic_api_key
hf_api_key = st.secrets.hf_api_key
logger = get_logger(__name__)
def chat_with_doc(model, vector_store: SupabaseVectorStore, stats_db, stats):
if 'chat_history' not in st.session_state:
st.session_state['chat_history'] = []
query = st.text_area("## Ask a question (" + stats + " queries answered so far)", max_chars=500)
columns = st.columns(2)
with columns[0]:
button = st.button("Ask")
with columns[1]:
clear_history = st.button("Clear History", type='secondary')
st.markdown("---\n\n")
if clear_history:
# Clear memory in Langchain
memory.clear()
st.session_state['chat_history'] = []
st.experimental_rerun()
if button:
qa = None
add_usage(stats_db, "chat", "prompt" + query, {"model": model, "temperature": st.session_state['temperature']})
if model.startswith("gpt"):
logger.info('Using OpenAI model %s', model)
qa = ConversationalRetrievalChain.from_llm(
OpenAI(
model_name=st.session_state['model'], openai_api_key=openai_api_key, temperature=st.session_state['temperature'], max_tokens=st.session_state['max_tokens']), vector_store.as_retriever(), memory=memory, verbose=True)
elif anthropic_api_key and model.startswith("claude"):
logger.info('Using Anthropics model %s', model)
qa = ConversationalRetrievalChain.from_llm(
ChatAnthropic(
model=st.session_state['model'], anthropic_api_key=anthropic_api_key, temperature=st.session_state['temperature'], max_tokens_to_sample=st.session_state['max_tokens']), vector_store.as_retriever(), memory=memory, verbose=True, max_tokens_limit=102400)
elif hf_api_key:
logger.info('Using HF model %s', model)
# print(st.session_state['max_tokens'])
endpoint_url = ("https://api-inference.huggingface.co/models/"+ model)
model_kwargs = {"temperature" : st.session_state['temperature'],
"max_new_tokens" : st.session_state['max_tokens'],
"return_full_text" : False}
hf = HuggingFaceEndpoint(
endpoint_url=endpoint_url,
task="text-generation",
huggingfacehub_api_token=hf_api_key,
model_kwargs=model_kwargs
)
qa = ConversationalRetrievalChain.from_llm(hf, retriever=vector_store.as_retriever(search_kwargs={"score_threshold": 0.6, "k": 4,"filter": {"user": st.session_state["username"]}}), memory=memory, verbose=True, return_source_documents=True)
print("Question>")
print(query)
st.session_state['chat_history'].append(("You", query))
# Generate model's response and add it to chat history
model_response = qa({"question": query})
logger.info('Result: %s', model_response["answer"])
sources = model_response["source_documents"]
logger.info('Sources: %s', model_response["source_documents"])
if len(sources) > 0:
st.session_state['chat_history'].append(("Safety Copilot", model_response["answer"]))
else:
st.session_state['chat_history'].append(("Safety Copilot", "I am sorry, I do not have enough information to provide an answer. If there is a public source of data that you would like to add, please email [email protected]."))
# Display chat history
st.empty()
chat_history = st.session_state['chat_history']
for speaker, text in chat_history:
st.markdown(f"**{speaker}:** {text}") |