Spaces:
Sleeping
Sleeping
John Graham Reynolds
commited on
Commit
·
4abddf8
1
Parent(s):
f45b463
try to change path to css file and add newer, non-experimental decorator for caching
Browse files
app.py
CHANGED
@@ -31,12 +31,13 @@ EXAMPLE_PROMPTS = [
|
|
31 |
]
|
32 |
|
33 |
TITLE = "VUMC Chatbot"
|
34 |
-
DESCRIPTION="""Welcome to the first generation Vanderbilt AI assistant! This AI assistant is built atop the Databricks DBRX large language model
|
35 |
and is augmented with additional organization-specific knowledge. Specifically, it has been preliminarily augmented with knowledge of Vanderbilt University Medical Center
|
36 |
terms like **Data Lake**, **EDW** (Enterprise Data Warehouse), **HCERA** (Health Care and Education Reconciliation Act), and **thousands more!** The model has **no access to PHI**.
|
37 |
Try querying the model with any of the examples prompts below for a simple introduction to both Vanderbilt-specific and general knowledge queries. The purpose of this
|
38 |
-
model is to allow VUMC employees access to an intelligent assistant that improves and expedites VUMC work.
|
39 |
-
Feedback and ideas are very welcome!
|
|
|
40 |
|
41 |
GENERAL_ERROR_MSG = "An error occurred. Please refresh the page to start a new conversation."
|
42 |
|
@@ -58,7 +59,7 @@ st.markdown(DESCRIPTION)
|
|
58 |
st.markdown("\n")
|
59 |
|
60 |
# use this to format later
|
61 |
-
with open("style.css") as css:
|
62 |
st.markdown( f'<style>{css.read()}</style>' , unsafe_allow_html= True)
|
63 |
|
64 |
if "messages" not in st.session_state:
|
@@ -80,7 +81,7 @@ def get_system_prompt():
|
|
80 |
# make sure we cache this so that it doesnt redownload each time, hindering Space start time if sleeping
|
81 |
# try adding this st caching decorator to ensure the embeddings class gets cached after downloading the entirety of the model
|
82 |
# does this cache to the given folder though? It does appear to populate the folder as expected after being run
|
83 |
-
@st.
|
84 |
def load_embedding_model():
|
85 |
embeddings = HuggingFaceEmbeddings(model_name="BAAI/bge-large-en", cache_folder="./langchain_cache/")
|
86 |
return embeddings
|
@@ -89,6 +90,7 @@ embeddings = load_embedding_model()
|
|
89 |
# instantiate the vector store for similarity search in our chain
|
90 |
# need to make this a function and decorate it with @st.experimental_memo as above?
|
91 |
# We are only calling this initially when the Space starts. Can we expedite this process for users when opening up this Space?
|
|
|
92 |
vector_store = DatabricksVectorSearch(
|
93 |
endpoint=VS_ENDPOINT_NAME,
|
94 |
index_name=VS_INDEX_NAME,
|
|
|
31 |
]
|
32 |
|
33 |
TITLE = "VUMC Chatbot"
|
34 |
+
DESCRIPTION="""Welcome to the first generation Vanderbilt AI assistant! \n This AI assistant is built atop the Databricks DBRX large language model
|
35 |
and is augmented with additional organization-specific knowledge. Specifically, it has been preliminarily augmented with knowledge of Vanderbilt University Medical Center
|
36 |
terms like **Data Lake**, **EDW** (Enterprise Data Warehouse), **HCERA** (Health Care and Education Reconciliation Act), and **thousands more!** The model has **no access to PHI**.
|
37 |
Try querying the model with any of the examples prompts below for a simple introduction to both Vanderbilt-specific and general knowledge queries. The purpose of this
|
38 |
+
model is to allow VUMC employees access to an intelligent assistant that improves and expedites VUMC work. \n
|
39 |
+
Feedback and ideas are very welcome! Please provide any feedback, ideas, or issues to the email: **[email protected]**.
|
40 |
+
We hope to gradually improve this AI assistant to create a large-scale, all-inclusive tool to compliment the work of all VUMC staff."""
|
41 |
|
42 |
GENERAL_ERROR_MSG = "An error occurred. Please refresh the page to start a new conversation."
|
43 |
|
|
|
59 |
st.markdown("\n")
|
60 |
|
61 |
# use this to format later
|
62 |
+
with open("./style.css") as css:
|
63 |
st.markdown( f'<style>{css.read()}</style>' , unsafe_allow_html= True)
|
64 |
|
65 |
if "messages" not in st.session_state:
|
|
|
81 |
# make sure we cache this so that it doesnt redownload each time, hindering Space start time if sleeping
|
82 |
# try adding this st caching decorator to ensure the embeddings class gets cached after downloading the entirety of the model
|
83 |
# does this cache to the given folder though? It does appear to populate the folder as expected after being run
|
84 |
+
@st.cache_data # will this work here?
|
85 |
def load_embedding_model():
|
86 |
embeddings = HuggingFaceEmbeddings(model_name="BAAI/bge-large-en", cache_folder="./langchain_cache/")
|
87 |
return embeddings
|
|
|
90 |
# instantiate the vector store for similarity search in our chain
|
91 |
# need to make this a function and decorate it with @st.experimental_memo as above?
|
92 |
# We are only calling this initially when the Space starts. Can we expedite this process for users when opening up this Space?
|
93 |
+
# @st.cache_data # TODO add this in
|
94 |
vector_store = DatabricksVectorSearch(
|
95 |
endpoint=VS_ENDPOINT_NAME,
|
96 |
index_name=VS_INDEX_NAME,
|