This page only contains information on the
st.cache_resource API. For a deeper dive into caching and how to use it, check out Caching.
Decorator to cache functions that return global resources (e.g. database connections, ML models).
Cached objects are shared across all users, sessions, and reruns. They must be thread-safe because they can be accessed from multiple threads concurrently. If thread safety is an issue, consider using st.session_state to store resources per session instead.
You can clear a function's cache with func.clear() or clear the entire cache with st.cache_resource.clear().
To cache data, use st.cache_data instead. Learn more about caching at https://docs.streamlit.io/library/advanced-features/caching.
st.cache_resource(func, *, ttl, max_entries, show_spinner, validate, experimental_allow_widgets)
The function that creates the cached resource. Streamlit hashes the function's source code.
ttl (float or timedelta or None)
The maximum number of seconds to keep an entry in the cache, or None if cache entries should not expire. The default is None.
max_entries (int or None)
The maximum number of entries to keep in the cache, or None for an unbounded cache. (When a new entry is added to a full cache, the oldest cached entry will be removed.) The default is None.
show_spinner (boolean or string)
Enable the spinner. Default is True to show a spinner when there is a "cache miss" and the cached resource is being created. If string, value of show_spinner param will be used for spinner text.
validate (callable or None)
An optional validation function for cached data. validate is called each time the cached value is accessed. It receives the cached value as its only parameter and it must return a boolean. If validate returns False, the current cached value is discarded, and the decorated function is called to compute a new value. This is useful e.g. to check the health of database connections.
Allow widgets to be used in the cached function. Defaults to False. Support for widgets in cached functions is currently experimental. Setting this parameter to True may lead to excessive memory use since the widget value is treated as an additional input parameter to the cache. We may remove support for this option at any time without notice.
import streamlit as st @st.cache_resource def get_database_session(url): # Create a database session object that points to the URL. return session s1 = get_database_session(SESSION_URL_1) # Actually executes the function, since this is the first time it was # encountered. s2 = get_database_session(SESSION_URL_1) # Does not execute the function. Instead, returns its previously computed # value. This means that now the connection object in s1 is the same as in s2. s3 = get_database_session(SESSION_URL_2) # This is a different URL, so the function executes.
By default, all parameters to a cache_resource function must be hashable. Any parameter whose name begins with _ will not be hashed. You can use this as an "escape hatch" for parameters that are not hashable:import streamlit as st @st.cache_resource def get_database_session(_sessionmaker, url): # Create a database connection object that points to the URL. return connection s1 = get_database_session(create_sessionmaker(), DATA_URL_1) # Actually executes the function, since this is the first time it was # encountered. s2 = get_database_session(create_sessionmaker(), DATA_URL_1) # Does not execute the function. Instead, returns its previously computed # value - even though the _sessionmaker parameter was different # in both calls.
A cache_resource function's cache can be procedurally cleared:import streamlit as st @st.cache_resource def get_database_session(_sessionmaker, url): # Create a database connection object that points to the URL. return connection get_database_session.clear() # Clear all cached entries for this function.
Using Streamlit commands in cached functions
Since version 1.16.0, cached functions can contain Streamlit commands! For example, you can do this:
from transformers import pipeline @st.cache_resource def load_model(): model = pipeline("sentiment-analysis") st.success("Loaded NLP model from Hugging Face!") # 👈 Show a success message return model
As we know, Streamlit only runs this function if it hasn’t been cached before. On this first run, the
st.success message will appear in the app. But what happens on subsequent runs? It still shows up! Streamlit realizes that there is an
st. command inside the cached function, saves it during the first run, and replays it on subsequent runs. Replaying static elements works for both caching decorators.
You can also use this functionality to cache entire parts of your UI:
@st.cache_resource def load_model(): st.header("Data analysis") model = torchvision.models.resnet50(weights=ResNet50_Weights.DEFAULT) st.success("Loaded model!") st.write("Turning on evaluation mode...") model.eval() st.write("Here's the model:") return model
You can also use interactive input widgets like
st.text_input in cached functions. Widget replay is an experimental feature at the moment. To enable it, you need to set the
@st.cache_data(experimental_allow_widgets=True) # 👈 Set the parameter def load_model(): pretrained = st.checkbox("Use pre-trained model:") # 👈 Add a checkbox model = torchvision.models.resnet50(weights=ResNet50_Weights.DEFAULT, pretrained=pretrained) return model
Streamlit treats the checkbox like an additional input parameter to the cached function. If you uncheck it, Streamlit will see if it has already cached the function for this checkbox state. If yes, it will return the cached value. If not, it will rerun the function using the new slider value.
Using widgets in cached functions is extremely powerful because it lets you cache entire parts of your app. But it can be dangerous! Since Streamlit treats the widget value as an additional input parameter, it can easily lead to excessive memory usage. Imagine your cached function has five sliders and returns a 100 MB DataFrame. Then we’ll add 100 MB to the cache for every permutation of these five slider values – even if the sliders do not influence the returned data! These additions can make your cache explode very quickly. Please be aware of this limitation if you use widgets in cached functions. We recommend using this feature only for isolated parts of your UI where the widgets directly influence the cached return value.
Support for widgets in cached functions is currently experimental. We may change or remove it anytime without warning. Please use it with care!
Two widgets are currently not supported in cached functions:
st.camera_input. We may support them in the future. Feel free to open a GitHub issue if you need them!