Introducing Gradio 5.0
Read MoreIntroducing Gradio 5.0
Read MoreNew to Gradio? Start here: Getting Started
See the Release History
To install Gradio from main, run the following command:
pip install https://gradio-builds.s3.amazonaws.com/ac9bf5ec9208b92579f36ee94a247ae3e676c02f/gradio-5.6.0-py3-none-any.whl
*Note: Setting share=True
in
launch()
will not work.
gradio.Chatbot(type="messages", ···)
The data format accepted by the Chatbot is dictated by the type
parameter.
This parameter can take two values, 'tuples'
and 'messages'
.
The 'tuples'
type is deprecated and will be removed in a future version of Gradio.
If the type
is 'messages'
, then the data sent to/from the chatbot will be a list of dictionaries
with role
and content
keys. This format is compliant with the format expected by most LLM APIs (HuggingChat, OpenAI, Claude).
The role
key is either 'user'
or 'assistant'
and the content
key can be one of the following:
path
and alt_text
keys. In this case, the file at path
will be displayed in the chat history. Image, audio, and video files are fully embedded and visualized in the chat bubble.
The path
key can point to a valid publicly available URL. The alt_text
key is optional but it’s good practice to provide alt text.def generate_response(history):
# A plain text response
history.append(
{"role": "assistant", content="I am happy to provide you that report and plot."}
)
# Embed the quaterly sales report in the chat
history.append(
{"role": "assistant", content={"path": "quaterly_sales.txt", "alt_text": "Sales Report for Q2 2024"}}
)
# Make a plot of sales data
history.append(
{"role": "assistant", content=gr.Plot(value=make_plot_from_file('quaterly_sales.txt'))}
)
return history
For convenience, you can use the ChatMessage
dataclass so that your text editor can give you autocomplete hints and typechecks.
from gradio import ChatMessage
def generate_response(history):
history.append(
ChatMessage(role="assistant",
content="How can I help you?")
)
return history
If type
is 'tuples'
, then the data sent to/from the chatbot will be a list of tuples.
The first element of each tuple is the user message and the second element is the bot’s response.
Each element can be a string (markdown/html is supported),
a tuple (in which case the first element is a filepath that will be displayed in the chatbot),
or a gradio component (see the Examples section for more details).
value: list[MessageDict | Message] | TupleFormat | Callable | None
= None
Default list of messages to show in chatbot, where each message is of the format {"role": "user", "content": "Help me."}. Role can be one of "user", "assistant", or "system". Content should be either text, or media passed as a Gradio component, e.g. {"content": gr.Image("lion.jpg")}. If callable, the function will be called whenever the app loads to set the initial value of the component.
type: Literal['messages', 'tuples'] | None
= None
The format of the messages passed into the chat history parameter of `fn`. If "messages", passes the value as a list of dictionaries with openai-style "role" and "content" keys. The "content" key's value should be one of the following - (1) strings in valid Markdown (2) a dictionary with a "path" key and value corresponding to the file to display or (3) an instance of a Gradio component. At the moment Image, Plot, Video, Gallery, Audio, and HTML are supported. The "role" key should be one of 'user' or 'assistant'. Any other roles will not be displayed in the output. If this parameter is 'tuples', expects a `list[list[str | None | tuple]]`, i.e. a list of lists. The inner list should have 2 elements: the user message and the response message, but this format is deprecated.
label: str | None
= None
the label for this component. Appears above the component and is also used as the header if there are a table of examples for this component. If None and used in a `gr.Interface`, the label will be the name of the parameter this component is assigned to.
every: Timer | float | None
= None
Continously calls `value` to recalculate it if `value` is a function (has no effect otherwise). Can provide a Timer whose tick resets `value`, or a float that provides the regular interval for the reset Timer.
inputs: Component | list[Component] | set[Component] | None
= None
Components that are used as inputs to calculate `value` if `value` is a function (has no effect otherwise). `value` is recalculated any time the inputs change.
show_label: bool | None
= None
if True, will display label.
container: bool
= True
If True, will place the component in a container - providing some extra padding around the border.
scale: int | None
= None
relative size compared to adjacent Components. For example if Components A and B are in a Row, and A has scale=2, and B has scale=1, A will be twice as wide as B. Should be an integer. scale applies in Rows, and to top-level Components in Blocks where fill_height=True.
min_width: int
= 160
minimum pixel width, will wrap if not sufficient screen space to satisfy this value. If a certain scale value results in this Component being narrower than min_width, the min_width parameter will be respected first.
visible: bool
= True
If False, component will be hidden.
elem_id: str | None
= None
An optional string that is assigned as the id of this component in the HTML DOM. Can be used for targeting CSS styles.
elem_classes: list[str] | str | None
= None
An optional list of strings that are assigned as the classes of this component in the HTML DOM. Can be used for targeting CSS styles.
autoscroll: bool
= True
If True, will automatically scroll to the bottom of the textbox when the value changes, unless the user scrolls up. If False, will not scroll to the bottom of the textbox when the value changes.
render: bool
= True
If False, component will not render be rendered in the Blocks context. Should be used if the intention is to assign event listeners now but render the component later.
key: int | str | None
= None
if assigned, will be used to assume identity across a re-render. Components that have the same key across a re-render will have their value preserved.
height: int | str | None
= 400
The height of the component, specified in pixels if a number is passed, or in CSS units if a string is passed. If messages exceed the height, the component will scroll.
max_height: int | str | None
= None
The maximum height of the component, specified in pixels if a number is passed, or in CSS units if a string is passed. If messages exceed the height, the component will scroll. If messages are shorter than the height, the component will shrink to fit the content. Will not have any effect if `height` is set and is smaller than `max_height`.
min_height: int | str | None
= None
The minimum height of the component, specified in pixels if a number is passed, or in CSS units if a string is passed. If messages exceed the height, the component will expand to fit the content. Will not have any effect if `height` is set and is larger than `min_height`.
latex_delimiters: list[dict[str, str | bool]] | None
= None
A list of dicts of the form {"left": open delimiter (str), "right": close delimiter (str), "display": whether to display in newline (bool)} that will be used to render LaTeX expressions. If not provided, `latex_delimiters` is set to `[{ "left": "$$", "right": "$$", "display": True }]`, so only expressions enclosed in $$ delimiters will be rendered as LaTeX, and in a new line. Pass in an empty list to disable LaTeX rendering. For more information, see the [KaTeX documentation](https://katex.org/docs/autorender.html).
rtl: bool
= False
If True, sets the direction of the rendered text to right-to-left. Default is False, which renders text left-to-right.
show_share_button: bool | None
= None
If True, will show a share icon in the corner of the component that allows user to share outputs to Hugging Face Spaces Discussions. If False, icon does not appear. If set to None (default behavior), then the icon appears if this Gradio app is launched on Spaces, but not otherwise.
show_copy_button: bool
= False
If True, will show a copy button for each chatbot message.
avatar_images: tuple[str | Path | None, str | Path | None] | None
= None
Tuple of two avatar image paths or URLs for user and bot (in that order). Pass None for either the user or bot image to skip. Must be within the working directory of the Gradio app or an external URL.
sanitize_html: bool
= True
If False, will disable HTML sanitization for chatbot messages. This is not recommended, as it can lead to security vulnerabilities.
render_markdown: bool
= True
If False, will disable Markdown rendering for chatbot messages.
bubble_full_width: bool
= True
If False, the chat bubble will fit to the content of the message. If True (default), the chat bubble will be the full width of the component.
line_breaks: bool
= True
If True (default), will enable Github-flavored Markdown line breaks in chatbot messages. If False, single new lines will be ignored. Only applies if `render_markdown` is True.
layout: Literal['panel', 'bubble'] | None
= None
If "panel", will display the chatbot in a llm style layout. If "bubble", will display the chatbot with message bubbles, with the user and bot messages on alterating sides. Will default to "bubble".
placeholder: str | None
= None
a placeholder message to display in the chatbot when it is empty. Centered vertically and horizontally in the Chatbot. Supports Markdown and HTML. If None, no placeholder is displayed.
examples: list[ExampleMessage] | None
= None
A list of example messages to display in the chatbot before any user/assistant messages are shown. Each example should be a dictionary with an optional "text" key representing the message that should be populated in the Chatbot when clicked, an optional "files" key, whose value should be a list of files to populate in the Chatbot, an optional "icon" key, whose value should be a filepath or URL to an image to display in the example box, and an optional "display_text" key, whose value should be the text to display in the example box. If "display_text" is not provided, the value of "text" will be displayed.
show_copy_all_button: <class 'inspect._empty'>
= False
If True, will show a copy all button that copies all chatbot messages to the clipboard.
allow_file_downloads: <class 'inspect._empty'>
= True
If True, will show a download button for chatbot messages that contain media. Defaults to True.
Class | Interface String Shortcut | Initialization |
---|---|---|
| "chatbot" | Uses default values |
Displaying Thoughts/Tool Usage
When type
is messages
, you can provide additional metadata regarding any tools used to generate the response.
This is useful for displaying the thought process of LLM agents. For example,
def generate_response(history):
history.append(
ChatMessage(role="assistant",
content="The weather API says it is 20 degrees Celcius in New York.",
metadata={"title": "🛠️ Used tool Weather API"})
)
return history
Would be displayed as following:
You can also specify metadata with a plain python dictionary,
def generate_response(history):
history.append(
dict(role="assistant",
content="The weather API says it is 20 degrees Celcius in New York.",
metadata={"title": "🛠️ Used tool Weather API"})
)
return history
Using Gradio Components Inside gr.Chatbot
The Chatbot
component supports using many of the core Gradio components (such as gr.Image
, gr.Plot
, gr.Audio
, and gr.HTML
) inside of the chatbot. Simply include one of these components in your list of tuples. Here’s an example:
import gradio as gr
def load():
return [
("Here's an audio", gr.Audio("https://github.com/gradio-app/gradio/raw/main/test/test_files/audio_sample.wav")),
("Here's an video", gr.Video("https://github.com/gradio-app/gradio/raw/main/demo/video_component/files/world.mp4"))
]
with gr.Blocks() as demo:
chatbot = gr.Chatbot()
button = gr.Button("Load audio and video")
button.click(load, None, chatbot)
demo.launch()
import gradio as gr
import random
import time
with gr.Blocks() as demo:
chatbot = gr.Chatbot(type="messages")
msg = gr.Textbox()
clear = gr.ClearButton([msg, chatbot])
def respond(message, chat_history):
bot_message = random.choice(["How are you?", "Today is a great day", "I'm very hungry"])
chat_history.append({"role": "user", "content": message})
chat_history.append({"role": "assistant", "content": bot_message})
time.sleep(2)
return "", chat_history
msg.submit(respond, [msg, chatbot], [msg, chatbot])
if __name__ == "__main__":
demo.launch()
Event listeners allow you to respond to user interactions with the UI components you've defined in a Gradio Blocks app. When a user interacts with an element, such as changing a slider value or uploading an image, a function is called.
The Chatbot component supports the following event listeners. Each event listener takes the same parameters, which are listed in the Event Parameters table below.
Listener | Description |
---|---|
| Triggered when the value of the Chatbot changes either because of user input (e.g. a user types in a textbox) OR because of a function update (e.g. an image receives a value from the output of an event trigger). See |
| Event listener for when the user selects or deselects the Chatbot. Uses event data gradio.SelectData to carry |
| This listener is triggered when the user likes/dislikes from within the Chatbot. This event has EventData of type gradio.LikeData that carries information, accessible through LikeData.index and LikeData.value. See EventData documentation on how to use this event data. |
| This listener is triggered when the user clicks the retry button in the chatbot message. |
| This listener is triggered when the user clicks the undo button in the chatbot message. |
| This listener is triggered when the user clicks on an example from within the Chatbot. This event has SelectData of type gradio.SelectData that carries information, accessible through SelectData.index and SelectData.value. See SelectData documentation on how to use this event data. |
| This listener is triggered when the user clears the Chatbot using the clear button for the component. |
fn: Callable | None | Literal['decorator']
= "decorator"
the function to call when this event is triggered. Often a machine learning model's prediction function. Each parameter of the function corresponds to one input component, and the function should return a single value or a tuple of values, with each element in the tuple corresponding to one output component.
inputs: Component | BlockContext | list[Component | BlockContext] | Set[Component | BlockContext] | None
= None
List of gradio.components to use as inputs. If the function takes no inputs, this should be an empty list.
outputs: Component | BlockContext | list[Component | BlockContext] | Set[Component | BlockContext] | None
= None
List of gradio.components to use as outputs. If the function returns no outputs, this should be an empty list.
api_name: str | None | Literal[False]
= None
defines how the endpoint appears in the API docs. Can be a string, None, or False. If set to a string, the endpoint will be exposed in the API docs with the given name. If None (default), the name of the function will be used as the API endpoint. If False, the endpoint will not be exposed in the API docs and downstream apps (including those that `gr.load` this app) will not be able to use this event.
scroll_to_output: bool
= False
If True, will scroll to output component on completion
show_progress: Literal['full', 'minimal', 'hidden']
= "full"
how to show the progress animation while event is running: "full" shows a spinner which covers the output component area as well as a runtime display in the upper right corner, "minimal" only shows the runtime display, "hidden" shows no progress animation at all
queue: bool
= True
If True, will place the request on the queue, if the queue has been enabled. If False, will not put this event on the queue, even if the queue has been enabled. If None, will use the queue setting of the gradio app.
batch: bool
= False
If True, then the function should process a batch of inputs, meaning that it should accept a list of input values for each parameter. The lists should be of equal length (and be up to length `max_batch_size`). The function is then *required* to return a tuple of lists (even if there is only 1 output component), with each list in the tuple corresponding to one output component.
max_batch_size: int
= 4
Maximum number of inputs to batch together if this is called from the queue (only relevant if batch=True)
preprocess: bool
= True
If False, will not run preprocessing of component data before running 'fn' (e.g. leaving it as a base64 string if this method is called with the `Image` component).
postprocess: bool
= True
If False, will not run postprocessing of component data before returning 'fn' output to the browser.
cancels: dict[str, Any] | list[dict[str, Any]] | None
= None
A list of other events to cancel when this listener is triggered. For example, setting cancels=[click_event] will cancel the click_event, where click_event is the return value of another components .click method. Functions that have not yet run (or generators that are iterating) will be cancelled, but functions that are currently running will be allowed to finish.
trigger_mode: Literal['once', 'multiple', 'always_last'] | None
= None
If "once" (default for all events except `.change()`) would not allow any submissions while an event is pending. If set to "multiple", unlimited submissions are allowed while pending, and "always_last" (default for `.change()` and `.key_up()` events) would allow a second submission after the pending event is complete.
js: str | None
= None
Optional frontend js method to run before running 'fn'. Input arguments for js method are values of 'inputs' and 'outputs', return should be a list of values for output components.
concurrency_limit: int | None | Literal['default']
= "default"
If set, this is the maximum number of this event that can be running simultaneously. Can be set to None to mean no concurrency_limit (any number of this event can be running simultaneously). Set to "default" to use the default concurrency limit (defined by the `default_concurrency_limit` parameter in `Blocks.queue()`, which itself is 1 by default).
concurrency_id: str | None
= None
If set, this is the id of the concurrency group. Events with the same concurrency_id will be limited by the lowest set concurrency_limit.
show_api: bool
= True
whether to show this event in the "view API" page of the Gradio app, or in the ".view_api()" method of the Gradio clients. Unlike setting api_name to False, setting show_api to False will still allow downstream apps as well as the Clients to use this event. If fn is None, show_api will automatically be set to False.
time_limit: int | None
= None
stream_every: float
= 0.5
like_user_message: bool
= False