Prompt API
The promptStreaming() method of the LanguageModel interface sends input to the language model and returns a ReadableStream that delivers the model's response incrementally as it is generated.
This is useful for displaying responses to users incrementally for outputs that take a long time to complete, or for any scenario where perceived latency should be minimized. Consume the stream using for await...of or by attaching a reader via ReadableStream.getReader().
Syntax
promptStreaming(input)
promptStreaming(input, options)
Parameters
input- : The content to prompt the model with. This is either:
- A string — Shorthand for a single textual message.
- An array of objects, each representing a single message in a conversation with a language model.
Objects may have the following properties:
role- : A string indicating the point of view the message is phrased from. Must be one of:
system- : A system-level instruction that guides the model's overall behavior. This must be the first instruction passed to the model.
user- : A message from the user, which the API should respond to.
assistant- : An input that provides context for the AI assistant, such as its persona or the format of its responses. Such messages mainly serve to provide context/history, and further shape how the model responds.
- : A string indicating the point of view the message is phrased from. Must be one of:
content- : A string representing a textual prompt, or an array of objects. Each object includes the following properties:
type- : An enumerated value representing the type of content. This can be one of:
audio- : Audio content.
image- : Image content.
text- : Textual content.
tool-call- : A tool invocation issued by the model.
tool-response- : The result of a tool invocation.
- : An enumerated value representing the type of content. This can be one of:
value- : The content of the message. If the
typeistext, this is always a string. If thetypeisaudioorimage, thevaluecan be one of several different object types; see What data types are accepted?.
- : The content of the message. If the
- : A string representing a textual prompt, or an array of objects. Each object includes the following properties:
prefix- : A boolean, defaulting to
false. Whentrue, the message is treated as a prefix for the model's next generated response rather than a complete turn.
- : A boolean, defaulting to
- : The content to prompt the model with. This is either:
options- : Options for creating a prompt. Properties include:
responseConstraint- : An object following the structure defined by JSON Schema defining the precise format the model's output should be delivered in. When provided and
omitResponseConstraintInputisfalse, any implementation-defined constraint-description message is included in the measurement.
- : An object following the structure defined by JSON Schema defining the precise format the model's output should be delivered in. When provided and
omitResponseConstraintInput- : A boolean; when
true, the automatic constraint-description message is excluded from the measurement.
- : A boolean; when
signal- : An AbortSignal to cancel the operation.
- : Options for creating a prompt. Properties include:
Return value
A ReadableStream of String chunks. Each chunk represents a piece of the model's response as it is generated. The stream closes when generation completes.
Exceptions
Errors are surfaced as stream errors rather than as rejected promises. Consumers should handle errors using a stream's standard error-handling mechanisms.
AbortErrorDOMException- : Thrown if the operation was canceled via the
signaloption.
- : Thrown if the operation was canceled via the
NotAllowedErrorDOMException- : Thrown if usage of the method is blocked by a Permissions-Policy/language-model Permissions-Policy.
NotSupportedErrorDOMException- : Thrown if:
- A message's
roleisassistantand itstypeis anything other thantext. - A message's
typeistextand itsvalueis not a string. - The input or output text is in a language the user agent doesn't support for prompting.
- A message's
typeisimageoraudiobut the type was not listed inexpectedInputs, or thevalueis not an accepted data type.
- A message's
- : Thrown if:
OperationErrorDOMException- : Thrown if the prompt fails for any other reason not listed in the other exception types.
QuotaExceededErrorDOMException- : Thrown if the prompt would cause the session's context usage to exceed the model's LanguageModel.contextWindow.
SyntaxErrorDOMException- : Thrown if:
- No messages are included in the messages array.
- A message's
prefixproperty is set totrueand:- The message's
roleis notassistant. - The message is not the last item in the messages array.
- The message's
- : Thrown if:
- TypeError
- : Thrown if:
omitResponseConstraintInputistruebutresponseConstraintis not provided.- A message's
roleissystembut it was not the first message passed to the context.
- : Thrown if:
Description
The promptStreaming() method adds the provided input to the context window and generates a response. The entire response is receives incrementally as a ReadableStream.
To receive the response as one complete string, use LanguageModel.prompt() instead. To add content to the context window without generating a response, use LanguageModel.append().
Each call to promptStreaming() adds to the session's context. To branch from a given state without affecting the original session, call LanguageModel.clone().
Examples
Streaming a response to the page
This example writes out chunks from a promptStreaming() call's ReadableStream as they arrive.
const session = await LanguageModel.create();
const output = document.querySelector("#output");
const stream = session.promptStreaming("Write a short poem about the ocean.");
for await (const chunk of stream) {
output.textContent += chunk;
}
See also Using the Prompt API > Complete streaming example.
Streaming with an abort signal
This example shows how to use an AbortController with promptStreaming().
const controller = new AbortController();
document
.querySelector("#stop")
.addEventListener("click", () => controller.abort());
const stream = session.promptStreaming("Tell me a long story.", {
signal: controller.signal,
});
try {
for await (const chunk of stream) {
output.textContent += chunk;
}
} catch (err) {
if (err.name === "AbortError") {
console.log("Streaming was stopped by the user.");
}
}
Collecting streamed chunks into a single string
In this example, chunks from a ReadableStream are collected before the whole stream is written out.
const session = await LanguageModel.create();
const stream = session.promptStreaming("Explain quantum entanglement.");
const chunks = [];
for await (const chunk of stream) {
chunks.push(chunk);
}
const fullResponse = chunks.join("");
console.log(fullResponse);
Specifications
Browser compatibility
See also
- LanguageModel.prompt()
- ReadableStream
- Prompt API
- Using the Prompt API
- Adding context with initial and ongoing prompt inputs
来源:MDN Web Docs / Mozilla 贡献者 · 作者与贡献记录 · CC BY-SA 2.5。离线整理了版式、链接和媒体显示。
