Use local LLMs in JS/TS/Node
LM Studio Client SDK - Pre-Release
lmstudio.js
is in pre-release alpha, and is undergoing rapid and continuous development. Expect breaking changes!
Follow along for our upcoming announcements about lmstudio.js
on Twitter and Discord. Read the Docs.
Discuss all things lmstudio.js in #dev-chat in LM Studio's Community Discord server.
npm install @lmstudio/sdk
npx lmstudio install-cli # open a new terminal window after installation...
lms create
import { LMStudioClient } from "@lmstudio/sdk";
const client = new LMStudioClient();
async function main() {
const modelPath = "lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF";
const llama3 = await client.llm.load(modelPath, { config: { gpuOffload: "max" } });
const prediction = llama3.respond([
{ role: "system", content: "Always answer in rhymes." },
{ role: "user", content: "Please introduce yourself." },
]);
for await (const text of prediction) {
process.stdout.write(text);
}
const { stats } = await prediction;
console.log(stats);
}
main();
lms
is the CLI tool for LM Studio. It is shipped with the latest versions of LM Studio. To set it up, run:
npx lmstudio install-cli
To check if the bootstrapping was successful, run the following in a ๐ new terminal window ๐:
lms
Note
lms
is only shipped with the latest version of LM Studio (v0.2.22 and onwards). Please make sure you have the latest version installed.
Start the server by running:
lms server start
If you are developing a web application and/or need to enable CORS (Cross Origin Resource Sharing), run this instead:
lms server start --cors=true
lms server start --port 12345
This example loads a model "lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF"
and predicts text with it.
import { LMStudioClient } from "@lmstudio/sdk";
const client = new LMStudioClient();
// Load a model
const llama3 = await client.llm.load("lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF");
// Create a text completion prediction
const prediction = llama3.complete("The meaning of life is");
// Stream the response
for await (const text of prediction) {
process.stdout.write(text);
}
Note
About process.stdout.write
process.stdout.write
is a Node.js-specific function that allows you to print text without a newline.
On the browser, you might want to do something like:
// Get the element where you want to display the output
const outputElement = document.getElementById("output");
for await (const text of prediction) {
outputElement.textContent += text;
}
This example shows how to connect to LM Studio running on a different port (e.g., 8080).
import { LMStudioClient } from "@lmstudio/sdk";
const client = new LMStudioClient({
baseUrl: "ws://127.0.0.1:8080",
});
// client.llm.load(...);
By default, when your client disconnects from LM Studio, all models loaded by that client are unloaded. You can prevent this by setting the noHup
option to true
.
await client.llm.load("lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF", {
noHup: true,
});
// The model stays loaded even after the client disconnects
You can set an identifier for a model when loading it. This identifier can be used to refer to the model later.
await client.llm.load("lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF", {
identifier: "my-model",
});
// You can refer to the model later using the identifier
const myModel = await client.llm.get("my-model");
// myModel.complete(...);
By default, the load configuration for a model comes from the preset associated with the model (Can be changed on the "My Models" page in LM Studio).
const llama3 = await client.llm.load("lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF", {
config: {
contextLength: 1024,
gpuOffload: 0.5, // Offloads 50% of the computation to the GPU
},
});
// llama3.complete(...);
The preset determines the default load configuration and the default inference configuration for a model. By default, the preset associated with the model is used. (Can be changed on the "My Models" page in LM Studio). You can change the preset used by specifying the preset
option.
const llama3 = await client.llm.load("lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF", {
preset: "My ChatML",
});
You can track the loading progress of a model by providing an onProgress
callback.
const llama3 = await client.llm.load("lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF", {
verbose: false, // Disables the default progress logging
onProgress: progress => {
console.log(`Progress: ${(progress * 100).toFixed(1)}%`);
},
});
If you wish to find all models that are available to be loaded, you can use the listDownloadedModel
method on the system
object.
const downloadedModels = await client.system.listDownloadedModels();
const downloadedLLMs = downloadedModels.filter(model => model.type === "llm");
// Load the first model
const model = await client.llm.load(downloadedLLMs[0].path);
// model.complete(...);
You can cancel a load by using an AbortController.
const controller = new AbortController();
try {
const llama3 = await client.llm.load("lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF", {
signal: controller.signal,
});
// llama3.complete(...);
} catch (error) {
console.error(error);
}
// Somewhere else in your code:
controller.abort();
Note
About AbortController
AbortController is a standard JavaScript API that allows you to cancel asynchronous operations. It is supported in modern browsers and Node.js. For more information, see the MDN Web Docs.
You can unload a model by calling the unload
method.
const llama3 = await client.llm.load("lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF", {
identifier: "my-model",
});
// ...Do stuff...
await client.llm.unload("my-model");
Note, by default, all models loaded by a client are unloaded when the client disconnects. Therefore, unless you want to precisely control the lifetime of a model, you do not need to unload them manually.
Note
Keeping a Model Loaded After Disconnection
If you wish to keep a model loaded after disconnection, you can set the noHup
option to true
when loading the model.
To look up an already loaded model by its identifier, use the following:
const myModel = await client.llm.get({ identifier: "my-model" });
// Or just
const myModel = await client.llm.get("my-model");
// myModel.complete(...);
To look up an already loaded model by its path, use the following:
// Matches any quantization
const llama3 = await client.llm.get({ path: "lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF" });
// Or if a specific quantization is desired:
const llama3 = await client.llm.get({
path: "lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF/Meta-Llama-3-8B-Instruct-Q4_K_M.gguf",
});
// llama3.complete(...);
If you do not have a specific model in mind, and just want to use any loaded model, you can simply pass in an empty object to client.llm.get
.
const anyModel = await client.llm.get({});
// anyModel.complete(...);
To list all loaded models, use the client.llm.listLoaded
method.
const loadedModels = await client.llm.listLoaded();
if (loadedModels.length === 0) {
throw new Error("No models loaded");
}
// Use the first one
const firstModel = await client.llm.get({ identifier: loadedModels[0].identifier });
// firstModel.complete(...);
Example loadedModels Response:
[ { "identifier": "lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF", "path": "lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF", }, { "identifier": "microsoft/Phi-3-mini-4k-instruct-gguf/Phi-3-mini-4k-instruct-q4.gguf", "path": "microsoft/Phi-3-mini-4k-instruct-gguf/Phi-3-mini-4k-instruct-q4.gguf", }, ]
To perform text completion, use the complete
method:
const prediction = model.complete("The meaning of life is");
for await (const text of prediction) {
process.stdout.write(text);
}
By default, the inference parameters in the preset is used for the prediction. You can override them like this:
const prediction = anyModel.complete("Meaning of life is", {
contextOverflowPolicy: "stopAtLimit",
maxPredictedTokens: 100,
prePrompt: "Some pre-prompt",
stopStrings: ["\n"],
temperature: 0.7,
});
// ...Do stuff with the prediction...
To perform a conversation, use the respond
method:
const prediction = anyModel.respond([
{ role: "system", content: "Answer the following questions." },
{ role: "user", content: "What is the meaning of life?" },
]);
for await (const text of prediction) {
process.stdout.write(text);
}
Similarly, you can override the inference parameters for the conversation (Note the available options are different from text completion):
const prediction = anyModel.respond(
[
{ role: "system", content: "Answer the following questions." },
{ role: "user", content: "What is the meaning of life?" },
],
{
contextOverflowPolicy: "stopAtLimit",
maxPredictedTokens: 100,
stopStrings: ["\n"],
temperature: 0.7,
inputPrefix: "Q: ",
inputSuffix: "\nA:",
},
);
// ...Do stuff with the prediction...
Important
Always Provide the Full History/Context
LLMs are stateless. They do not remember or retain information from previous inputs. Therefore, when predicting with an LLM, you should always provide the full history/context.
If you wish to get the prediction statistics, you can await on the prediction object to get a PredictionResult
, through which you can access the stats via the stats
property.
const prediction = model.complete("The meaning of life is");
for await (const text of prediction) {
process.stdout.write(text);
}
const { stats } = await prediction;
console.log(stats);
Note
No Extra Waiting
When you have already consumed the prediction stream, awaiting on the prediction object will not cause any extra waiting, as the result is cached within the prediction object.
On the other hand, if you only care about the final result, you don't need to iterate through the stream. Instead, you can await on the prediction object directly to get the final result.
const prediction = model.complete("The meaning of life is");
const result = await prediction;
const content = result.content;
const stats = result.stats;
// Or just:
const { content, stats } = await model.complete("The meaning of life is");
console.log(stats)
Example output for stats:
{ "stopReason": "eosFound", "tokensPerSecond": 26.644333102146646, "numGpuLayers": 33, "timeToFirstTokenSec": 0.146, "promptTokensCount": 5, "predictedTokensCount": 694, "totalTokensCount": 699 }
LM Studio supports structured prediction, which will force the model to produce content that conforms to a specific structure. To enable structured prediction, you should set the structured
field. It is available for both complete
and respond
methods.
Here is an example of how to use structured prediction:
const prediction = model.complete("Here is a joke in JSON:", {
maxPredictedTokens: 100,
structured: { type: "json" },
});
const result = await prediction;
try {
// Although the LLM is guaranteed to only produce valid JSON, when it is interrupted, the
// partial result might not be. Always check for errors. (See caveats below)
const parsed = JSON.parse(result.content);
console.info(parsed);
} catch (e) {
console.error(e);
}
Example output:
{ "title": "The Shawshank Redemption", "genre": [ "drama", "thriller" ], "release_year": 1994, "cast": [ { "name": "Tim Robbins", "role": "Andy Dufresne" }, { "name": "Morgan Freeman", "role": "Ellis Boyd" } ] }
Sometimes, any JSON is not enough. You might want to enforce a specific JSON schema. You can do this by providing a JSON schema to the structured
field. Read more about JSON schema at json-schema.org.
const bookSchema = {
type: "object",
properties: {
bookTitle: { type: "string" },
author: { type: "string" },
genre: { type: "string" },
pageCount: { type: "number" },
},
required: ["bookTitle", "author", "genre"],
};
const prediction = model.complete("Books that were turned into movies:", {
maxPredictedTokens: 100,
structured: { type: "json", jsonSchema: bookSchema },
});
const result = await prediction;
try {
const parsed = JSON.parse(result.content);
console.info(parsed); // see example response below
console.info("The bookTitle is", parsed.bookTitle); // The bookTitle is The Help
console.info("The author is", parsed.author); // The author is Tina
console.info("The genre is", parsed.genre); // The genre is Historical Fiction
console.info("The pageCount is", parsed.pageCount); // The pageCount is 320
} catch (e) {
console.error(e);
}
Example response for parsed:
{ "author": "J.K. Rowling", "bookTitle": "Harry Potter and the Philosopher's Stone", "genre": "Fantasy", "pageCount": 320 }
Important
Caveats with Structured Prediction
- Although the model is forced to generate predictions that conform to the specified structure, the prediction may be interrupted (for example, if the user stops the prediction). When that happens, the partial result may not conform to the specified structure. Thus, always check the prediction result before using it, for example, by wrapping the
JSON.parse
inside a try-catch block. - In certain cases, the model may get stuck. For example, when forcing it to generate valid JSON, it may generate a opening brace
{
but never generate a closing brace}
. In such cases, the prediction will go on forever until the context length is reached, which can take a long time. Therefore, it is recommended to always set amaxPredictedTokens
limit. This also contributes to the point above.
A prediction may be canceled by calling the cancel
method on the prediction object.
const prediction = model.complete("The meaning of life is");
// ...Do stuff...
prediction.cancel();
When a prediction is canceled, the prediction will stop normally but with stopReason
set to "userStopped"
. You can detect cancellation like so:
for await (const text of prediction) {
process.stdout.write(text);
}
const { stats } = await prediction;
if (stats.stopReason === "userStopped") {
console.log("Prediction was canceled by the user");
}