Generation

Let the language model generate text.

Optionally, constrain the output with a regex expression.

Generating and Capturing

const {
  captured: { joke },
} = await model
  .user("Tell me a joke")
  .assistant((m) => m.gen("joke", { maxTokens: 128 }))
  .run();

The captured text is available in the captured object.

console.log(joke);

` Why don't scientists trust atoms?

Because they make up everything!`

Stop sequence

const {
  captured: { oneLine },
} = await model
  .user(
    "Tell me a joke, answer with a single paragraph (line).",
  )
  .assistant((m) =>
    m.gen("oneLine", { maxTokens: 128, stop: "\n" }),
  )
  .run();

console.log(oneLine);

` Why did the tomato turn red? Because it saw the salad dressing!`

Regex Constraint

import { BULLET_LIST_REGEX } from "@lmscript/client/regex";

const {
  captured: { mdList },
} = await model
  .user(
    "Tell me a list of 5 jokes. Answer with a markdown list, where each item of the list has a joke, in a single line.",
  )
  .assistant((m) =>
    m.gen("mdList", {
      maxTokens: 128,
      stop: "\n\n",
      regex: BULLET_LIST_REGEX,
    }),
  )
  .run();

console.log(mdList);

`- Why did the tomato turn red? Because it saw the salad dressing!
- Why did the scarecrow win an award? Because he was outstanding in his field!
- Why don't scientists trust atoms? Because they make up everything!
- Why did the bicycle fall over? Because it was two-tired!
- Why did the chicken cross the playground? To get to the other slide!`

Generate without a name

const { captured: noNameCaptured } = await model
  .user("Tell me a joke.")
  .assistant((m) => m.gen({ maxTokens: 128 }))
  .run();

console.log(noNameCaptured);

{}

Generating with prefix

Call the .push method before the .gen method to add a prefix to the generated text.

Note that none of the backends implement token healing yet.

Therefore, you must be careful with prefixes. In general, being careful not to terminate a prefix with space should be enough.

const {
  captured: { sureJoke },
} = await model
  .user("Tell me a joke")
  .assistant((m) =>
    m.push("Sure.").gen("sureJoke", { maxTokens: 128 }),
  )
  .run();

The captured text is available in the captured object.

console.log(sureJoke);

` Here's a classic one for you:

Why did the tomato turn red?

Because it saw the salad dressing!

I hope that brought a smile to your face! Do you have any other requests or questions? I'm here to help.`

Generating without chat template

This example does not use the chat template. The result is not good, but it is just an example.

The .gen method can be used without chat templates, or inside a chat template role function.

const {
  captured: { continuation },
} = await model
  .push("Here is a small joke, in a single line:")
  .gen("continuation", { maxTokens: 32 })
  .run();

The captured text is available in the captured object.

console.log(continuation);

`

Why did the tomato turn red?

Because it saw the salad dressing!

This is a classic example of a pun, which is`