首页 > web前端 > js教程 > Search And (LLM) Transform

Search And (LLM) Transform

DDD
发布: 2024-09-19 06:22:32
原创
1016 人浏览过

Search And (LLM) Transform

This article shows an evolution of the "search and replace" feature from text editor,
where the "replace" step has been replaced by a LLM transformation.
The example is using GenAISCript.

It can be useful to batch apply text transformations that are not easily done with
regular expressions.

For example, when we added the ability to use a string command in
the exec command, we needed to convert all calls using argument arrays to this new syntax:

host.exec("cmd", ["arg0", "arg1", "arg2"])
登录后复制

to

host.exec(`cmd arg0 arg1 arg2`)`
登录后复制

While it's possible to match this function call with a regular expression

host\.exec\s*\([^,]+,\s*\[[^\]]+\]\s*\)
登录后复制

it's not easy to formulate the replacement string... unless you can describe it in natural language:

Convert the call to a single string command shell in TypeScript
登录后复制

Here are some example of the transformations where the LLM correctly handled variables.

  • concatenate the arguments of a function call into a single string
- const { stdout } = await host.exec("git", ["diff"])
+ const { stdout } = await host.exec(`git diff`)
登录后复制
  • concatenate the arguments and use the ${} syntax to interpolate variables
- const { stdout: commits } = await host.exec("git", [
-     "log",
-     "--author",
-     author,
-     "--until",
-     until,
-     "--format=oneline",
- ])
+ const { stdout: commits } = 
+   await host.exec(`git log --author ${author} --until ${until} --format=oneline`)
登录后复制

Search

The search step is done with the workspace.grep
that allows to efficiently search for a pattern in files (this is the same search engine
that powers the Visual Studio Code search).

const { pattern, glob } = env.vars
const patternRx = new RegExp(pattern, "g")
const { files } = await workspace.grep(patternRx, glob)
登录后复制

Compute Transforms

The second step is to apply the regular expression to the file content
and pre-compute the LLM transformation of each match using an inline prompt.

const { transform } = env.vars
...
const patches = {} // map of match -> transformed
for (const file of files) {
    const { content } = await workspace.readText(file.filename)
    for (const match of content.matchAll(patternRx)) {
        const res = await runPrompt(
            (ctx) => {
                ctx.$`
            ## Task

            Your task is to transform the MATCH with the following TRANSFORM.
            Return the transformed text.
            - do NOT add enclosing quotes.

            ## Context
            `
                ctx.def("MATCHED", match[0])
                ctx.def("TRANSFORM", transform)
            },
            { label: match[0], system: [], cache: "search-and-transform" }
        )
        ...
登录后复制

Since the LLM sometimes decides to wrap the answer in quotes, we need to remove them.

    ...
    const transformed = res.fences?.[0].content ?? res.text
    patches[match[0]] = transformed
登录后复制

Transform

Finally, with the transforms pre-computed, we apply a final regex replace to
patch the old file content with the transformed strings.

    const newContent = content.replace(
        patternRx,
        (match) => patches[match] ?? match
    )
    await workspace.writeText(file.filename, newContent)
}
登录后复制

Parameters

The script takes three parameters: a file glob, a pattern to search for, and a LLM transformation to apply.
We declare these parameters in the script metadata and extract them from the env.vars object.

script({ ...,
    parameters: {
        glob: {
            type: "string",
            description: "The glob pattern to filter files",
            default: "*",
        },
        pattern: {
            type: "string",
            description: "The text pattern (regular expression) to search for",
        },
        transform: {
            type: "string",
            description: "The LLM transformation to apply to the match",
        },
    },
})
const { pattern, glob, transform } = env.vars
登录后复制

Running

To run this script, you can use the --vars option to pass the pattern and the transform.

genaiscript run st --vars 'pattern=host\.exec\s*\([^,]+,\s*\[[^\]]+\]\s*\)' 'transform=Convert the call to a single string command shell in TypeScript'
登录后复制

以上是Search And (LLM) Transform的详细内容。更多信息请关注PHP中文网其他相关文章!

来源:dev.to
本站声明
本文内容由网友自发贡献,版权归原作者所有,本站不承担相应法律责任。如您发现有涉嫌抄袭侵权的内容,请联系admin@php.cn
热门教程
更多>
最新下载
更多>
网站特效
网站源码
网站素材
前端模板