import express from "express"
import { HfInference } from '@huggingface/inference'
import { daisy } from "./daisy.mts"
const hfi = new HfInference(process.env.HF_API_TOKEN)
const hf = hfi.endpoint(process.env.HF_ENDPOINT_URL)
const app = express()
const port = 7860
const minPromptSize = 16 // if you change this, you will need to also change in public/index.html
const timeoutInSec = 30 * 60
console.log("timeout set to 30 minutes")
app.use(express.static("public"))
const pending: {
total: number;
queue: string[];
} = {
total: 0,
queue: [],
}
const endRequest = (id: string, reason: string) => {
if (!id || !pending.queue.includes(id)) {
return
}
pending.queue = pending.queue.filter(i => i !== id)
console.log(`request ${id} ended (${reason})`)
}
app.get("/debug", (req, res) => {
res.write(JSON.stringify({
nbTotal: pending.total,
nbPending: pending.queue.length,
queue: pending.queue,
}))
res.end()
})
app.get("/app", async (req, res) => {
const model = `${req.query.model || 'OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5'}`
console.log('model:', model)
const endpoint = `${req.query.endpoint || ''}`
console.log('endpoint:', endpoint)
if (`${req.query.prompt}`.length < minPromptSize) {
res.write(`prompt too short, please enter at least ${minPromptSize} characters`)
res.end()
return
}
const id = `${pending.total++}`
console.log(`new request ${id}`)
pending.queue.push(id)
const prefix = `Generated content {
endRequest(id, `timed out after ${timeoutInSec}s`)
}, timeoutInSec * 1000)
const finalPrompt = `# Task
Generate the following: ${req.query.prompt}
# API Documentation
${daisy}
# Guidelines
- Never repeat the instruction, instead directly write the final code
- Use a color scheme consistent with the brief and theme
- You must use Tailwind CSS and Daisy UI for the CSS classes, vanilla JS and Alpine.js for the JS.
- All the JS code will be written directly inside the page, using
- You MUST use English, not Latin! (I repeat: do NOT write lorem ipsum!)
- No need to write code comments, so please make the code compact (short function names etc)
- Use a central layout by wrapping everything in a \`
\`
# HTML output
')) {
break
}
if (result.includes('<|end|>') || result.includes('<|assistant|>')) {
// it ended, but we probably don't have a valid HTML
break
}
}
endRequest(id, `normal end of the LLM stream for request ${id}`)
} catch (e) {
console.log(e)
endRequest(id, `premature end of the LLM stream for request ${id} (${e})`)
}
try {
res.end()
} catch (err) {
console.log(`couldn't end the HTTP stream for request ${id} (${err})`)
}
})
app.listen(port, () => { console.log(`Open http://localhost:${port}`) })