llama.cpp/examples/server/public/index.html
Tobias Lütke 31cfbb1013
Expose generation timings from server & update completions.js (#2116)
* use javascript generators as much cleaner API

Also add ways to access completion as promise and EventSource

* export llama_timings as struct and expose them in server

* update readme, update baked includes

* llama : uniform variable names + struct init

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-07-05 16:51:13 -04:00

381 lines
10 KiB
HTML

<html>
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1, maximum-scale=1" />
<title>llama.cpp - chat</title>
<style>
body {
background-color: #fff;
color: #000;
font-family: system-ui;
font-size: 90%;
}
#container {
margin: 0em auto;
display: flex;
flex-direction: column;
justify-content: space-between;
height: 100%;
}
main {
margin: 3px;
display: flex;
flex-direction: column;
justify-content: space-between;
gap: 1em;
flex-grow: 1;
overflow-y: auto;
border: 1px solid #ccc;
border-radius: 5px;
padding: 0.5em;
}
body {
max-width: 600px;
min-width: 300px;
line-height: 1.2;
margin: 0 auto;
padding: 0 0.5em;
}
p {
overflow-wrap: break-word;
word-wrap: break-word;
hyphens: auto;
margin-top: 0.5em;
margin-bottom: 0.5em;
}
#write form {
margin: 1em 0 0 0;
display: flex;
flex-direction: column;
gap: 0.5em;
align-items: stretch;
}
.right {
display: flex;
flex-direction: row;
gap: 0.5em;
justify-content: flex-end;
}
fieldset {
border: none;
padding: 0;
margin: 0;
}
textarea {
padding: 5px;
flex-grow: 1;
width: 100%;
}
pre code {
display: block;
background-color: #222;
color: #ddd;
}
code {
font-family: monospace;
padding: 0.1em 0.3em;
border-radius: 3px;
}
fieldset label {
margin: 0.5em 0;
display: block;
}
header, footer {
text-align: center;
}
footer {
font-size: 80%;
color: #888;
}
</style>
<script type="module">
import {
html, h, signal, effect, computed, render, useSignal, useEffect, useRef
} from '/index.js';
import { llama } from '/completion.js';
const session = signal({
prompt: "This is a conversation between user and llama, a friendly chatbot. respond in simple markdown.",
template: "{{prompt}}\n\n{{history}}\n{{char}}:",
historyTemplate: "{{name}}: {{message}}",
transcript: [],
type: "chat",
char: "llama",
user: "User",
})
const params = signal({
n_predict: 400,
temperature: 0.7,
repeat_last_n: 256,
repeat_penalty: 1.18,
top_k: 40,
top_p: 0.5,
})
const llamaStats = signal(null)
const controller = signal(null)
const generating = computed(() => controller.value == null )
const chatStarted = computed(() => session.value.transcript.length > 0)
const transcriptUpdate = (transcript) => {
session.value = {
...session.value,
transcript
}
}
// simple template replace
const template = (str, extraSettings) => {
let settings = session.value;
if (extraSettings) {
settings = { ...settings, ...extraSettings };
}
return String(str).replaceAll(/\{\{(.*?)\}\}/g, (_, key) => template(settings[key]));
}
// send message to server
const chat = async (msg) => {
if (controller.value) {
console.log('already running...');
return;
}
controller.value = new AbortController();
transcriptUpdate([...session.value.transcript, ["{{user}}", msg]])
const prompt = template(session.value.template, {
message: msg,
history: session.value.transcript.flatMap(([name, message]) => template(session.value.historyTemplate, {name, message})).join("\n"),
});
let currentMessage = '';
const history = session.value.transcript
const llamaParams = {
...params.value,
stop: ["</s>", template("{{char}}:"), template("{{user}}:")],
}
for await (const chunk of llama(prompt, llamaParams, { controller: controller.value })) {
const data = chunk.data;
currentMessage += data.content;
// remove leading whitespace
currentMessage = currentMessage.replace(/^\s+/, "")
transcriptUpdate([...history, ["{{char}}", currentMessage]])
if (data.stop) {
console.log("Completion finished: '", currentMessage, "', summary: ", data);
}
if (data.timings) {
llamaStats.value = data.timings;
}
}
controller.value = null;
}
function MessageInput() {
const message = useSignal("")
const stop = (e) => {
e.preventDefault();
if (controller.value) {
controller.value.abort();
controller.value = null;
}
}
const reset = (e) => {
stop(e);
transcriptUpdate([]);
}
const submit = (e) => {
stop(e);
chat(message.value);
message.value = "";
}
const enterSubmits = (event) => {
if (event.which === 13 && !event.shiftKey) {
submit(event);
}
}
return html`
<form onsubmit=${submit}>
<div>
<textarea type="text" rows=2 onkeypress=${enterSubmits} value="${message}" oninput=${(e) => message.value = e.target.value} placeholder="Say something..."/>
</div>
<div class="right">
<button type="submit" disabled=${!generating.value} >Send</button>
<button onclick=${stop} disabled=${generating}>Stop</button>
<button onclick=${reset}>Reset</button>
</div>
</form>
`
}
const ChatLog = (props) => {
const messages = session.value.transcript;
const container = useRef(null)
useEffect(() => {
// scroll to bottom (if needed)
if (container.current && container.current.scrollHeight <= container.current.scrollTop + container.current.offsetHeight + 300) {
container.current.scrollTo(0, container.current.scrollHeight)
}
}, [messages])
const chatLine = ([user, msg]) => {
return html`<p key=${msg}><strong>${template(user)}:</strong> <${Markdownish} text=${template(msg)} /></p>`
};
return html`
<section id="chat" ref=${container}>
${messages.flatMap(chatLine)}
</section>`;
};
const ConfigForm = (props) => {
const updateSession = (el) => session.value = { ...session.value, [el.target.name]: el.target.value }
const updateParams = (el) => params.value = { ...params.value, [el.target.name]: el.target.value }
const updateParamsFloat = (el) => params.value = { ...params.value, [el.target.name]: parseFloat(el.target.value) }
return html`
<form>
<fieldset>
<div>
<label for="prompt">Prompt</label>
<textarea type="text" name="prompt" value="${session.value.prompt}" rows=4 oninput=${updateSession}/>
</div>
<div>
<label for="user">User name</label>
<input type="text" name="user" value="${session.value.user}" oninput=${updateSession} />
</div>
<div>
<label for="bot">Bot name</label>
<input type="text" name="char" value="${session.value.char}" oninput=${updateSession} />
</div>
<div>
<label for="template">Prompt template</label>
<textarea id="template" name="template" value="${session.value.template}" rows=4 oninput=${updateSession}/>
</div>
<div>
<label for="template">Chat history template</label>
<textarea id="template" name="historyTemplate" value="${session.value.historyTemplate}" rows=1 oninput=${updateSession}/>
</div>
<div>
<label for="temperature">Temperature</label>
<input type="range" id="temperature" min="0.0" max="1.0" step="0.01" name="temperature" value="${params.value.temperature}" oninput=${updateParamsFloat} />
<span>${params.value.temperature}</span>
</div>
<div>
<label for="nPredict">Predictions</label>
<input type="range" id="nPredict" min="1" max="2048" step="1" name="n_predict" value="${params.value.n_predict}" oninput=${updateParamsFloat} />
<span>${params.value.n_predict}</span>
</div>
<div>
<label for="repeat_penalty">Penalize repeat sequence</label>
<input type="range" id="repeat_penalty" min="0.0" max="2.0" step="0.01" name="repeat_penalty" value="${params.value.repeat_penalty}" oninput=${updateParamsFloat} />
<span>${params.value.repeat_penalty}</span>
</div>
<div>
<label for="repeat_last_n">Consider N tokens for penalize</label>
<input type="range" id="repeat_last_n" min="0.0" max="2048" name="repeat_last_n" value="${params.value.repeat_last_n}" oninput=${updateParamsFloat} />
<span>${params.value.repeat_last_n}</span>
</div>
</fieldset>
</form>
`
}
// poor mans markdown replacement
const Markdownish = (params) => {
const md = params.text
.replace(/^#{1,6} (.*)$/gim, '<h3>$1</h3>')
.replace(/\*\*(.*?)\*\*/g, '<strong>$1</strong>')
.replace(/__(.*?)__/g, '<strong>$1</strong>')
.replace(/\*(.*?)\*/g, '<em>$1</em>')
.replace(/_(.*?)_/g, '<em>$1</em>')
.replace(/```.*?\n([\s\S]*?)```/g, '<pre><code>$1</code></pre>')
.replace(/`(.*?)`/g, '<code>$1</code>')
.replace(/\n/gim, '<br />');
return html`<span dangerouslySetInnerHTML=${{ __html: md }} />`;
};
const ModelGenerationInfo = (params) => {
if (!llamaStats.value) {
return html`<span/>`
}
return html`
<span>
${llamaStats.value.predicted_per_token_ms.toFixed()}ms per token, ${llamaStats.value.predicted_per_second.toFixed(2)} tokens per second
</span>
`
}
function App(props) {
return html`
<div id="container">
<header>
<h1>llama.cpp</h1>
</header>
<main id="content">
<${chatStarted.value ? ChatLog : ConfigForm} />
</main>
<section id="write">
<${MessageInput} />
</section>
<footer>
<p><${ModelGenerationInfo} /></p>
<p>Powered by <a href="https://github.com/ggerganov/llama.cpp">llama.cpp</a> and <a href="https://ggml.ai">ggml.ai</a>.</p>
</footer>
</div>
`;
}
render(h(App), document.body);
</script>
</head>
<body>
</body>
</html>