index.html 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478
  1. <html>
  2. <head>
  3. <meta charset="UTF-8">
  4. <meta name="viewport" content="width=device-width, initial-scale=1, maximum-scale=1" />
  5. <meta name="color-scheme" content="light dark">
  6. <title>llama.cpp - chat</title>
  7. <style>
  8. body {
  9. font-family: system-ui;
  10. font-size: 90%;
  11. }
  12. #container {
  13. margin: 0em auto;
  14. display: flex;
  15. flex-direction: column;
  16. justify-content: space-between;
  17. height: 100%;
  18. }
  19. main {
  20. margin: 3px;
  21. display: flex;
  22. flex-direction: column;
  23. justify-content: space-between;
  24. gap: 1em;
  25. flex-grow: 1;
  26. overflow-y: auto;
  27. border: 1px solid #ccc;
  28. border-radius: 5px;
  29. padding: 0.5em;
  30. }
  31. body {
  32. max-width: 600px;
  33. min-width: 300px;
  34. line-height: 1.2;
  35. margin: 0 auto;
  36. padding: 0 0.5em;
  37. }
  38. p {
  39. overflow-wrap: break-word;
  40. word-wrap: break-word;
  41. hyphens: auto;
  42. margin-top: 0.5em;
  43. margin-bottom: 0.5em;
  44. }
  45. #write form {
  46. margin: 1em 0 0 0;
  47. display: flex;
  48. flex-direction: column;
  49. gap: 0.5em;
  50. align-items: stretch;
  51. }
  52. .right {
  53. display: flex;
  54. flex-direction: row;
  55. gap: 0.5em;
  56. justify-content: flex-end;
  57. }
  58. fieldset {
  59. border: none;
  60. padding: 0;
  61. margin: 0;
  62. }
  63. fieldset.two {
  64. display: grid;
  65. grid-template: "a a";
  66. gap: 1em;
  67. }
  68. fieldset.three {
  69. display: grid;
  70. grid-template: "a a a";
  71. gap: 1em;
  72. }
  73. details {
  74. border: 1px solid #aaa;
  75. border-radius: 4px;
  76. padding: 0.5em 0.5em 0;
  77. margin-top: 0.5em;
  78. }
  79. summary {
  80. font-weight: bold;
  81. margin: -0.5em -0.5em 0;
  82. padding: 0.5em;
  83. cursor: pointer;
  84. }
  85. details[open] {
  86. padding: 0.5em;
  87. }
  88. textarea {
  89. padding: 5px;
  90. flex-grow: 1;
  91. width: 100%;
  92. }
  93. pre code {
  94. display: block;
  95. background-color: #222;
  96. color: #ddd;
  97. }
  98. code {
  99. font-family: monospace;
  100. padding: 0.1em 0.3em;
  101. border-radius: 3px;
  102. }
  103. fieldset label {
  104. margin: 0.5em 0;
  105. display: block;
  106. }
  107. header, footer {
  108. text-align: center;
  109. }
  110. footer {
  111. font-size: 80%;
  112. color: #888;
  113. }
  114. </style>
  115. <script type="module">
  116. import {
  117. html, h, signal, effect, computed, render, useSignal, useEffect, useRef
  118. } from '/index.js';
  119. import { llama } from '/completion.js';
  120. import { SchemaConverter } from '/json-schema-to-grammar.mjs';
  121. const session = signal({
  122. prompt: "This is a conversation between user and llama, a friendly chatbot. respond in simple markdown.",
  123. template: "{{prompt}}\n\n{{history}}\n{{char}}:",
  124. historyTemplate: "{{name}}: {{message}}",
  125. transcript: [],
  126. type: "chat",
  127. char: "llama",
  128. user: "User",
  129. })
  130. const params = signal({
  131. n_predict: 400,
  132. temperature: 0.7,
  133. repeat_last_n: 256, // 0 = disable penalty, -1 = context size
  134. repeat_penalty: 1.18, // 1.0 = disabled
  135. top_k: 40, // <= 0 to use vocab size
  136. top_p: 0.5, // 1.0 = disabled
  137. tfs_z: 1.0, // 1.0 = disabled
  138. typical_p: 1.0, // 1.0 = disabled
  139. presence_penalty: 0.0, // 0.0 = disabled
  140. frequency_penalty: 0.0, // 0.0 = disabled
  141. mirostat: 0, // 0/1/2
  142. mirostat_tau: 5, // target entropy
  143. mirostat_eta: 0.1, // learning rate
  144. grammar: '',
  145. })
  146. const llamaStats = signal(null)
  147. const controller = signal(null)
  148. const generating = computed(() => controller.value == null )
  149. const chatStarted = computed(() => session.value.transcript.length > 0)
  150. const transcriptUpdate = (transcript) => {
  151. session.value = {
  152. ...session.value,
  153. transcript
  154. }
  155. }
  156. // simple template replace
  157. const template = (str, extraSettings) => {
  158. let settings = session.value;
  159. if (extraSettings) {
  160. settings = { ...settings, ...extraSettings };
  161. }
  162. return String(str).replaceAll(/\{\{(.*?)\}\}/g, (_, key) => template(settings[key]));
  163. }
  164. // send message to server
  165. const chat = async (msg) => {
  166. if (controller.value) {
  167. console.log('already running...');
  168. return;
  169. }
  170. controller.value = new AbortController();
  171. transcriptUpdate([...session.value.transcript, ["{{user}}", msg]])
  172. const prompt = template(session.value.template, {
  173. message: msg,
  174. history: session.value.transcript.flatMap(([name, message]) => template(session.value.historyTemplate, {name, message})).join("\n"),
  175. });
  176. let currentMessage = '';
  177. const history = session.value.transcript
  178. const llamaParams = {
  179. ...params.value,
  180. stop: ["</s>", template("{{char}}:"), template("{{user}}:")],
  181. }
  182. for await (const chunk of llama(prompt, llamaParams, { controller: controller.value })) {
  183. const data = chunk.data;
  184. currentMessage += data.content;
  185. // remove leading whitespace
  186. currentMessage = currentMessage.replace(/^\s+/, "")
  187. transcriptUpdate([...history, ["{{char}}", currentMessage]])
  188. if (data.stop) {
  189. console.log("Completion finished: '", currentMessage, "', summary: ", data);
  190. }
  191. if (data.timings) {
  192. llamaStats.value = data.timings;
  193. }
  194. }
  195. controller.value = null;
  196. }
  197. function MessageInput() {
  198. const message = useSignal("")
  199. const stop = (e) => {
  200. e.preventDefault();
  201. if (controller.value) {
  202. controller.value.abort();
  203. controller.value = null;
  204. }
  205. }
  206. const reset = (e) => {
  207. stop(e);
  208. transcriptUpdate([]);
  209. }
  210. const submit = (e) => {
  211. stop(e);
  212. chat(message.value);
  213. message.value = "";
  214. }
  215. const enterSubmits = (event) => {
  216. if (event.which === 13 && !event.shiftKey) {
  217. submit(event);
  218. }
  219. }
  220. return html`
  221. <form onsubmit=${submit}>
  222. <div>
  223. <textarea type="text" rows=2 onkeypress=${enterSubmits} value="${message}" oninput=${(e) => message.value = e.target.value} placeholder="Say something..."/>
  224. </div>
  225. <div class="right">
  226. <button type="submit" disabled=${!generating.value} >Send</button>
  227. <button onclick=${stop} disabled=${generating}>Stop</button>
  228. <button onclick=${reset}>Reset</button>
  229. </div>
  230. </form>
  231. `
  232. }
  233. const ChatLog = (props) => {
  234. const messages = session.value.transcript;
  235. const container = useRef(null)
  236. useEffect(() => {
  237. // scroll to bottom (if needed)
  238. const parent = container.current.parentElement;
  239. if (parent && parent.scrollHeight <= parent.scrollTop + parent.offsetHeight + 300) {
  240. parent.scrollTo(0, parent.scrollHeight)
  241. }
  242. }, [messages])
  243. const chatLine = ([user, msg]) => {
  244. return html`<p key=${msg}><strong>${template(user)}:</strong> <${Markdownish} text=${template(msg)} /></p>`
  245. };
  246. return html`
  247. <section id="chat" ref=${container}>
  248. ${messages.flatMap(chatLine)}
  249. </section>`;
  250. };
  251. const ConfigForm = (props) => {
  252. const updateSession = (el) => session.value = { ...session.value, [el.target.name]: el.target.value }
  253. const updateParams = (el) => params.value = { ...params.value, [el.target.name]: el.target.value }
  254. const updateParamsFloat = (el) => params.value = { ...params.value, [el.target.name]: parseFloat(el.target.value) }
  255. const updateParamsInt = (el) => params.value = { ...params.value, [el.target.name]: Math.floor(parseFloat(el.target.value)) }
  256. const grammarJsonSchemaPropOrder = signal('')
  257. const updateGrammarJsonSchemaPropOrder = (el) => grammarJsonSchemaPropOrder.value = el.target.value
  258. const convertJSONSchemaGrammar = () => {
  259. try {
  260. const schema = JSON.parse(params.value.grammar)
  261. const converter = new SchemaConverter(
  262. grammarJsonSchemaPropOrder.value
  263. .split(',')
  264. .reduce((acc, cur, i) => ({...acc, [cur.trim()]: i}), {})
  265. )
  266. converter.visit(schema, '')
  267. params.value = {
  268. ...params.value,
  269. grammar: converter.formatGrammar(),
  270. }
  271. } catch (e) {
  272. alert(`Convert failed: ${e.message}`)
  273. }
  274. }
  275. const FloatField = ({label, max, min, name, step, value}) => {
  276. return html`
  277. <div>
  278. <label for="${name}">${label}</label>
  279. <input type="range" id="${name}" min="${min}" max="${max}" step="${step}" name="${name}" value="${value}" oninput=${updateParamsFloat} />
  280. <span>${value}</span>
  281. </div>
  282. `
  283. };
  284. const IntField = ({label, max, min, name, value}) => {
  285. return html`
  286. <div>
  287. <label for="${name}">${label}</label>
  288. <input type="range" id="${name}" min="${min}" max="${max}" name="${name}" value="${value}" oninput=${updateParamsInt} />
  289. <span>${value}</span>
  290. </div>
  291. `
  292. };
  293. return html`
  294. <form>
  295. <fieldset>
  296. <div>
  297. <label for="prompt">Prompt</label>
  298. <textarea type="text" name="prompt" value="${session.value.prompt}" rows=4 oninput=${updateSession}/>
  299. </div>
  300. </fieldset>
  301. <fieldset class="two">
  302. <div>
  303. <label for="user">User name</label>
  304. <input type="text" name="user" value="${session.value.user}" oninput=${updateSession} />
  305. </div>
  306. <div>
  307. <label for="bot">Bot name</label>
  308. <input type="text" name="char" value="${session.value.char}" oninput=${updateSession} />
  309. </div>
  310. </fieldset>
  311. <fieldset>
  312. <div>
  313. <label for="template">Prompt template</label>
  314. <textarea id="template" name="template" value="${session.value.template}" rows=4 oninput=${updateSession}/>
  315. </div>
  316. <div>
  317. <label for="template">Chat history template</label>
  318. <textarea id="template" name="historyTemplate" value="${session.value.historyTemplate}" rows=1 oninput=${updateSession}/>
  319. </div>
  320. <div>
  321. <label for="template">Grammar</label>
  322. <textarea id="grammar" name="grammar" placeholder="Use gbnf or JSON Schema+convert" value="${params.value.grammar}" rows=4 oninput=${updateParams}/>
  323. <input type="text" name="prop-order" placeholder="order: prop1,prop2,prop3" oninput=${updateGrammarJsonSchemaPropOrder} />
  324. <button type="button" onclick=${convertJSONSchemaGrammar}>Convert JSON Schema</button>
  325. </div>
  326. </fieldset>
  327. <fieldset class="two">
  328. ${IntField({label: "Predictions", max: 2048, min: -1, name: "n_predict", value: params.value.n_predict})}
  329. ${FloatField({label: "Temperature", max: 1.5, min: 0.0, name: "temperature", step: 0.01, value: params.value.temperature})}
  330. ${FloatField({label: "Penalize repeat sequence", max: 2.0, min: 0.0, name: "repeat_penalty", step: 0.01, value: params.value.repeat_penalty})}
  331. ${IntField({label: "Consider N tokens for penalize", max: 2048, min: 0, name: "repeat_last_n", value: params.value.repeat_last_n})}
  332. ${IntField({label: "Top-K sampling", max: 100, min: -1, name: "top_k", value: params.value.top_k})}
  333. ${FloatField({label: "Top-P sampling", max: 1.0, min: 0.0, name: "top_p", step: 0.01, value: params.value.top_p})}
  334. </fieldset>
  335. <details>
  336. <summary>More options</summary>
  337. <fieldset class="two">
  338. ${FloatField({label: "TFS-Z", max: 1.0, min: 0.0, name: "tfs_z", step: 0.01, value: params.value.tfs_z})}
  339. ${FloatField({label: "Typical P", max: 1.0, min: 0.0, name: "typical_p", step: 0.01, value: params.value.typical_p})}
  340. ${FloatField({label: "Presence penalty", max: 1.0, min: 0.0, name: "presence_penalty", step: 0.01, value: params.value.presence_penalty})}
  341. ${FloatField({label: "Frequency penalty", max: 1.0, min: 0.0, name: "frequency_penalty", step: 0.01, value: params.value.frequency_penalty})}
  342. </fieldset>
  343. <hr />
  344. <fieldset class="three">
  345. <div>
  346. <label><input type="radio" name="mirostat" value="0" checked=${params.value.mirostat == 0} oninput=${updateParamsInt} /> no Mirostat</label>
  347. <label><input type="radio" name="mirostat" value="1" checked=${params.value.mirostat == 1} oninput=${updateParamsInt} /> Mirostat v1</label>
  348. <label><input type="radio" name="mirostat" value="2" checked=${params.value.mirostat == 2} oninput=${updateParamsInt} /> Mirostat v2</label>
  349. </div>
  350. ${FloatField({label: "Mirostat tau", max: 10.0, min: 0.0, name: "mirostat_tau", step: 0.01, value: params.value.mirostat_tau})}
  351. ${FloatField({label: "Mirostat eta", max: 1.0, min: 0.0, name: "mirostat_eta", step: 0.01, value: params.value.mirostat_eta})}
  352. </fieldset>
  353. </details>
  354. </form>
  355. `
  356. }
  357. // poor mans markdown replacement
  358. const Markdownish = (params) => {
  359. const md = params.text
  360. .replace(/&/g, '&amp;')
  361. .replace(/</g, '&lt;')
  362. .replace(/>/g, '&gt;')
  363. .replace(/^#{1,6} (.*)$/gim, '<h3>$1</h3>')
  364. .replace(/\*\*(.*?)\*\*/g, '<strong>$1</strong>')
  365. .replace(/__(.*?)__/g, '<strong>$1</strong>')
  366. .replace(/\*(.*?)\*/g, '<em>$1</em>')
  367. .replace(/_(.*?)_/g, '<em>$1</em>')
  368. .replace(/```.*?\n([\s\S]*?)```/g, '<pre><code>$1</code></pre>')
  369. .replace(/`(.*?)`/g, '<code>$1</code>')
  370. .replace(/\n/gim, '<br />');
  371. return html`<span dangerouslySetInnerHTML=${{ __html: md }} />`;
  372. };
  373. const ModelGenerationInfo = (params) => {
  374. if (!llamaStats.value) {
  375. return html`<span/>`
  376. }
  377. return html`
  378. <span>
  379. ${llamaStats.value.predicted_per_token_ms.toFixed()}ms per token, ${llamaStats.value.predicted_per_second.toFixed(2)} tokens per second
  380. </span>
  381. `
  382. }
  383. function App(props) {
  384. return html`
  385. <div id="container">
  386. <header>
  387. <h1>llama.cpp</h1>
  388. </header>
  389. <main id="content">
  390. <${chatStarted.value ? ChatLog : ConfigForm} />
  391. </main>
  392. <section id="write">
  393. <${MessageInput} />
  394. </section>
  395. <footer>
  396. <p><${ModelGenerationInfo} /></p>
  397. <p>Powered by <a href="https://github.com/ggerganov/llama.cpp">llama.cpp</a> and <a href="https://ggml.ai">ggml.ai</a>.</p>
  398. </footer>
  399. </div>
  400. `;
  401. }
  402. render(h(App), document.body);
  403. </script>
  404. </head>
  405. <body>
  406. </body>
  407. </html>