mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-12-24 10:24:35 +00:00
server : add completion mode (no chat) (#3582)
This commit is contained in:
Aarni Koskela
GitHub
parent
6b3ae4da92
commit
b016596d90
File diff suppressed because it is too large
Load Diff
@ -136,6 +136,11 @@
|
||||
display: block;
|
||||
}
|
||||
|
||||
fieldset label.slim {
|
||||
margin: 0 0.5em;
|
||||
display: inline;
|
||||
}
|
||||
|
||||
header, footer {
|
||||
text-align: center;
|
||||
}
|
||||
@ -145,6 +150,14 @@
|
||||
color: #888;
|
||||
}
|
||||
|
||||
.mode-chat textarea[name=prompt] {
|
||||
height: 4.5em;
|
||||
}
|
||||
|
||||
.mode-completion textarea[name=prompt] {
|
||||
height: 10em;
|
||||
}
|
||||
|
||||
|
||||
@keyframes loading-bg-wipe {
|
||||
0% {
|
||||
@ -187,7 +200,7 @@
|
||||
template: "{{prompt}}\n\n{{history}}\n{{char}}:",
|
||||
historyTemplate: "{{name}}: {{message}}",
|
||||
transcript: [],
|
||||
type: "chat",
|
||||
type: "chat", // "chat" | "completion"
|
||||
char: "Llama",
|
||||
user: "User",
|
||||
})
|
||||
@ -365,13 +378,44 @@
|
||||
return String(str).replaceAll(/\{\{(.*?)\}\}/g, (_, key) => template(settings[key]));
|
||||
}
|
||||
|
||||
async function runLlama(prompt, llamaParams, char) {
|
||||
const currentMessages = [];
|
||||
const history = session.value.transcript;
|
||||
if (controller.value) {
|
||||
throw new Error("already running");
|
||||
}
|
||||
controller.value = new AbortController();
|
||||
for await (const chunk of llama(prompt, llamaParams, {controller: controller.value})) {
|
||||
const data = chunk.data;
|
||||
|
||||
if (data.stop) {
|
||||
while (
|
||||
currentMessages.length > 0 &&
|
||||
currentMessages[currentMessages.length - 1].content.match(/\n$/) != null
|
||||
) {
|
||||
currentMessages.pop();
|
||||
}
|
||||
transcriptUpdate([...history, [char, currentMessages]])
|
||||
console.log("Completion finished: '", currentMessages.map(msg => msg.content).join(''), "', summary: ", data);
|
||||
} else {
|
||||
currentMessages.push(data);
|
||||
transcriptUpdate([...history, [char, currentMessages]])
|
||||
}
|
||||
|
||||
if (data.timings) {
|
||||
llamaStats.value = data.timings;
|
||||
}
|
||||
}
|
||||
|
||||
controller.value = null;
|
||||
}
|
||||
|
||||
// send message to server
|
||||
const chat = async (msg) => {
|
||||
if (controller.value) {
|
||||
console.log('already running...');
|
||||
return;
|
||||
}
|
||||
controller.value = new AbortController();
|
||||
|
||||
transcriptUpdate([...session.value.transcript, ["{{user}}", msg]])
|
||||
|
||||
@ -391,42 +435,25 @@
|
||||
).join("\n"),
|
||||
});
|
||||
|
||||
const currentMessages = [];
|
||||
const history = session.value.transcript
|
||||
|
||||
const llamaParams = {
|
||||
await runLlama(prompt, {
|
||||
...params.value,
|
||||
stop: ["</s>", template("{{char}}:"), template("{{user}}:")],
|
||||
}, "{{char}}");
|
||||
}
|
||||
|
||||
for await (const chunk of llama(prompt, llamaParams, { controller: controller.value })) {
|
||||
const data = chunk.data;
|
||||
|
||||
if (data.stop) {
|
||||
while (
|
||||
currentMessages.length > 0 &&
|
||||
currentMessages[currentMessages.length - 1].content.match(/\n$/) != null
|
||||
) {
|
||||
currentMessages.pop();
|
||||
const runCompletion = async () => {
|
||||
if (controller.value) {
|
||||
console.log('already running...');
|
||||
return;
|
||||
}
|
||||
transcriptUpdate([...history, ["{{char}}", currentMessages]])
|
||||
console.log("Completion finished: '", currentMessages.map(msg => msg.content).join(''), "', summary: ", data);
|
||||
} else {
|
||||
currentMessages.push(data);
|
||||
transcriptUpdate([...history, ["{{char}}", currentMessages]])
|
||||
const {prompt} = session.value;
|
||||
transcriptUpdate([...session.value.transcript, ["", prompt]]);
|
||||
await runLlama(prompt, {
|
||||
...params.value,
|
||||
stop: [],
|
||||
}, "");
|
||||
}
|
||||
|
||||
if (data.timings) {
|
||||
llamaStats.value = data.timings;
|
||||
}
|
||||
}
|
||||
|
||||
controller.value = null;
|
||||
}
|
||||
|
||||
function MessageInput() {
|
||||
const message = useSignal("")
|
||||
|
||||
const stop = (e) => {
|
||||
e.preventDefault();
|
||||
if (controller.value) {
|
||||
@ -440,6 +467,9 @@
|
||||
transcriptUpdate([]);
|
||||
}
|
||||
|
||||
function MessageInput() {
|
||||
const message = useSignal("")
|
||||
|
||||
const submit = (e) => {
|
||||
stop(e);
|
||||
chat(message.value);
|
||||
@ -474,6 +504,19 @@
|
||||
`
|
||||
}
|
||||
|
||||
function CompletionControls() {
|
||||
const submit = (e) => {
|
||||
stop(e);
|
||||
runCompletion();
|
||||
}
|
||||
return html`
|
||||
<div>
|
||||
<button onclick=${submit} type="button" disabled=${generating.value}>Start</button>
|
||||
<button onclick=${stop} disabled=${!generating.value}>Stop</button>
|
||||
<button onclick=${reset}>Reset</button>
|
||||
</div>`;
|
||||
}
|
||||
|
||||
const ChatLog = (props) => {
|
||||
const messages = session.value.transcript;
|
||||
const container = useRef(null)
|
||||
@ -497,7 +540,11 @@
|
||||
data;
|
||||
message = html`<${Markdownish} text=${template(text)} />`
|
||||
}
|
||||
if(user) {
|
||||
return html`<p key=${index}><strong>${template(user)}:</strong> ${message}</p>`
|
||||
} else {
|
||||
return html`<p key=${index}>${message}</p>`
|
||||
}
|
||||
};
|
||||
|
||||
return html`
|
||||
@ -574,18 +621,31 @@
|
||||
userTemplateAutosave()
|
||||
}, [session.value, params.value])
|
||||
|
||||
return html`
|
||||
<form>
|
||||
<fieldset>
|
||||
<${UserTemplateResetButton}/>
|
||||
</fieldset>
|
||||
const GrammarControl = () => (
|
||||
html`
|
||||
<div>
|
||||
<label for="template">Grammar</label>
|
||||
<textarea id="grammar" name="grammar" placeholder="Use gbnf or JSON Schema+convert" value="${params.value.grammar}" rows=4 oninput=${updateParams}/>
|
||||
<input type="text" name="prop-order" placeholder="order: prop1,prop2,prop3" oninput=${updateGrammarJsonSchemaPropOrder} />
|
||||
<button type="button" onclick=${convertJSONSchemaGrammar}>Convert JSON Schema</button>
|
||||
</div>
|
||||
`
|
||||
);
|
||||
|
||||
const PromptControlFieldSet = () => (
|
||||
html`
|
||||
<fieldset>
|
||||
<div>
|
||||
<label for="prompt">Prompt</label>
|
||||
<textarea type="text" name="prompt" value="${session.value.prompt}" rows=4 oninput=${updateSession}/>
|
||||
<label htmlFor="prompt">Prompt</label>
|
||||
<textarea type="text" name="prompt" value="${session.value.prompt}" oninput=${updateSession}/>
|
||||
</div>
|
||||
</fieldset>
|
||||
`
|
||||
);
|
||||
|
||||
const ChatConfigForm = () => (
|
||||
html`
|
||||
${PromptControlFieldSet()}
|
||||
|
||||
<fieldset class="two">
|
||||
<div>
|
||||
@ -609,15 +669,30 @@
|
||||
<label for="template">Chat history template</label>
|
||||
<textarea id="template" name="historyTemplate" value="${session.value.historyTemplate}" rows=1 oninput=${updateSession}/>
|
||||
</div>
|
||||
${GrammarControl()}
|
||||
</fieldset>
|
||||
`
|
||||
);
|
||||
|
||||
const CompletionConfigForm = () => (
|
||||
html`
|
||||
${PromptControlFieldSet()}
|
||||
<fieldset>${GrammarControl()}</fieldset>
|
||||
`
|
||||
);
|
||||
|
||||
return html`
|
||||
<form>
|
||||
<fieldset class="two">
|
||||
<${UserTemplateResetButton}/>
|
||||
<div>
|
||||
<label for="template">Grammar</label>
|
||||
<textarea id="grammar" name="grammar" placeholder="Use gbnf or JSON Schema+convert" value="${params.value.grammar}" rows=4 oninput=${updateParams}/>
|
||||
<input type="text" name="prop-order" placeholder="order: prop1,prop2,prop3" oninput=${updateGrammarJsonSchemaPropOrder} />
|
||||
<button type="button" onclick=${convertJSONSchemaGrammar}>Convert JSON Schema</button>
|
||||
<label class="slim"><input type="radio" name="type" value="chat" checked=${session.value.type === "chat"} oninput=${updateSession} /> Chat</label>
|
||||
<label class="slim"><input type="radio" name="type" value="completion" checked=${session.value.type === "completion"} oninput=${updateSession} /> Completion</label>
|
||||
</div>
|
||||
</fieldset>
|
||||
|
||||
${session.value.type === 'chat' ? ChatConfigForm() : CompletionConfigForm()}
|
||||
|
||||
<fieldset class="two">
|
||||
${IntField({label: "Predictions", max: 2048, min: -1, name: "n_predict", value: params.value.n_predict})}
|
||||
${FloatField({label: "Temperature", max: 1.5, min: 0.0, name: "temperature", step: 0.01, value: params.value.temperature})}
|
||||
@ -851,7 +926,7 @@
|
||||
function App(props) {
|
||||
|
||||
return html`
|
||||
<div>
|
||||
<div class="mode-${session.value.type}">
|
||||
<header>
|
||||
<h1>llama.cpp</h1>
|
||||
</header>
|
||||
@ -861,7 +936,7 @@
|
||||
</main>
|
||||
|
||||
<section id="write">
|
||||
<${MessageInput} />
|
||||
<${session.value.type === 'chat' ? MessageInput : CompletionControls} />
|
||||
</section>
|
||||
|
||||
<footer>
|
||||
|
||||
Loading…
Reference in New Issue
Block a user