feat: vllm support

This commit is contained in:
jialin
2024-09-25 16:17:25 +08:00
parent 0652f850d6
commit 3cd96ba2e8
47 changed files with 1610 additions and 252 deletions
@@ -99,7 +99,7 @@ const GroundLeft: React.FC<MessageProps> = forwardRef((props, ref) => {
if (!chunk) {
return;
}
if (_.get(chunk, 'choices.0.finish_reason')) {
if (!_.get(chunk, 'choices', []).length) {
setTokenResult({
...chunk?.usage
});
@@ -142,6 +142,9 @@ const GroundLeft: React.FC<MessageProps> = forwardRef((props, ref) => {
: [];
contentRef.current = '';
setMessageList((pre) => {
return [...pre, ...currentMessageRef.current];
});
const formatMessages = _.map(
[...messageList, ...currentMessageRef.current],
(item: MessageItem) => {
@@ -217,9 +220,10 @@ const GroundLeft: React.FC<MessageProps> = forwardRef((props, ref) => {
setTokenResult(null);
};
const handleSendMessage = (message: { role: string; content: string }) => {
const handleSendMessage = (message: Omit<MessageItem, 'uid'>) => {
console.log('message:', message);
const currentMessage = message.content ? message : undefined;
const currentMessage =
message.content || message.imgs?.length ? message : undefined;
submitMessage(currentMessage);
};
@@ -110,7 +110,9 @@ const MessageInput: React.FC<MessageInputProps> = ({
const inputRef = useRef<any>(null);
const isDisabled = useMemo(() => {
return disabled ? true : !message.content && isEmpty;
return disabled
? true
: !message.content && isEmpty && !message.imgs?.length;
}, [disabled, message.content, isEmpty]);
const resetMessage = () => {
@@ -214,9 +216,6 @@ const MessageInput: React.FC<MessageInputProps> = ({
dataUrl: img
};
});
// setImgList((pre) => {
// return [...pre, ...list];
// });
setMessage({
...message,
imgs: [...(message.imgs || []), ...list]
@@ -253,7 +252,7 @@ const MessageInput: React.FC<MessageInputProps> = ({
if (text) {
setMessage?.({
...message,
content: text
content: message.content + text
});
} else {
getPasteContent(e);
@@ -446,7 +445,7 @@ const MessageInput: React.FC<MessageInputProps> = ({
<div className="input-box">
<TextArea
ref={inputRef}
autoSize={{ minRows: 3, maxRows: 3 }}
autoSize={{ minRows: 3, maxRows: 8 }}
onChange={(e) => handleInputChange(e.target.value)}
value={message.content}
size="large"
@@ -83,7 +83,7 @@ const ModelItem: React.FC<ModelItemProps> = forwardRef(
if (!chunk) {
return;
}
if (_.get(chunk, 'choices.0.finish_reason')) {
if (!_.get(chunk, 'choices', [].length)) {
setTokenResult({
...chunk?.usage
});
@@ -120,6 +120,9 @@ const ModelItem: React.FC<ModelItemProps> = forwardRef(
}
]
: [];
setMessageList((preList) => {
return [...preList, ...currentMessageRef.current];
});
console.log('currentMessageRef.current 1:', currentMessageRef.current);
console.log('currentMessage==========4', messageList);
const messages = _.map(
@@ -208,12 +211,11 @@ const ModelItem: React.FC<ModelItemProps> = forwardRef(
}
}, []);
const handleSubmit = (currentMessage: {
role: string;
content: string;
}) => {
console.log('currentMessage==========2', currentMessage);
const currentMsg = currentMessage.content ? currentMessage : undefined;
const handleSubmit = (currentMessage: Omit<MessageItem, 'uid'>) => {
const currentMsg =
currentMessage.content || currentMessage.imgs?.length
? currentMessage
: undefined;
submitMessage(currentMsg);
};
@@ -20,6 +20,7 @@ const ReferenceParams = (props: ReferenceParamsProps) => {
if (!usage) {
return null;
}
console.log('ReferenceParams usage:', usage);
return (
<div className="reference-params">
<span className="usage">
@@ -48,14 +49,18 @@ const ReferenceParams = (props: ReferenceParamsProps) => {
<Tooltip
title={
<Space>
<span>TPOT: {_.round(usage.time_per_output_token_ms, 2)} ms</span>
<span>TTFT: {_.round(usage.time_to_first_token_ms, 2)} ms</span>
<span>
TPOT: {_.round(usage.time_per_output_token_ms, 2) || 0} ms
</span>
<span>
TTFT: {_.round(usage.time_to_first_token_ms, 2) || 0} ms
</span>
</Space>
}
>
<span>
{intl.formatMessage({ id: 'playground.tokenoutput' })}:{' '}
{_.round(usage.tokens_per_second, 2)} Tokens/s
{_.round(usage.tokens_per_second, 2) || 0} Tokens/s
</span>
</Tooltip>
</span>
@@ -1,6 +1,6 @@
import EditorWrap from '@/components/editor-wrap';
import HighlightCode from '@/components/highlight-code';
import { BulbOutlined } from '@ant-design/icons';
import Editor from '@monaco-editor/react';
import { useIntl } from '@umijs/max';
import { Button, Modal } from 'antd';
import _ from 'lodash';
@@ -16,6 +16,18 @@ type ViewModalProps = {
onCancel: () => void;
};
const langMap = {
shell: 'bash',
python: 'python',
javascript: 'javascript'
};
const langOptions = [
{ label: 'Curl', value: langMap.shell },
{ label: 'Python', value: langMap.python },
{ label: 'Nodejs', value: langMap.javascript }
];
const ViewCodeModal: React.FC<ViewModalProps> = (props) => {
const {
title,
@@ -31,7 +43,7 @@ const ViewCodeModal: React.FC<ViewModalProps> = (props) => {
const editorRef = useRef(null);
const [loaded, setLoaded] = useState(false);
const [codeValue, setCodeValue] = useState('');
const [lang, setLang] = useState('shell');
const [lang, setLang] = useState(langMap.shell);
const BaseURL = `${window.location.origin}/v1-openai`;
const ClientType = apiType === 'chat' ? 'chat.completions' : 'embeddings';
@@ -39,24 +51,6 @@ const ViewCodeModal: React.FC<ViewModalProps> = (props) => {
const logcommand =
apiType === 'chat' ? 'choices[0].message.content' : 'data[0].embedding';
const langOptions = [
{ label: 'Curl', value: 'shell' },
{ label: 'Python', value: 'python' },
{ label: 'Nodejs', value: 'javascript' }
];
const formatCode = () => {
if (editorRef.current) {
setTimeout(() => {
editorRef.current
?.getAction?.('editor.action.formatDocument')
?.run()
.then(() => {
console.log('format success');
});
}, 100);
}
};
const generateCode = () => {
const systemList = systemMessage
? [
@@ -91,7 +85,7 @@ const ViewCodeModal: React.FC<ViewModalProps> = (props) => {
]
};
});
if (lang === 'shell') {
if (lang === langMap.shell) {
const messages = [...systemList, ...formatMessageList];
const code = `curl ${window.location.origin}/v1-openai/${api} \\\n-H "Content-Type: application/json" \\\n-H "Authorization: Bearer $\{YOUR_GPUSTACK_API_KEY}" \\\n-d '${JSON.stringify(
{
@@ -102,7 +96,7 @@ const ViewCodeModal: React.FC<ViewModalProps> = (props) => {
2
)}'`;
setCodeValue(code);
} else if (lang === 'javascript') {
} else if (lang === langMap.javascript) {
const messages = [...systemList, ...formatMessageList];
const code = `const OpenAI = require("openai");\n\nconst openai = new OpenAI({\n "apiKey": "YOUR_GPUSTACK_API_KEY",\n "baseURL": "${BaseURL}"\n});\n\nasync function main(){\n const params = ${JSON.stringify(
{
@@ -113,7 +107,7 @@ const ViewCodeModal: React.FC<ViewModalProps> = (props) => {
4
)};\nconst response = await openai.${ClientType}.create(params);\n console.log(response.${logcommand});\n}\nmain();`;
setCodeValue(code);
} else if (lang === 'python') {
} else if (lang === langMap.python) {
const formattedParams = _.keys(parameters).reduce(
(acc: string, key: string) => {
if (parameters[key] === null) {
@@ -138,20 +132,6 @@ const ViewCodeModal: React.FC<ViewModalProps> = (props) => {
const code = `from openai import OpenAI\n\nclient = OpenAI(\n base_url="${BaseURL}", \n api_key="YOUR_GPUSTACK_API_KEY"\n)\n\nresponse = client.${ClientType}.create(\n${formattedParams} ${messages})\nprint(response.${logcommand})`;
setCodeValue(code);
}
formatCode();
};
const handleEditorDidMount = (editor: any, monaco: any) => {
editorRef.current = editor;
setLoaded(true);
console.log('loaded====', editor, monaco);
};
const handleBeforeMount = (monaco: any) => {
monaco.languages.typescript.javascriptDefaults.setDiagnosticsOptions({
noSemanticValidation: false,
noSyntaxValidation: false,
diagnosticCodesToIgnore: [80001]
});
};
const handleOnChangeLang = (value: string) => {
@@ -159,7 +139,7 @@ const ViewCodeModal: React.FC<ViewModalProps> = (props) => {
};
const handleClose = () => {
setLang('shell');
setLang(langMap.shell);
onCancel();
};
const editorConfig = {
@@ -203,21 +183,22 @@ const ViewCodeModal: React.FC<ViewModalProps> = (props) => {
<EditorWrap
copyText={codeValue}
langOptions={langOptions}
defaultValue="shell"
showHeader={loaded}
defaultValue={langMap.shell}
showHeader={true}
onChangeLang={handleOnChangeLang}
styles={{
wrapper: {
backgroundColor: 'var(--color-editor-dark)'
}
}}
>
<Editor
<HighlightCode
height={380}
theme="vs-dark"
className="monaco-editor"
defaultLanguage="shell"
language={lang}
value={codeValue}
options={editorConfig}
beforeMount={handleBeforeMount}
onMount={handleEditorDidMount}
/>
theme="dark"
code={codeValue}
lang={lang}
copyable={false}
></HighlightCode>
</EditorWrap>
<div
style={{ marginTop: 10, display: 'flex', alignItems: 'baseline' }}