feat: Add LM Studio Support and Thinking Model Panel
LM Studio Integration: - Added LM Studio provider with OpenAI-compatible API support - Dynamic model discovery via /v1/models endpoint - Support for both chat and embeddings models - Docker-compatible networking configuration Thinking Model Panel: - Added collapsible UI panel for model's chain of thought - Parses responses with <think> tags to separate reasoning - Maintains backward compatibility with regular responses - Styled consistently with app theme for light/dark modes - Preserves all existing message functionality (sources, markdown, etc.) These improvements enhance the app's compatibility with local LLMs and provide better visibility into model reasoning processes while maintaining existing functionality.
This commit is contained in:
parent
0737701de0
commit
6edac6938c
5 changed files with 191 additions and 18 deletions
|
@ -17,8 +17,9 @@ interface Config {
|
||||||
GEMINI: string;
|
GEMINI: string;
|
||||||
};
|
};
|
||||||
API_ENDPOINTS: {
|
API_ENDPOINTS: {
|
||||||
SEARXNG: string;
|
|
||||||
OLLAMA: string;
|
OLLAMA: string;
|
||||||
|
LMSTUDIO: string;
|
||||||
|
SEARXNG: string;
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -51,6 +52,8 @@ export const getSearxngApiEndpoint = () =>
|
||||||
|
|
||||||
export const getOllamaApiEndpoint = () => loadConfig().API_ENDPOINTS.OLLAMA;
|
export const getOllamaApiEndpoint = () => loadConfig().API_ENDPOINTS.OLLAMA;
|
||||||
|
|
||||||
|
export const getLMStudioApiEndpoint = () => loadConfig().API_ENDPOINTS.LMSTUDIO;
|
||||||
|
|
||||||
export const updateConfig = (config: RecursivePartial<Config>) => {
|
export const updateConfig = (config: RecursivePartial<Config>) => {
|
||||||
const currentConfig = loadConfig();
|
const currentConfig = loadConfig();
|
||||||
|
|
||||||
|
@ -72,6 +75,27 @@ export const updateConfig = (config: RecursivePartial<Config>) => {
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/*
|
||||||
|
export const updateConfig = (config: RecursivePartial<Config>) => {
|
||||||
|
const currentConfig = loadConfig();
|
||||||
|
|
||||||
|
// Merge existing config with new values
|
||||||
|
const mergedConfig: RecursivePartial<Config> = {
|
||||||
|
GENERAL: {
|
||||||
|
...currentConfig.GENERAL,
|
||||||
|
...config.GENERAL,
|
||||||
|
},
|
||||||
|
API_KEYS: {
|
||||||
|
...currentConfig.API_KEYS,
|
||||||
|
...config.API_KEYS,
|
||||||
|
},
|
||||||
|
API_ENDPOINTS: {
|
||||||
|
...currentConfig.API_ENDPOINTS,
|
||||||
|
...config.API_ENDPOINTS,
|
||||||
|
},
|
||||||
|
};
|
||||||
|
*/
|
||||||
|
|
||||||
fs.writeFileSync(
|
fs.writeFileSync(
|
||||||
path.join(__dirname, `../${configFileName}`),
|
path.join(__dirname, `../${configFileName}`),
|
||||||
toml.stringify(config),
|
toml.stringify(config),
|
||||||
|
|
|
@ -4,6 +4,7 @@ import { loadOpenAIChatModels, loadOpenAIEmbeddingsModels } from './openai';
|
||||||
import { loadAnthropicChatModels } from './anthropic';
|
import { loadAnthropicChatModels } from './anthropic';
|
||||||
import { loadTransformersEmbeddingsModels } from './transformers';
|
import { loadTransformersEmbeddingsModels } from './transformers';
|
||||||
import { loadGeminiChatModels, loadGeminiEmbeddingsModels } from './gemini';
|
import { loadGeminiChatModels, loadGeminiEmbeddingsModels } from './gemini';
|
||||||
|
import { loadLMStudioChatModels, loadLMStudioEmbeddingsModels } from './lmstudio';
|
||||||
|
|
||||||
const chatModelProviders = {
|
const chatModelProviders = {
|
||||||
openai: loadOpenAIChatModels,
|
openai: loadOpenAIChatModels,
|
||||||
|
@ -11,6 +12,7 @@ const chatModelProviders = {
|
||||||
ollama: loadOllamaChatModels,
|
ollama: loadOllamaChatModels,
|
||||||
anthropic: loadAnthropicChatModels,
|
anthropic: loadAnthropicChatModels,
|
||||||
gemini: loadGeminiChatModels,
|
gemini: loadGeminiChatModels,
|
||||||
|
lm_studio: loadLMStudioChatModels,
|
||||||
};
|
};
|
||||||
|
|
||||||
const embeddingModelProviders = {
|
const embeddingModelProviders = {
|
||||||
|
@ -18,6 +20,7 @@ const embeddingModelProviders = {
|
||||||
local: loadTransformersEmbeddingsModels,
|
local: loadTransformersEmbeddingsModels,
|
||||||
ollama: loadOllamaEmbeddingsModels,
|
ollama: loadOllamaEmbeddingsModels,
|
||||||
gemini: loadGeminiEmbeddingsModels,
|
gemini: loadGeminiEmbeddingsModels,
|
||||||
|
lm_studio: loadLMStudioEmbeddingsModels,
|
||||||
};
|
};
|
||||||
|
|
||||||
export const getAvailableChatModelProviders = async () => {
|
export const getAvailableChatModelProviders = async () => {
|
||||||
|
|
89
src/lib/providers/lmstudio.ts
Normal file
89
src/lib/providers/lmstudio.ts
Normal file
|
@ -0,0 +1,89 @@
|
||||||
|
import { OpenAIEmbeddings } from '@langchain/openai';
|
||||||
|
import { ChatOpenAI } from '@langchain/openai';
|
||||||
|
import { getKeepAlive, getLMStudioApiEndpoint } from '../../config';
|
||||||
|
import logger from '../../utils/logger';
|
||||||
|
import axios from 'axios';
|
||||||
|
|
||||||
|
interface LMStudioModel {
|
||||||
|
id: string;
|
||||||
|
// add other properties if LM Studio API provides them
|
||||||
|
}
|
||||||
|
|
||||||
|
interface ChatModelConfig {
|
||||||
|
displayName: string;
|
||||||
|
model: ChatOpenAI;
|
||||||
|
}
|
||||||
|
|
||||||
|
export const loadLMStudioChatModels = async (): Promise<Record<string, ChatModelConfig>> => {
|
||||||
|
const lmStudioEndpoint = getLMStudioApiEndpoint();
|
||||||
|
|
||||||
|
if (!lmStudioEndpoint) {
|
||||||
|
logger.debug('LM Studio endpoint not configured, skipping');
|
||||||
|
return {};
|
||||||
|
}
|
||||||
|
|
||||||
|
try {
|
||||||
|
const response = await axios.get<{ data: LMStudioModel[] }>(`${lmStudioEndpoint}/models`, {
|
||||||
|
headers: {
|
||||||
|
'Content-Type': 'application/json',
|
||||||
|
},
|
||||||
|
});
|
||||||
|
|
||||||
|
const lmStudioModels = response.data.data;
|
||||||
|
|
||||||
|
const chatModels = lmStudioModels.reduce<Record<string, ChatModelConfig>>((acc, model) => {
|
||||||
|
acc[model.id] = {
|
||||||
|
displayName: model.id,
|
||||||
|
model: new ChatOpenAI({
|
||||||
|
openAIApiKey: 'lm-studio',
|
||||||
|
configuration: {
|
||||||
|
baseURL: lmStudioEndpoint,
|
||||||
|
},
|
||||||
|
modelName: model.id,
|
||||||
|
temperature: 0.7,
|
||||||
|
}),
|
||||||
|
};
|
||||||
|
return acc;
|
||||||
|
}, {});
|
||||||
|
|
||||||
|
return chatModels;
|
||||||
|
} catch (err) {
|
||||||
|
logger.error(`Error loading LM Studio models: ${err}`);
|
||||||
|
return {};
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
|
export const loadLMStudioEmbeddingsModels = async () => {
|
||||||
|
const lmStudioEndpoint = getLMStudioApiEndpoint();
|
||||||
|
|
||||||
|
if (!lmStudioEndpoint) return {};
|
||||||
|
|
||||||
|
try {
|
||||||
|
const response = await axios.get(`${lmStudioEndpoint}/models`, {
|
||||||
|
headers: {
|
||||||
|
'Content-Type': 'application/json',
|
||||||
|
},
|
||||||
|
});
|
||||||
|
|
||||||
|
const lmStudioModels = response.data.data;
|
||||||
|
|
||||||
|
const embeddingsModels = lmStudioModels.reduce((acc, model) => {
|
||||||
|
acc[model.id] = {
|
||||||
|
displayName: model.id,
|
||||||
|
model: new OpenAIEmbeddings({
|
||||||
|
openAIApiKey: 'lm-studio', // Dummy key required by LangChain
|
||||||
|
configuration: {
|
||||||
|
baseURL: lmStudioEndpoint,
|
||||||
|
},
|
||||||
|
modelName: model.id,
|
||||||
|
}),
|
||||||
|
};
|
||||||
|
return acc;
|
||||||
|
}, {});
|
||||||
|
|
||||||
|
return embeddingsModels;
|
||||||
|
} catch (err) {
|
||||||
|
logger.error(`Error loading LM Studio embeddings model: ${err}`);
|
||||||
|
return {};
|
||||||
|
}
|
||||||
|
};
|
|
@ -6,6 +6,7 @@ import {
|
||||||
import {
|
import {
|
||||||
getGroqApiKey,
|
getGroqApiKey,
|
||||||
getOllamaApiEndpoint,
|
getOllamaApiEndpoint,
|
||||||
|
getLMStudioApiEndpoint,
|
||||||
getAnthropicApiKey,
|
getAnthropicApiKey,
|
||||||
getGeminiApiKey,
|
getGeminiApiKey,
|
||||||
getOpenaiApiKey,
|
getOpenaiApiKey,
|
||||||
|
@ -51,6 +52,7 @@ router.get('/', async (_, res) => {
|
||||||
|
|
||||||
config['openaiApiKey'] = getOpenaiApiKey();
|
config['openaiApiKey'] = getOpenaiApiKey();
|
||||||
config['ollamaApiUrl'] = getOllamaApiEndpoint();
|
config['ollamaApiUrl'] = getOllamaApiEndpoint();
|
||||||
|
config['lmStudioApiUrl'] = getLMStudioApiEndpoint();
|
||||||
config['anthropicApiKey'] = getAnthropicApiKey();
|
config['anthropicApiKey'] = getAnthropicApiKey();
|
||||||
config['groqApiKey'] = getGroqApiKey();
|
config['groqApiKey'] = getGroqApiKey();
|
||||||
config['geminiApiKey'] = getGeminiApiKey();
|
config['geminiApiKey'] = getGeminiApiKey();
|
||||||
|
@ -74,6 +76,7 @@ router.post('/', async (req, res) => {
|
||||||
},
|
},
|
||||||
API_ENDPOINTS: {
|
API_ENDPOINTS: {
|
||||||
OLLAMA: config.ollamaApiUrl,
|
OLLAMA: config.ollamaApiUrl,
|
||||||
|
LMSTUDIO: config.lmStudioApiUrl,
|
||||||
},
|
},
|
||||||
};
|
};
|
||||||
|
|
||||||
|
|
|
@ -11,6 +11,8 @@ import {
|
||||||
StopCircle,
|
StopCircle,
|
||||||
Layers3,
|
Layers3,
|
||||||
Plus,
|
Plus,
|
||||||
|
Brain,
|
||||||
|
ChevronDown,
|
||||||
} from 'lucide-react';
|
} from 'lucide-react';
|
||||||
import Markdown from 'markdown-to-jsx';
|
import Markdown from 'markdown-to-jsx';
|
||||||
import Copy from './MessageActions/Copy';
|
import Copy from './MessageActions/Copy';
|
||||||
|
@ -41,26 +43,48 @@ const MessageBox = ({
|
||||||
}) => {
|
}) => {
|
||||||
const [parsedMessage, setParsedMessage] = useState(message.content);
|
const [parsedMessage, setParsedMessage] = useState(message.content);
|
||||||
const [speechMessage, setSpeechMessage] = useState(message.content);
|
const [speechMessage, setSpeechMessage] = useState(message.content);
|
||||||
|
const [thinking, setThinking] = useState<string>('');
|
||||||
|
const [answer, setAnswer] = useState<string>('');
|
||||||
|
const [isThinkingExpanded, setIsThinkingExpanded] = useState(false);
|
||||||
|
|
||||||
useEffect(() => {
|
useEffect(() => {
|
||||||
const regex = /\[(\d+)\]/g;
|
const regex = /\[(\d+)\]/g;
|
||||||
|
|
||||||
if (
|
// First check for thinking content
|
||||||
message.role === 'assistant' &&
|
const match = message.content.match(/<think>(.*?)<\/think>(.*)/s);
|
||||||
message?.sources &&
|
if (match) {
|
||||||
message.sources.length > 0
|
const [_, thinkingContent, answerContent] = match;
|
||||||
) {
|
setThinking(thinkingContent.trim());
|
||||||
return setParsedMessage(
|
setAnswer(answerContent.trim());
|
||||||
|
|
||||||
|
// Process the answer part for sources if needed
|
||||||
|
if (message.role === 'assistant' && message?.sources && message.sources.length > 0) {
|
||||||
|
setParsedMessage(
|
||||||
|
answerContent.trim().replace(
|
||||||
|
regex,
|
||||||
|
(_, number) =>
|
||||||
|
`<a href="${message.sources?.[number - 1]?.metadata?.url}" target="_blank" className="bg-light-secondary dark:bg-dark-secondary px-1 rounded ml-1 no-underline text-xs text-black/70 dark:text-white/70 relative">${number}</a>`,
|
||||||
|
),
|
||||||
|
);
|
||||||
|
} else {
|
||||||
|
setParsedMessage(answerContent.trim());
|
||||||
|
}
|
||||||
|
setSpeechMessage(answerContent.trim().replace(regex, ''));
|
||||||
|
} else {
|
||||||
|
// No thinking content - process as before
|
||||||
|
if (message.role === 'assistant' && message?.sources && message.sources.length > 0) {
|
||||||
|
setParsedMessage(
|
||||||
message.content.replace(
|
message.content.replace(
|
||||||
regex,
|
regex,
|
||||||
(_, number) =>
|
(_, number) =>
|
||||||
`<a href="${message.sources?.[number - 1]?.metadata?.url}" target="_blank" className="bg-light-secondary dark:bg-dark-secondary px-1 rounded ml-1 no-underline text-xs text-black/70 dark:text-white/70 relative">${number}</a>`,
|
`<a href="${message.sources?.[number - 1]?.metadata?.url}" target="_blank" className="bg-light-secondary dark:bg-dark-secondary px-1 rounded ml-1 no-underline text-xs text-black/70 dark:text-white/70 relative">${number}</a>`,
|
||||||
),
|
),
|
||||||
);
|
);
|
||||||
}
|
} else {
|
||||||
|
|
||||||
setSpeechMessage(message.content.replace(regex, ''));
|
|
||||||
setParsedMessage(message.content);
|
setParsedMessage(message.content);
|
||||||
|
}
|
||||||
|
setSpeechMessage(message.content.replace(regex, ''));
|
||||||
|
}
|
||||||
}, [message.content, message.sources, message.role]);
|
}, [message.content, message.sources, message.role]);
|
||||||
|
|
||||||
const { speechStatus, start, stop } = useSpeech({ text: speechMessage });
|
const { speechStatus, start, stop } = useSpeech({ text: speechMessage });
|
||||||
|
@ -81,6 +105,37 @@ const MessageBox = ({
|
||||||
ref={dividerRef}
|
ref={dividerRef}
|
||||||
className="flex flex-col space-y-6 w-full lg:w-9/12"
|
className="flex flex-col space-y-6 w-full lg:w-9/12"
|
||||||
>
|
>
|
||||||
|
{thinking && (
|
||||||
|
<div className="flex flex-col space-y-2 mb-4">
|
||||||
|
<button
|
||||||
|
onClick={() => setIsThinkingExpanded(!isThinkingExpanded)}
|
||||||
|
className="flex flex-row items-center space-x-2 group text-black/70 dark:text-white/70 hover:text-black dark:hover:text-white transition duration-200"
|
||||||
|
>
|
||||||
|
<Brain size={20} />
|
||||||
|
<h3 className="font-medium text-xl">View Thinking</h3>
|
||||||
|
<ChevronDown
|
||||||
|
size={16}
|
||||||
|
className={cn(
|
||||||
|
"transition-transform duration-200",
|
||||||
|
isThinkingExpanded ? "rotate-180" : ""
|
||||||
|
)}
|
||||||
|
/>
|
||||||
|
</button>
|
||||||
|
|
||||||
|
{isThinkingExpanded && (
|
||||||
|
<div className="rounded-lg bg-light-secondary/50 dark:bg-dark-secondary/50 p-4">
|
||||||
|
<Markdown
|
||||||
|
className={cn(
|
||||||
|
'prose dark:prose-invert text-sm leading-relaxed',
|
||||||
|
'max-w-none break-words'
|
||||||
|
)}
|
||||||
|
>
|
||||||
|
{thinking}
|
||||||
|
</Markdown>
|
||||||
|
</div>
|
||||||
|
)}
|
||||||
|
</div>
|
||||||
|
)}
|
||||||
{message.sources && message.sources.length > 0 && (
|
{message.sources && message.sources.length > 0 && (
|
||||||
<div className="flex flex-col space-y-2">
|
<div className="flex flex-col space-y-2">
|
||||||
<div className="flex flex-row items-center space-x-2">
|
<div className="flex flex-row items-center space-x-2">
|
||||||
|
@ -199,4 +254,3 @@ const MessageBox = ({
|
||||||
);
|
);
|
||||||
};
|
};
|
||||||
|
|
||||||
export default MessageBox;
|
|
||||||
|
|
Loading…
Add table
Reference in a new issue