mirror of
https://github.com/ItzCrazyKns/Perplexica.git
synced 2025-07-18 14:38:32 +00:00
Add DeepSeek and LMStudio providers
- Integrate DeepSeek and LMStudio AI providers - Add message processing utilities for improved handling - Implement reasoning panel for message actions - Add logging functionality to UI - Update configurations and dependencies
This commit is contained in:
haddadrm
251
src/lib/deepseekChat.ts
Normal file
251
src/lib/deepseekChat.ts
Normal file
@ -0,0 +1,251 @@
|
||||
import { BaseChatModel, BaseChatModelCallOptions } from '@langchain/core/language_models/chat_models';
|
||||
import { CallbackManagerForLLMRun } from '@langchain/core/callbacks/manager';
|
||||
import { AIMessage, AIMessageChunk, BaseMessage, HumanMessage, SystemMessage } from '@langchain/core/messages';
|
||||
import { ChatResult, ChatGenerationChunk } from '@langchain/core/outputs';
|
||||
import axios from 'axios';
|
||||
|
||||
import { BaseChatModelParams } from '@langchain/core/language_models/chat_models';
|
||||
|
||||
interface DeepSeekChatParams extends BaseChatModelParams {
|
||||
apiKey: string;
|
||||
baseURL: string;
|
||||
modelName: string;
|
||||
temperature?: number;
|
||||
max_tokens?: number;
|
||||
top_p?: number;
|
||||
frequency_penalty?: number;
|
||||
presence_penalty?: number;
|
||||
}
|
||||
|
||||
export class DeepSeekChat extends BaseChatModel<BaseChatModelCallOptions & { stream?: boolean }> {
|
||||
private apiKey: string;
|
||||
private baseURL: string;
|
||||
private modelName: string;
|
||||
private temperature: number;
|
||||
private maxTokens: number;
|
||||
private topP: number;
|
||||
private frequencyPenalty: number;
|
||||
private presencePenalty: number;
|
||||
|
||||
constructor(params: DeepSeekChatParams) {
|
||||
super(params);
|
||||
this.apiKey = params.apiKey;
|
||||
this.baseURL = params.baseURL;
|
||||
this.modelName = params.modelName;
|
||||
this.temperature = params.temperature ?? 0.7;
|
||||
this.maxTokens = params.max_tokens ?? 8192;
|
||||
this.topP = params.top_p ?? 1;
|
||||
this.frequencyPenalty = params.frequency_penalty ?? 0;
|
||||
this.presencePenalty = params.presence_penalty ?? 0;
|
||||
}
|
||||
|
||||
async _generate(
|
||||
messages: BaseMessage[],
|
||||
options: this['ParsedCallOptions'],
|
||||
runManager?: CallbackManagerForLLMRun
|
||||
): Promise<ChatResult> {
|
||||
const formattedMessages = messages.map(msg => ({
|
||||
role: this.getRole(msg),
|
||||
content: msg.content.toString(),
|
||||
}));
|
||||
const response = await this.callDeepSeekAPI(formattedMessages, options.stream);
|
||||
|
||||
if (options.stream) {
|
||||
return this.processStreamingResponse(response, messages, options, runManager);
|
||||
} else {
|
||||
const choice = response.data.choices[0];
|
||||
let content = choice.message.content || '';
|
||||
if (choice.message.reasoning_content) {
|
||||
content = `<think>\n${choice.message.reasoning_content}\n</think>\n\n${content}`;
|
||||
}
|
||||
|
||||
// Report usage stats if available
|
||||
if (response.data.usage && runManager) {
|
||||
runManager.handleLLMEnd({
|
||||
generations: [],
|
||||
llmOutput: {
|
||||
tokenUsage: {
|
||||
completionTokens: response.data.usage.completion_tokens,
|
||||
promptTokens: response.data.usage.prompt_tokens,
|
||||
totalTokens: response.data.usage.total_tokens
|
||||
}
|
||||
}
|
||||
});
|
||||
}
|
||||
return {
|
||||
generations: [
|
||||
{
|
||||
text: content,
|
||||
message: new AIMessage(content),
|
||||
},
|
||||
],
|
||||
};
|
||||
}
|
||||
}
|
||||
|
||||
private getRole(msg: BaseMessage): string {
|
||||
if (msg instanceof SystemMessage) return 'system';
|
||||
if (msg instanceof HumanMessage) return 'user';
|
||||
if (msg instanceof AIMessage) return 'assistant';
|
||||
return 'user'; // Default to user
|
||||
}
|
||||
|
||||
private async callDeepSeekAPI(messages: Array<{ role: string; content: string }>, streaming?: boolean) {
|
||||
return axios.post(
|
||||
`${this.baseURL}/chat/completions`,
|
||||
{
|
||||
messages,
|
||||
model: this.modelName,
|
||||
stream: streaming,
|
||||
temperature: this.temperature,
|
||||
max_tokens: this.maxTokens,
|
||||
top_p: this.topP,
|
||||
frequency_penalty: this.frequencyPenalty,
|
||||
presence_penalty: this.presencePenalty,
|
||||
response_format: { type: 'text' },
|
||||
...(streaming && {
|
||||
stream_options: {
|
||||
include_usage: true
|
||||
}
|
||||
})
|
||||
},
|
||||
{
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
'Authorization': `Bearer ${this.apiKey}`,
|
||||
},
|
||||
responseType: streaming ? 'text' : 'json',
|
||||
}
|
||||
);
|
||||
}
|
||||
|
||||
public async *_streamResponseChunks(messages: BaseMessage[], options: this['ParsedCallOptions'], runManager?: CallbackManagerForLLMRun) {
|
||||
const response = await this.callDeepSeekAPI(messages.map(msg => ({
|
||||
role: this.getRole(msg),
|
||||
content: msg.content.toString(),
|
||||
})), true);
|
||||
|
||||
let thinkState = -1; // -1: not started, 0: thinking, 1: answered
|
||||
let currentContent = '';
|
||||
|
||||
// Split the response into lines
|
||||
const lines = response.data.split('\n');
|
||||
for (const line of lines) {
|
||||
if (!line.startsWith('data: ')) continue;
|
||||
const jsonStr = line.slice(6);
|
||||
if (jsonStr === '[DONE]') break;
|
||||
|
||||
try {
|
||||
console.log('Received chunk:', jsonStr);
|
||||
const chunk = JSON.parse(jsonStr);
|
||||
const delta = chunk.choices[0].delta;
|
||||
console.log('Parsed delta:', delta);
|
||||
|
||||
// Handle usage stats in final chunk
|
||||
if (chunk.usage && !chunk.choices?.length) {
|
||||
runManager?.handleLLMEnd?.({
|
||||
generations: [],
|
||||
llmOutput: {
|
||||
tokenUsage: {
|
||||
completionTokens: chunk.usage.completion_tokens,
|
||||
promptTokens: chunk.usage.prompt_tokens,
|
||||
totalTokens: chunk.usage.total_tokens
|
||||
}
|
||||
}
|
||||
});
|
||||
continue;
|
||||
}
|
||||
|
||||
// Handle reasoning content
|
||||
if (delta.reasoning_content) {
|
||||
if (thinkState === -1) {
|
||||
thinkState = 0;
|
||||
const startTag = '<think>\n';
|
||||
currentContent += startTag;
|
||||
console.log('Emitting think start:', startTag);
|
||||
runManager?.handleLLMNewToken(startTag);
|
||||
const chunk = new ChatGenerationChunk({
|
||||
text: startTag,
|
||||
message: new AIMessageChunk(startTag),
|
||||
generationInfo: {}
|
||||
});
|
||||
yield chunk;
|
||||
}
|
||||
currentContent += delta.reasoning_content;
|
||||
console.log('Emitting reasoning:', delta.reasoning_content);
|
||||
runManager?.handleLLMNewToken(delta.reasoning_content);
|
||||
const chunk = new ChatGenerationChunk({
|
||||
text: delta.reasoning_content,
|
||||
message: new AIMessageChunk(delta.reasoning_content),
|
||||
generationInfo: {}
|
||||
});
|
||||
yield chunk;
|
||||
}
|
||||
|
||||
// Handle regular content
|
||||
if (delta.content) {
|
||||
if (thinkState === 0) {
|
||||
thinkState = 1;
|
||||
const endTag = '\n</think>\n\n';
|
||||
currentContent += endTag;
|
||||
console.log('Emitting think end:', endTag);
|
||||
runManager?.handleLLMNewToken(endTag);
|
||||
const chunk = new ChatGenerationChunk({
|
||||
text: endTag,
|
||||
message: new AIMessageChunk(endTag),
|
||||
generationInfo: {}
|
||||
});
|
||||
yield chunk;
|
||||
}
|
||||
currentContent += delta.content;
|
||||
console.log('Emitting content:', delta.content);
|
||||
runManager?.handleLLMNewToken(delta.content);
|
||||
const chunk = new ChatGenerationChunk({
|
||||
text: delta.content,
|
||||
message: new AIMessageChunk(delta.content),
|
||||
generationInfo: {}
|
||||
});
|
||||
yield chunk;
|
||||
}
|
||||
} catch (error) {
|
||||
const errorMessage = error instanceof Error ? error.message : 'Failed to parse chunk';
|
||||
console.error(`Streaming error: ${errorMessage}`);
|
||||
if (error instanceof Error && error.message.includes('DeepSeek API Error')) {
|
||||
throw error;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Handle any unclosed think block
|
||||
if (thinkState === 0) {
|
||||
const endTag = '\n</think>\n\n';
|
||||
currentContent += endTag;
|
||||
runManager?.handleLLMNewToken(endTag);
|
||||
const chunk = new ChatGenerationChunk({
|
||||
text: endTag,
|
||||
message: new AIMessageChunk(endTag),
|
||||
generationInfo: {}
|
||||
});
|
||||
yield chunk;
|
||||
}
|
||||
}
|
||||
|
||||
private async processStreamingResponse(response: any, messages: BaseMessage[], options: this['ParsedCallOptions'], runManager?: CallbackManagerForLLMRun): Promise<ChatResult> {
|
||||
let accumulatedContent = '';
|
||||
for await (const chunk of this._streamResponseChunks(messages, options, runManager)) {
|
||||
accumulatedContent += chunk.message.content;
|
||||
}
|
||||
return {
|
||||
generations: [
|
||||
{
|
||||
text: accumulatedContent,
|
||||
message: new AIMessage(accumulatedContent),
|
||||
},
|
||||
],
|
||||
};
|
||||
}
|
||||
|
||||
_llmType(): string {
|
||||
return 'deepseek';
|
||||
}
|
||||
}
|
||||
69
src/lib/providers/deepseek.ts
Normal file
69
src/lib/providers/deepseek.ts
Normal file
@ -0,0 +1,69 @@
|
||||
import { DeepSeekChat } from '../deepseekChat';
|
||||
import logger from '../../utils/logger';
|
||||
import { getDeepseekApiKey } from '../../config';
|
||||
import axios from 'axios';
|
||||
|
||||
interface DeepSeekModel {
|
||||
id: string;
|
||||
object: string;
|
||||
owned_by: string;
|
||||
}
|
||||
|
||||
interface ModelListResponse {
|
||||
object: 'list';
|
||||
data: DeepSeekModel[];
|
||||
}
|
||||
|
||||
interface ChatModelConfig {
|
||||
displayName: string;
|
||||
model: DeepSeekChat;
|
||||
}
|
||||
|
||||
const MODEL_DISPLAY_NAMES: Record<string, string> = {
|
||||
'deepseek-reasoner': 'DeepSeek R1',
|
||||
'deepseek-chat': 'DeepSeek V3'
|
||||
};
|
||||
|
||||
export const loadDeepSeekChatModels = async (): Promise<Record<string, ChatModelConfig>> => {
|
||||
const deepSeekEndpoint = 'https://api.deepseek.com';
|
||||
|
||||
const apiKey = getDeepseekApiKey();
|
||||
if (!apiKey) return {};
|
||||
|
||||
if (!deepSeekEndpoint || !apiKey) {
|
||||
logger.debug('DeepSeek endpoint or API key not configured, skipping');
|
||||
return {};
|
||||
}
|
||||
|
||||
try {
|
||||
const response = await axios.get<{ data: DeepSeekModel[] }>(`${deepSeekEndpoint}/models`, {
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
'Authorization': `Bearer ${apiKey}`,
|
||||
},
|
||||
});
|
||||
|
||||
const deepSeekModels = response.data.data;
|
||||
|
||||
const chatModels = deepSeekModels.reduce<Record<string, ChatModelConfig>>((acc, model) => {
|
||||
// Only include models we have display names for
|
||||
if (model.id in MODEL_DISPLAY_NAMES) {
|
||||
acc[model.id] = {
|
||||
displayName: MODEL_DISPLAY_NAMES[model.id],
|
||||
model: new DeepSeekChat({
|
||||
apiKey,
|
||||
baseURL: deepSeekEndpoint,
|
||||
modelName: model.id,
|
||||
temperature: 0.7,
|
||||
}),
|
||||
};
|
||||
}
|
||||
return acc;
|
||||
}, {});
|
||||
|
||||
return chatModels;
|
||||
} catch (err) {
|
||||
logger.error(`Error loading DeepSeek models: ${String(err)}`);
|
||||
return {};
|
||||
}
|
||||
};
|
||||
@ -4,6 +4,8 @@ import { loadOpenAIChatModels, loadOpenAIEmbeddingsModels } from './openai';
|
||||
import { loadAnthropicChatModels } from './anthropic';
|
||||
import { loadTransformersEmbeddingsModels } from './transformers';
|
||||
import { loadGeminiChatModels, loadGeminiEmbeddingsModels } from './gemini';
|
||||
import { loadDeepSeekChatModels } from './deepseek';
|
||||
import { loadLMStudioChatModels, loadLMStudioEmbeddingsModels } from './lmstudio';
|
||||
import {
|
||||
getCustomOpenaiApiKey,
|
||||
getCustomOpenaiApiUrl,
|
||||
@ -17,6 +19,8 @@ const chatModelProviders = {
|
||||
ollama: loadOllamaChatModels,
|
||||
anthropic: loadAnthropicChatModels,
|
||||
gemini: loadGeminiChatModels,
|
||||
deepseek: loadDeepSeekChatModels,
|
||||
lm_studio: loadLMStudioChatModels,
|
||||
};
|
||||
|
||||
const embeddingModelProviders = {
|
||||
@ -24,6 +28,7 @@ const embeddingModelProviders = {
|
||||
local: loadTransformersEmbeddingsModels,
|
||||
ollama: loadOllamaEmbeddingsModels,
|
||||
gemini: loadGeminiEmbeddingsModels,
|
||||
lm_studio: loadLMStudioEmbeddingsModels,
|
||||
};
|
||||
|
||||
export const getAvailableChatModelProviders = async () => {
|
||||
|
||||
96
src/lib/providers/lmstudio.ts
Normal file
96
src/lib/providers/lmstudio.ts
Normal file
@ -0,0 +1,96 @@
|
||||
import { ChatOpenAI, OpenAIEmbeddings } from '@langchain/openai';
|
||||
import { getLMStudioApiEndpoint, getKeepAlive } from '../../config';
|
||||
import logger from '../../utils/logger';
|
||||
import axios from 'axios';
|
||||
|
||||
interface LMStudioModel {
|
||||
id: string;
|
||||
name?: string;
|
||||
}
|
||||
|
||||
const ensureV1Endpoint = (endpoint: string): string =>
|
||||
endpoint.endsWith('/v1') ? endpoint : `${endpoint}/v1`;
|
||||
|
||||
const checkServerAvailability = async (endpoint: string): Promise<boolean> => {
|
||||
try {
|
||||
const keepAlive = getKeepAlive();
|
||||
await axios.get(`${ensureV1Endpoint(endpoint)}/models`, {
|
||||
timeout: parseInt(keepAlive) * 1000 || 5000,
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
});
|
||||
return true;
|
||||
} catch {
|
||||
return false;
|
||||
}
|
||||
};
|
||||
|
||||
export const loadLMStudioChatModels = async () => {
|
||||
const endpoint = getLMStudioApiEndpoint();
|
||||
const keepAlive = getKeepAlive();
|
||||
|
||||
if (!endpoint) return {};
|
||||
if (!await checkServerAvailability(endpoint)) return {};
|
||||
|
||||
try {
|
||||
const response = await axios.get(`${ensureV1Endpoint(endpoint)}/models`, {
|
||||
timeout: parseInt(keepAlive) * 1000 || 5000,
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
});
|
||||
|
||||
const chatModels = response.data.data.reduce((acc: Record<string, any>, model: LMStudioModel) => {
|
||||
acc[model.id] = {
|
||||
displayName: model.name || model.id,
|
||||
model: new ChatOpenAI({
|
||||
openAIApiKey: 'lm-studio',
|
||||
configuration: {
|
||||
baseURL: ensureV1Endpoint(endpoint),
|
||||
},
|
||||
modelName: model.id,
|
||||
temperature: 0.7,
|
||||
streaming: true,
|
||||
maxRetries: 3
|
||||
}),
|
||||
};
|
||||
return acc;
|
||||
}, {});
|
||||
|
||||
return chatModels;
|
||||
} catch (err) {
|
||||
logger.error(`Error loading LM Studio models: ${err}`);
|
||||
return {};
|
||||
}
|
||||
};
|
||||
|
||||
export const loadLMStudioEmbeddingsModels = async () => {
|
||||
const endpoint = getLMStudioApiEndpoint();
|
||||
const keepAlive = getKeepAlive();
|
||||
|
||||
if (!endpoint) return {};
|
||||
if (!await checkServerAvailability(endpoint)) return {};
|
||||
|
||||
try {
|
||||
const response = await axios.get(`${ensureV1Endpoint(endpoint)}/models`, {
|
||||
timeout: parseInt(keepAlive) * 1000 || 5000,
|
||||
headers: { 'Content-Type': 'application/json' },
|
||||
});
|
||||
|
||||
const embeddingsModels = response.data.data.reduce((acc: Record<string, any>, model: LMStudioModel) => {
|
||||
acc[model.id] = {
|
||||
displayName: model.name || model.id,
|
||||
model: new OpenAIEmbeddings({
|
||||
openAIApiKey: 'lm-studio',
|
||||
configuration: {
|
||||
baseURL: ensureV1Endpoint(endpoint),
|
||||
},
|
||||
modelName: model.id,
|
||||
}),
|
||||
};
|
||||
return acc;
|
||||
}, {});
|
||||
|
||||
return embeddingsModels;
|
||||
} catch (err) {
|
||||
logger.error(`Error loading LM Studio embeddings model: ${err}`);
|
||||
return {};
|
||||
}
|
||||
};
|
||||
Reference in New Issue
Block a user