anything-llm/server/utils/AiProviders/perplexity/index.js

142 lines
3.9 KiB
JavaScript
Raw Normal View History

const { NativeEmbedder } = require("../../EmbeddingEngines/native");
const {
handleDefaultStreamResponseV2,
} = require("../../helpers/chat/responses");
function perplexityModels() {
const { MODELS } = require("./models.js");
return MODELS || {};
}
class PerplexityLLM {
constructor(embedder = null, modelPreference = null) {
if (!process.env.PERPLEXITY_API_KEY)
throw new Error("No Perplexity API key was set.");
const { OpenAI: OpenAIApi } = require("openai");
this.openai = new OpenAIApi({
baseURL: "https://api.perplexity.ai",
apiKey: process.env.PERPLEXITY_API_KEY ?? null,
});
this.model =
modelPreference ||
process.env.PERPLEXITY_MODEL_PREF ||
"sonar-small-online"; // Give at least a unique model to the provider as last fallback.
this.limits = {
history: this.promptWindowLimit() * 0.15,
system: this.promptWindowLimit() * 0.15,
user: this.promptWindowLimit() * 0.7,
};
this.embedder = embedder ?? new NativeEmbedder();
this.defaultTemp = 0.7;
}
#appendContext(contextTexts = []) {
if (!contextTexts || !contextTexts.length) return "";
return (
"\nContext:\n" +
contextTexts
.map((text, i) => {
return `[CONTEXT ${i}]:\n${text}\n[END CONTEXT ${i}]\n\n`;
})
.join("")
);
}
allModelInformation() {
return perplexityModels();
}
streamingEnabled() {
return "streamGetChatCompletion" in this;
}
promptWindowLimit() {
const availableModels = this.allModelInformation();
return availableModels[this.model]?.maxLength || 4096;
}
async isValidChatCompletionModel(model = "") {
const availableModels = this.allModelInformation();
return availableModels.hasOwnProperty(model);
}
constructPrompt({
systemPrompt = "",
contextTexts = [],
chatHistory = [],
userPrompt = "",
}) {
const prompt = {
role: "system",
content: `${systemPrompt}${this.#appendContext(contextTexts)}`,
};
return [prompt, ...chatHistory, { role: "user", content: userPrompt }];
}
async isSafe(_input = "") {
// Not implemented so must be stubbed
return { safe: true, reasons: [] };
}
async getChatCompletion(messages = null, { temperature = 0.7 }) {
if (!(await this.isValidChatCompletionModel(this.model)))
throw new Error(
`Perplexity chat: ${this.model} is not valid for chat completion!`
);
const result = await this.openai.chat.completions
.create({
model: this.model,
messages,
temperature,
})
.catch((e) => {
throw new Error(e.response.data.error.message);
});
if (!result.hasOwnProperty("choices") || result.choices.length === 0)
return null;
return result.choices[0].message.content;
}
async streamGetChatCompletion(messages = null, { temperature = 0.7 }) {
if (!(await this.isValidChatCompletionModel(this.model)))
throw new Error(
`Perplexity chat: ${this.model} is not valid for chat completion!`
);
const streamRequest = await this.openai.chat.completions.create({
model: this.model,
stream: true,
messages,
temperature,
});
return streamRequest;
}
handleStream(response, stream, responseProps) {
return handleDefaultStreamResponseV2(response, stream, responseProps);
}
// Simple wrapper for dynamic embedder & normalize interface for all LLM implementations
async embedTextInput(textInput) {
return await this.embedder.embedTextInput(textInput);
}
async embedChunks(textChunks = []) {
return await this.embedder.embedChunks(textChunks);
}
async compressMessages(promptArgs = {}, rawHistory = []) {
const { messageArrayCompressor } = require("../../helpers/chat");
const messageArray = this.constructPrompt(promptArgs);
return await messageArrayCompressor(this, messageArray, rawHistory);
}
}
module.exports = {
PerplexityLLM,
perplexityModels,
};