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

194 lines
5.6 KiB
JavaScript
Raw Normal View History

Using OpenAI API locally (#335) * Using OpenAI API locally * Infinite prompt input and compression implementation (#332) * WIP on continuous prompt window summary * wip * Move chat out of VDB simplify chat interface normalize LLM model interface have compression abstraction Cleanup compressor TODO: Anthropic stuff * Implement compression for Anythropic Fix lancedb sources * cleanup vectorDBs and check that lance, chroma, and pinecone are returning valid metadata sources * Resolve Weaviate citation sources not working with schema * comment cleanup * disable import on hosted instances (#339) * disable import on hosted instances * Update UI on disabled import/export --------- Co-authored-by: timothycarambat <rambat1010@gmail.com> * Add support for gpt-4-turbo 128K model (#340) resolves #336 Add support for gpt-4-turbo 128K model * 315 show citations based on relevancy score (#316) * settings for similarity score threshold and prisma schema updated * prisma schema migration for adding similarityScore setting * WIP * Min score default change * added similarityThreshold checking for all vectordb providers * linting --------- Co-authored-by: shatfield4 <seanhatfield5@gmail.com> * rename localai to lmstudio * forgot files that were renamed * normalize model interface * add model and context window limits * update LMStudio tagline * Fully working LMStudio integration --------- Co-authored-by: Francisco Bischoff <984592+franzbischoff@users.noreply.github.com> Co-authored-by: Timothy Carambat <rambat1010@gmail.com> Co-authored-by: Sean Hatfield <seanhatfield5@gmail.com>
2023-11-09 21:33:21 +01:00
const { chatPrompt } = require("../../chats");
// hybrid of openAi LLM chat completion for LMStudio
class LMStudioLLM {
constructor(embedder = null) {
if (!process.env.LMSTUDIO_BASE_PATH)
throw new Error("No LMStudio API Base Path was set.");
const { Configuration, OpenAIApi } = require("openai");
const config = new Configuration({
basePath: process.env.LMSTUDIO_BASE_PATH?.replace(/\/+$/, ""), // here is the URL to your LMStudio instance
});
this.lmstudio = new OpenAIApi(config);
// When using LMStudios inference server - the model param is not required so
// we can stub it here.
this.model = "model-placeholder";
this.limits = {
history: this.promptWindowLimit() * 0.15,
system: this.promptWindowLimit() * 0.15,
user: this.promptWindowLimit() * 0.7,
};
if (!embedder)
throw new Error(
"INVALID LM STUDIO SETUP. No embedding engine has been set. Go to instance settings and set up an embedding interface to use LMStudio as your LLM."
);
this.embedder = embedder;
}
#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("")
);
}
streamingEnabled() {
return "streamChat" in this && "streamGetChatCompletion" in this;
}
Using OpenAI API locally (#335) * Using OpenAI API locally * Infinite prompt input and compression implementation (#332) * WIP on continuous prompt window summary * wip * Move chat out of VDB simplify chat interface normalize LLM model interface have compression abstraction Cleanup compressor TODO: Anthropic stuff * Implement compression for Anythropic Fix lancedb sources * cleanup vectorDBs and check that lance, chroma, and pinecone are returning valid metadata sources * Resolve Weaviate citation sources not working with schema * comment cleanup * disable import on hosted instances (#339) * disable import on hosted instances * Update UI on disabled import/export --------- Co-authored-by: timothycarambat <rambat1010@gmail.com> * Add support for gpt-4-turbo 128K model (#340) resolves #336 Add support for gpt-4-turbo 128K model * 315 show citations based on relevancy score (#316) * settings for similarity score threshold and prisma schema updated * prisma schema migration for adding similarityScore setting * WIP * Min score default change * added similarityThreshold checking for all vectordb providers * linting --------- Co-authored-by: shatfield4 <seanhatfield5@gmail.com> * rename localai to lmstudio * forgot files that were renamed * normalize model interface * add model and context window limits * update LMStudio tagline * Fully working LMStudio integration --------- Co-authored-by: Francisco Bischoff <984592+franzbischoff@users.noreply.github.com> Co-authored-by: Timothy Carambat <rambat1010@gmail.com> Co-authored-by: Sean Hatfield <seanhatfield5@gmail.com>
2023-11-09 21:33:21 +01:00
// Ensure the user set a value for the token limit
// and if undefined - assume 4096 window.
promptWindowLimit() {
const limit = process.env.LMSTUDIO_MODEL_TOKEN_LIMIT || 4096;
if (!limit || isNaN(Number(limit)))
throw new Error("No LMStudio token context limit was set.");
return Number(limit);
}
async isValidChatCompletionModel(_ = "") {
// LMStudio may be anything. The user must do it correctly.
// See comment about this.model declaration in constructor
return true;
}
constructPrompt({
systemPrompt = "",
contextTexts = [],
chatHistory = [],
userPrompt = "",
}) {
const prompt = {
role: "system",
content: `${systemPrompt}${this.#appendContext(contextTexts)}`,
Using OpenAI API locally (#335) * Using OpenAI API locally * Infinite prompt input and compression implementation (#332) * WIP on continuous prompt window summary * wip * Move chat out of VDB simplify chat interface normalize LLM model interface have compression abstraction Cleanup compressor TODO: Anthropic stuff * Implement compression for Anythropic Fix lancedb sources * cleanup vectorDBs and check that lance, chroma, and pinecone are returning valid metadata sources * Resolve Weaviate citation sources not working with schema * comment cleanup * disable import on hosted instances (#339) * disable import on hosted instances * Update UI on disabled import/export --------- Co-authored-by: timothycarambat <rambat1010@gmail.com> * Add support for gpt-4-turbo 128K model (#340) resolves #336 Add support for gpt-4-turbo 128K model * 315 show citations based on relevancy score (#316) * settings for similarity score threshold and prisma schema updated * prisma schema migration for adding similarityScore setting * WIP * Min score default change * added similarityThreshold checking for all vectordb providers * linting --------- Co-authored-by: shatfield4 <seanhatfield5@gmail.com> * rename localai to lmstudio * forgot files that were renamed * normalize model interface * add model and context window limits * update LMStudio tagline * Fully working LMStudio integration --------- Co-authored-by: Francisco Bischoff <984592+franzbischoff@users.noreply.github.com> Co-authored-by: Timothy Carambat <rambat1010@gmail.com> Co-authored-by: Sean Hatfield <seanhatfield5@gmail.com>
2023-11-09 21:33:21 +01:00
};
return [prompt, ...chatHistory, { role: "user", content: userPrompt }];
}
async isSafe(_input = "") {
// Not implemented so must be stubbed
return { safe: true, reasons: [] };
}
async sendChat(chatHistory = [], prompt, workspace = {}, rawHistory = []) {
if (!this.model)
throw new Error(
`LMStudio chat: ${this.model} is not valid or defined for chat completion!`
Using OpenAI API locally (#335) * Using OpenAI API locally * Infinite prompt input and compression implementation (#332) * WIP on continuous prompt window summary * wip * Move chat out of VDB simplify chat interface normalize LLM model interface have compression abstraction Cleanup compressor TODO: Anthropic stuff * Implement compression for Anythropic Fix lancedb sources * cleanup vectorDBs and check that lance, chroma, and pinecone are returning valid metadata sources * Resolve Weaviate citation sources not working with schema * comment cleanup * disable import on hosted instances (#339) * disable import on hosted instances * Update UI on disabled import/export --------- Co-authored-by: timothycarambat <rambat1010@gmail.com> * Add support for gpt-4-turbo 128K model (#340) resolves #336 Add support for gpt-4-turbo 128K model * 315 show citations based on relevancy score (#316) * settings for similarity score threshold and prisma schema updated * prisma schema migration for adding similarityScore setting * WIP * Min score default change * added similarityThreshold checking for all vectordb providers * linting --------- Co-authored-by: shatfield4 <seanhatfield5@gmail.com> * rename localai to lmstudio * forgot files that were renamed * normalize model interface * add model and context window limits * update LMStudio tagline * Fully working LMStudio integration --------- Co-authored-by: Francisco Bischoff <984592+franzbischoff@users.noreply.github.com> Co-authored-by: Timothy Carambat <rambat1010@gmail.com> Co-authored-by: Sean Hatfield <seanhatfield5@gmail.com>
2023-11-09 21:33:21 +01:00
);
const textResponse = await this.lmstudio
.createChatCompletion({
model: this.model,
temperature: Number(workspace?.openAiTemp ?? 0.7),
n: 1,
messages: await this.compressMessages(
{
systemPrompt: chatPrompt(workspace),
userPrompt: prompt,
chatHistory,
},
rawHistory
),
})
.then((json) => {
const res = json.data;
if (!res.hasOwnProperty("choices"))
throw new Error("LMStudio chat: No results!");
if (res.choices.length === 0)
throw new Error("LMStudio chat: No results length!");
return res.choices[0].message.content;
})
.catch((error) => {
throw new Error(
`LMStudio::createChatCompletion failed with: ${error.message}`
);
});
return textResponse;
}
async streamChat(chatHistory = [], prompt, workspace = {}, rawHistory = []) {
if (!this.model)
throw new Error(
`LMStudio chat: ${this.model} is not valid or defined for chat completion!`
);
const streamRequest = await this.lmstudio.createChatCompletion(
{
model: this.model,
temperature: Number(workspace?.openAiTemp ?? 0.7),
n: 1,
stream: true,
messages: await this.compressMessages(
{
systemPrompt: chatPrompt(workspace),
userPrompt: prompt,
chatHistory,
},
rawHistory
),
},
{ responseType: "stream" }
);
return streamRequest;
}
Using OpenAI API locally (#335) * Using OpenAI API locally * Infinite prompt input and compression implementation (#332) * WIP on continuous prompt window summary * wip * Move chat out of VDB simplify chat interface normalize LLM model interface have compression abstraction Cleanup compressor TODO: Anthropic stuff * Implement compression for Anythropic Fix lancedb sources * cleanup vectorDBs and check that lance, chroma, and pinecone are returning valid metadata sources * Resolve Weaviate citation sources not working with schema * comment cleanup * disable import on hosted instances (#339) * disable import on hosted instances * Update UI on disabled import/export --------- Co-authored-by: timothycarambat <rambat1010@gmail.com> * Add support for gpt-4-turbo 128K model (#340) resolves #336 Add support for gpt-4-turbo 128K model * 315 show citations based on relevancy score (#316) * settings for similarity score threshold and prisma schema updated * prisma schema migration for adding similarityScore setting * WIP * Min score default change * added similarityThreshold checking for all vectordb providers * linting --------- Co-authored-by: shatfield4 <seanhatfield5@gmail.com> * rename localai to lmstudio * forgot files that were renamed * normalize model interface * add model and context window limits * update LMStudio tagline * Fully working LMStudio integration --------- Co-authored-by: Francisco Bischoff <984592+franzbischoff@users.noreply.github.com> Co-authored-by: Timothy Carambat <rambat1010@gmail.com> Co-authored-by: Sean Hatfield <seanhatfield5@gmail.com>
2023-11-09 21:33:21 +01:00
async getChatCompletion(messages = null, { temperature = 0.7 }) {
if (!this.model)
throw new Error(
`LMStudio chat: ${this.model} is not valid or defined model for chat completion!`
);
const { data } = await this.lmstudio.createChatCompletion({
model: this.model,
messages,
temperature,
});
if (!data.hasOwnProperty("choices")) return null;
return data.choices[0].message.content;
}
async streamGetChatCompletion(messages = null, { temperature = 0.7 }) {
if (!this.model)
throw new Error(
`LMStudio chat: ${this.model} is not valid or defined model for chat completion!`
);
const streamRequest = await this.lmstudio.createChatCompletion(
{
model: this.model,
stream: true,
messages,
temperature,
},
{ responseType: "stream" }
);
return streamRequest;
}
Using OpenAI API locally (#335) * Using OpenAI API locally * Infinite prompt input and compression implementation (#332) * WIP on continuous prompt window summary * wip * Move chat out of VDB simplify chat interface normalize LLM model interface have compression abstraction Cleanup compressor TODO: Anthropic stuff * Implement compression for Anythropic Fix lancedb sources * cleanup vectorDBs and check that lance, chroma, and pinecone are returning valid metadata sources * Resolve Weaviate citation sources not working with schema * comment cleanup * disable import on hosted instances (#339) * disable import on hosted instances * Update UI on disabled import/export --------- Co-authored-by: timothycarambat <rambat1010@gmail.com> * Add support for gpt-4-turbo 128K model (#340) resolves #336 Add support for gpt-4-turbo 128K model * 315 show citations based on relevancy score (#316) * settings for similarity score threshold and prisma schema updated * prisma schema migration for adding similarityScore setting * WIP * Min score default change * added similarityThreshold checking for all vectordb providers * linting --------- Co-authored-by: shatfield4 <seanhatfield5@gmail.com> * rename localai to lmstudio * forgot files that were renamed * normalize model interface * add model and context window limits * update LMStudio tagline * Fully working LMStudio integration --------- Co-authored-by: Francisco Bischoff <984592+franzbischoff@users.noreply.github.com> Co-authored-by: Timothy Carambat <rambat1010@gmail.com> Co-authored-by: Sean Hatfield <seanhatfield5@gmail.com>
2023-11-09 21:33:21 +01:00
// 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 = {
LMStudioLLM,
};