mirror of
https://github.com/ParisNeo/lollms-webui.git
synced 2024-12-18 20:17:50 +00:00
1577 lines
52 KiB
JavaScript
1577 lines
52 KiB
JavaScript
// This requires axios
|
|
// In the html don't forget to import axios.min.js cdn
|
|
// <script src="https://cdn.jsdelivr.net/npm/axios/dist/axios.min.js"></script>
|
|
|
|
|
|
|
|
// JavaScript equivalent of the ELF_GENERATION_FORMAT enum
|
|
const ELF_GENERATION_FORMAT = {
|
|
LOLLMS: 0,
|
|
OPENAI: 1,
|
|
OLLAMA: 2,
|
|
LITELLM: 2
|
|
};
|
|
|
|
// JavaScript equivalent of the ELF_COMPLETION_FORMAT enum
|
|
const ELF_COMPLETION_FORMAT = {
|
|
Instruct: 0,
|
|
Chat: 1
|
|
};
|
|
|
|
// Ensuring the objects are immutable
|
|
Object.freeze(ELF_GENERATION_FORMAT);
|
|
Object.freeze(ELF_COMPLETION_FORMAT);
|
|
|
|
|
|
class LollmsClient {
|
|
constructor(
|
|
host_address = null,
|
|
model_name = null,
|
|
ctx_size = null,
|
|
personality = -1,
|
|
n_predict = null,
|
|
temperature = 0.1,
|
|
top_k = 50,
|
|
top_p = 0.95,
|
|
repeat_penalty = 0.8,
|
|
repeat_last_n = 40,
|
|
seed = null,
|
|
n_threads = 8,
|
|
service_key = "",
|
|
default_generation_mode = ELF_GENERATION_FORMAT.LOLLMS
|
|
) {
|
|
// Handle the import or initialization of tiktoken equivalent in JavaScript
|
|
// this.tokenizer = new TikTokenJS('gpt-3.5-turbo-1106'); // This is hypothetical
|
|
|
|
this.host_address = host_address;
|
|
this.model_name = model_name;
|
|
this.ctx_size = ctx_size;
|
|
this.n_predict = n_predict?n_predict:null;
|
|
this.personality = personality;
|
|
this.temperature = temperature;
|
|
this.top_k = top_k;
|
|
this.top_p = top_p;
|
|
this.repeat_penalty = repeat_penalty;
|
|
this.repeat_last_n = repeat_last_n;
|
|
this.seed = seed;
|
|
this.n_threads = n_threads;
|
|
this.service_key = service_key;
|
|
this.default_generation_mode = default_generation_mode;
|
|
this.minNPredict = 10
|
|
this.template = {
|
|
start_header_id_template: "!@>",
|
|
end_header_id_template: ": ",
|
|
separator_template: "\n",
|
|
start_user_header_id_template: "!@>",
|
|
end_user_header_id_template: ": ",
|
|
end_user_message_id_template: "",
|
|
start_ai_header_id_template: "!@>",
|
|
end_ai_header_id_template: ": ",
|
|
end_ai_message_id_template: "",
|
|
system_message_template: "system"
|
|
}
|
|
fetch('/template')
|
|
.then((response) => {
|
|
if (!response.ok) {
|
|
throw new Error('Network response was not ok ' + response.statusText);
|
|
}
|
|
return response.json();
|
|
})
|
|
.then((data) => {
|
|
console.log("data: ", data);
|
|
this.template = data;
|
|
})
|
|
.catch((error) => {
|
|
console.error('Error fetching template:', error);
|
|
});
|
|
|
|
}
|
|
updateSettings(settings) {
|
|
// Update each setting if it's provided in the settings object
|
|
if ('host_address' in settings) this.host_address = settings.host_address;
|
|
if ('model_name' in settings) this.model_name = settings.model_name;
|
|
if ('ctx_size' in settings) this.ctx_size = settings.ctx_size;
|
|
if ('n_predict' in settings) this.n_predict = settings.n_predict;
|
|
if ('personality' in settings) this.personality = settings.personality;
|
|
if ('temperature' in settings) this.temperature = settings.temperature;
|
|
if ('top_k' in settings) this.top_k = settings.top_k;
|
|
if ('top_p' in settings) this.top_p = settings.top_p;
|
|
if ('repeat_penalty' in settings) this.repeat_penalty = settings.repeat_penalty;
|
|
if ('repeat_last_n' in settings) this.repeat_last_n = settings.repeat_last_n;
|
|
if ('seed' in settings) this.seed = settings.seed;
|
|
if ('n_threads' in settings) this.n_threads = settings.n_threads;
|
|
if ('service_key' in settings) this.service_key = settings.service_key;
|
|
if ('default_generation_mode' in settings) this.default_generation_mode = settings.default_generation_mode;
|
|
|
|
// You might want to add some validation or type checking here
|
|
|
|
console.log('Settings updated:', settings);
|
|
}
|
|
|
|
separatorTemplate(){
|
|
return this.template.separator_template;
|
|
}
|
|
system_message(){
|
|
return this.template.start_header_id_template+this.template.system_message_template+this.template.end_header_id_template
|
|
}
|
|
ai_message(ai_name="assistant"){
|
|
return this.template.start_ai_header_id_template+ai_name+this.template.end_ai_header_id_template
|
|
}
|
|
user_message(user_name="user"){
|
|
return this.template.start_user_header_id_template+user_name+this.template.end_user_header_id_template
|
|
}
|
|
custom_message(message_name="message"){
|
|
return this.template.start_ai_header_id_template+message_name+this.template.end_ai_header_id_template
|
|
}
|
|
updateServerAddress(newAddress) {
|
|
this.serverAddress = newAddress;
|
|
}
|
|
async tokenize(prompt) {
|
|
/**
|
|
* Tokenizes the given prompt using the model's tokenizer.
|
|
*
|
|
* @param {string} prompt - The input prompt to be tokenized.
|
|
* @returns {Array} A list of tokens representing the tokenized prompt.
|
|
*/
|
|
console.log("Tokenizing",prompt)
|
|
const output = await axios.post("/lollms_tokenize", {"prompt": prompt});
|
|
return output.data.named_tokens
|
|
}
|
|
async detokenize(tokensList) {
|
|
/**
|
|
* Detokenizes the given list of tokens using the model's tokenizer.
|
|
*
|
|
* @param {Array} tokensList - A list of tokens to be detokenized.
|
|
* @returns {string} The detokenized text as a string.
|
|
*/
|
|
const output = await axios.post("/lollms_detokenize", {"tokens": tokensList});
|
|
console.log(output.data.text)
|
|
return output.data.text
|
|
}
|
|
cancel_generation() {
|
|
// TODO: implement
|
|
}
|
|
|
|
generate(prompt, {
|
|
n_predict = null,
|
|
stream = false,
|
|
temperature = 0.1,
|
|
top_k = 50,
|
|
top_p = 0.95,
|
|
repeat_penalty = 0.8,
|
|
repeat_last_n = 40,
|
|
seed = null,
|
|
n_threads = 8,
|
|
service_key = "",
|
|
streamingCallback = null
|
|
} = {}) {
|
|
switch (this.default_generation_mode) {
|
|
case ELF_GENERATION_FORMAT.LOLLMS:
|
|
return this.lollms_generate(prompt, this.host_address, this.model_name, -1, n_predict, stream, temperature, top_k, top_p, repeat_penalty, repeat_last_n, seed, n_threads, service_key, streamingCallback);
|
|
case ELF_GENERATION_FORMAT.OPENAI:
|
|
return this.openai_generate(prompt, this.host_address, this.model_name, -1, n_predict, stream, temperature, top_k, top_p, repeat_penalty, repeat_last_n, seed, n_threads, ELF_COMPLETION_FORMAT.INSTRUCT, service_key, streamingCallback);
|
|
case ELF_GENERATION_FORMAT.OLLAMA:
|
|
return this.ollama_generate(prompt, this.host_address, this.model_name, -1, n_predict, stream, temperature, top_k, top_p, repeat_penalty, repeat_last_n, seed, n_threads, ELF_COMPLETION_FORMAT.INSTRUCT, service_key, streamingCallback);
|
|
case ELF_GENERATION_FORMAT.LITELLM:
|
|
return this.litellm_generate(prompt, this.host_address, this.model_name, -1, n_predict, stream, temperature, top_k, top_p, repeat_penalty, repeat_last_n, seed, n_threads, ELF_COMPLETION_FORMAT.INSTRUCT, service_key, streamingCallback);
|
|
default:
|
|
throw new Error('Invalid generation mode');
|
|
}
|
|
}
|
|
generate_with_images(prompt, images, {
|
|
n_predict = null,
|
|
stream = false,
|
|
temperature = 0.1,
|
|
top_k = 50,
|
|
top_p = 0.95,
|
|
repeat_penalty = 0.8,
|
|
repeat_last_n = 40,
|
|
seed = null,
|
|
n_threads = 8,
|
|
service_key = "",
|
|
streamingCallback = null
|
|
} = {}) {
|
|
switch (this.default_generation_mode) {
|
|
case ELF_GENERATION_FORMAT.LOLLMS:
|
|
return this.lollms_generate_with_images(prompt, images, this.host_address, this.model_name, -1, n_predict, stream, temperature, top_k, top_p, repeat_penalty, repeat_last_n, seed, n_threads, service_key, streamingCallback);
|
|
case ELF_GENERATION_FORMAT.OPENAI:
|
|
return this.openai_generate_with_images(prompt, images, this.host_address, this.model_name, -1, n_predict, stream, temperature, top_k, top_p, repeat_penalty, repeat_last_n, seed, n_threads, ELF_COMPLETION_FORMAT.INSTRUCT, service_key, streamingCallback);
|
|
default:
|
|
throw new Error('Invalid generation mode');
|
|
}
|
|
}
|
|
async generateText(prompt, options = {}) {
|
|
// Destructure with default values from `this` if not provided in `options`
|
|
const {
|
|
host_address = this.host_address,
|
|
model_name = this.model_name,
|
|
personality = this.personality,
|
|
n_predict = this.n_predict,
|
|
stream = false,
|
|
temperature = this.temperature,
|
|
top_k = this.top_k,
|
|
top_p = this.top_p,
|
|
repeat_penalty = this.repeat_penalty,
|
|
repeat_last_n = this.repeat_last_n,
|
|
seed = this.seed,
|
|
n_threads = this.n_threads,
|
|
service_key = this.service_key,
|
|
streamingCallback = null
|
|
} = options;
|
|
|
|
try {
|
|
const result = await this.lollms_generate(
|
|
prompt,
|
|
host_address,
|
|
model_name,
|
|
personality,
|
|
n_predict,
|
|
stream,
|
|
temperature,
|
|
top_k,
|
|
top_p,
|
|
repeat_penalty,
|
|
repeat_last_n,
|
|
seed,
|
|
n_threads,
|
|
service_key,
|
|
streamingCallback
|
|
);
|
|
return result;
|
|
} catch (error) {
|
|
// Handle any errors that occur during generation
|
|
console.error('An error occurred during text generation:', error);
|
|
throw error; // Re-throw the error if you want to allow the caller to handle it as well
|
|
}
|
|
}
|
|
async lollms_generate(prompt, host_address = this.host_address, model_name = this.model_name, personality = this.personality, n_predict = this.n_predict, stream = false, temperature = this.temperature, top_k = this.top_k, top_p = this.top_p, repeat_penalty = this.repeat_penalty, repeat_last_n = this.repeat_last_n, seed = this.seed, n_threads = this.n_threads, service_key = this.service_key, streamingCallback = null) {
|
|
let url;
|
|
if(host_address!=null){
|
|
url = `${host_address}/lollms_generate`;
|
|
}
|
|
else{
|
|
url = `/lollms_generate`;
|
|
}
|
|
const headers = service_key !== "" ? {
|
|
'Content-Type': 'application/json; charset=utf-8',
|
|
'Authorization': `Bearer ${service_key}`,
|
|
} : {
|
|
'Content-Type': 'application/json',
|
|
};
|
|
console.log("n_predict:",n_predict)
|
|
console.log("self.n_predict:",this.n_predict)
|
|
const data = JSON.stringify({
|
|
prompt: prompt,
|
|
model_name: model_name,
|
|
personality: personality,
|
|
n_predict: n_predict?n_predict:this.n_predict,
|
|
stream: stream,
|
|
temperature: temperature,
|
|
top_k: top_k,
|
|
top_p: top_p,
|
|
repeat_penalty: repeat_penalty,
|
|
repeat_last_n: repeat_last_n,
|
|
seed: seed,
|
|
n_threads: n_threads
|
|
});
|
|
|
|
try {
|
|
const response = await fetch(url, {
|
|
method: 'POST',
|
|
headers: headers,
|
|
body: data
|
|
});
|
|
|
|
// Check if the response is okay
|
|
if (!response.ok) {
|
|
throw new Error('Network response was not ok ' + response.statusText);
|
|
}
|
|
|
|
// Read the response as plaintext
|
|
const responseBody = await response.text();
|
|
console.log(responseBody)
|
|
return responseBody ;
|
|
} catch (error) {
|
|
console.error(error);
|
|
return null;
|
|
}
|
|
}
|
|
async lollms_generate_with_images(prompt, images, host_address = this.host_address, model_name = this.model_name, personality = this.personality, n_predict = this.n_predict, stream = false, temperature = this.temperature, top_k = this.top_k, top_p = this.top_p, repeat_penalty = this.repeat_penalty, repeat_last_n = this.repeat_last_n, seed = this.seed, n_threads = this.n_threads, service_key = this.service_key, streamingCallback = null) {
|
|
let url;
|
|
if(host_address!=null){
|
|
url = `${host_address}/lollms_generate_with_images`;
|
|
}
|
|
else{
|
|
url = `/lollms_generate_with_images`;
|
|
}
|
|
const headers = service_key !== "" ? {
|
|
'Content-Type': 'application/json; charset=utf-8',
|
|
'Authorization': `Bearer ${service_key}`,
|
|
} : {
|
|
'Content-Type': 'application/json',
|
|
};
|
|
console.log("n_predict:",n_predict)
|
|
console.log("self.n_predict:",this.n_predict)
|
|
const data = JSON.stringify({
|
|
prompt: prompt,
|
|
images: images,
|
|
model_name: model_name,
|
|
personality: personality,
|
|
n_predict: n_predict?n_predict:this.n_predict,
|
|
stream: stream,
|
|
temperature: temperature,
|
|
top_k: top_k,
|
|
top_p: top_p,
|
|
repeat_penalty: repeat_penalty,
|
|
repeat_last_n: repeat_last_n,
|
|
seed: seed,
|
|
n_threads: n_threads
|
|
});
|
|
|
|
try {
|
|
const response = await fetch(url, {
|
|
method: 'POST',
|
|
headers: headers,
|
|
body: data
|
|
});
|
|
|
|
// Check if the response is okay
|
|
if (!response.ok) {
|
|
throw new Error('Network response was not ok ' + response.statusText);
|
|
}
|
|
|
|
// Read the response as plaintext
|
|
const responseBody = await response.text();
|
|
console.log(responseBody)
|
|
return responseBody ;
|
|
} catch (error) {
|
|
console.error(error);
|
|
return null;
|
|
}
|
|
}
|
|
|
|
|
|
async openai_generate(prompt, host_address = this.host_address, model_name = this.model_name, personality = this.personality, n_predict = this.n_predict, stream = false, temperature = this.temperature, top_k = this.top_k, top_p = this.top_p, repeat_penalty = this.repeat_penalty, repeat_last_n = this.repeat_last_n, seed = this.seed, n_threads = this.n_threads, ELF_COMPLETION_FORMAT = "vllm instruct", service_key = this.service_key, streamingCallback = null) {
|
|
const url = `${host_address}/generate_completion`;
|
|
const headers = service_key !== "" ? {
|
|
'Content-Type': 'application/json; charset=utf-8',
|
|
'Authorization': `Bearer ${service_key}`,
|
|
} : {
|
|
'Content-Type': 'application/json',
|
|
};
|
|
|
|
const data = JSON.stringify({
|
|
prompt: prompt,
|
|
model_name: model_name,
|
|
personality: personality,
|
|
n_predict: n_predict,
|
|
stream: stream,
|
|
temperature: temperature,
|
|
top_k: top_k,
|
|
top_p: top_p,
|
|
repeat_penalty: repeat_penalty,
|
|
repeat_last_n: repeat_last_n,
|
|
seed: seed,
|
|
n_threads: n_threads,
|
|
completion_format: ELF_COMPLETION_FORMAT
|
|
});
|
|
|
|
try {
|
|
const response = await fetch(url, {
|
|
method: 'POST',
|
|
headers: headers,
|
|
body: data
|
|
});
|
|
if (stream && streamingCallback) {
|
|
// Note: Streaming with Fetch API in the browser might not work as expected because Fetch API does not support HTTP/2 server push.
|
|
// You would need a different approach for real-time streaming.
|
|
streamingCallback(await response.json(), 'MSG_TYPE_CHUNK');
|
|
} else {
|
|
return await response.json();
|
|
}
|
|
} catch (error) {
|
|
console.error("Error generating completion:", error);
|
|
return null;
|
|
}
|
|
}
|
|
|
|
async openai_generate_with_images(prompt, images, options = {}) {
|
|
const {
|
|
host_address = this.host_address,
|
|
model_name = this.model_name,
|
|
personality = this.personality,
|
|
n_predict = this.n_predict,
|
|
stream = false,
|
|
temperature = this.temperature,
|
|
top_k = this.top_k,
|
|
top_p = this.top_p,
|
|
repeat_penalty = this.repeat_penalty,
|
|
repeat_last_n = this.repeat_last_n,
|
|
seed = this.seed,
|
|
n_threads = this.n_threads,
|
|
max_image_width = -1,
|
|
service_key = this.service_key,
|
|
streamingCallback = null,
|
|
} = options;
|
|
|
|
const headers = {
|
|
'Content-Type': 'application/json',
|
|
...(service_key ? { 'Authorization': `Bearer ${service_key}` } : {})
|
|
};
|
|
|
|
const data = {
|
|
model: model_name,
|
|
messages: [
|
|
{
|
|
role: "user",
|
|
content: [
|
|
{
|
|
type: "text",
|
|
text: prompt
|
|
},
|
|
...images.map(image_path => ({
|
|
type: "image_url",
|
|
image_url: {
|
|
url: `data:image/jpeg;base64,${this.encode_image(image_path, max_image_width)}`
|
|
}
|
|
}))
|
|
]
|
|
}
|
|
],
|
|
stream: true,
|
|
temperature: parseFloat(temperature),
|
|
max_tokens: n_predict
|
|
};
|
|
|
|
const completion_format_path = "/v1/chat/completions";
|
|
const url = `${host_address.endsWith("/") ? host_address.slice(0, -1) : host_address}${completion_format_path}`;
|
|
|
|
try {
|
|
const response = await fetch(url, {
|
|
method: 'POST',
|
|
headers: headers,
|
|
body: JSON.stringify(data)
|
|
});
|
|
|
|
if (!response.ok) {
|
|
const content = await response.json();
|
|
if (response.status === 400) {
|
|
this.error(content.error?.message || content.message);
|
|
} else if (response.status === 404) {
|
|
console.error(await response.text());
|
|
}
|
|
return;
|
|
}
|
|
|
|
const reader = response.body.getReader();
|
|
const decoder = new TextDecoder();
|
|
let text = "";
|
|
|
|
while (true) {
|
|
const { done, value } = await reader.read();
|
|
if (done) break;
|
|
|
|
const chunk = decoder.decode(value);
|
|
const lines = chunk.split('\n');
|
|
|
|
for (const line of lines) {
|
|
if (line.startsWith("data: ")) {
|
|
try {
|
|
const jsonData = JSON.parse(line.slice(5).trim());
|
|
const content = jsonData.choices[0]?.delta?.content || "";
|
|
text += content;
|
|
if (streamingCallback) {
|
|
if (!streamingCallback(content, "MSG_TYPE_CHUNK")) {
|
|
return text;
|
|
}
|
|
}
|
|
} catch (error) {
|
|
// Handle JSON parsing error
|
|
}
|
|
} else if (line.startsWith("{")) {
|
|
try {
|
|
const jsonData = JSON.parse(line);
|
|
if (jsonData.object === "error") {
|
|
this.error(jsonData.message);
|
|
return text;
|
|
}
|
|
} catch (error) {
|
|
this.error("Couldn't generate text, verify your key or model name");
|
|
}
|
|
} else {
|
|
text += line;
|
|
if (streamingCallback) {
|
|
if (!streamingCallback(line, "MSG_TYPE_CHUNK")) {
|
|
return text;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
return text;
|
|
} catch (error) {
|
|
console.error("Error in openai_generate_with_images:", error);
|
|
throw error;
|
|
}
|
|
}
|
|
|
|
async encode_image(image_path, max_image_width = -1) {
|
|
// In a browser environment, we'll use the File API and canvas
|
|
// For Node.js, you'd need to use libraries like 'sharp' or 'jimp'
|
|
|
|
return new Promise((resolve, reject) => {
|
|
const img = new Image();
|
|
img.onload = () => {
|
|
let width = img.width;
|
|
let height = img.height;
|
|
|
|
// Resize if necessary
|
|
if (max_image_width !== -1 && width > max_image_width) {
|
|
const ratio = max_image_width / width;
|
|
width = max_image_width;
|
|
height = Math.round(height * ratio);
|
|
}
|
|
|
|
// Create a canvas to draw the image
|
|
const canvas = document.createElement('canvas');
|
|
canvas.width = width;
|
|
canvas.height = height;
|
|
const ctx = canvas.getContext('2d');
|
|
|
|
// Draw the image on the canvas
|
|
ctx.drawImage(img, 0, 0, width, height);
|
|
|
|
// Convert to base64
|
|
const base64Image = canvas.toDataURL('image/jpeg').split(',')[1];
|
|
resolve(base64Image);
|
|
};
|
|
|
|
img.onerror = (error) => reject(error);
|
|
|
|
// Load the image from the provided path
|
|
img.src = image_path;
|
|
});
|
|
}
|
|
async generateCode(prompt, images = [], {
|
|
n_predict = null,
|
|
stream = false,
|
|
temperature = 0.1,
|
|
top_k = 50,
|
|
top_p = 0.95,
|
|
repeat_penalty = 0.8,
|
|
repeat_last_n = 40,
|
|
seed = null,
|
|
n_threads = 8,
|
|
service_key = "",
|
|
streamingCallback = null
|
|
} = {}){
|
|
let response;
|
|
const systemHeader = this.custom_message("Generation infos");
|
|
const codeInstructions = "Generated code must be put inside the adequate markdown code tag. Use this template:\n```language name\nCode\n```\nMake sure only a single code tag is generated at each dialogue turn.\n";
|
|
const fullPrompt = systemHeader + codeInstructions + this.separatorTemplate() + prompt;
|
|
|
|
if (images.length > 0) {
|
|
response = await this.generate_with_images(fullPrompt, images, {
|
|
n_predict: n_predict,
|
|
temperature: temperature,
|
|
top_k: top_k,
|
|
top_p: top_p,
|
|
repeat_penalty: repeat_penalty,
|
|
repeat_last_n: repeat_last_n,
|
|
callback: streamingCallback
|
|
});
|
|
} else {
|
|
response = await this.generate(fullPrompt, {
|
|
n_predict: n_predict,
|
|
temperature: temperature,
|
|
top_k: top_k,
|
|
top_p: top_p,
|
|
repeat_penalty: repeat_penalty,
|
|
repeat_last_n: repeat_last_n,
|
|
callback: streamingCallback
|
|
});
|
|
}
|
|
|
|
const codes = this.extractCodeBlocks(response);
|
|
if (codes.length > 0) {
|
|
let code = '';
|
|
if (!codes[0].is_complete) {
|
|
code = codes[0].content.split('\n').slice(0, -1).join('\n');
|
|
while (!codes[0].is_complete) {
|
|
console.warn("The AI did not finish the code, let's ask it to continue")
|
|
const continuePrompt = prompt + code + this.userFullHeader + "continue the code. Rewrite last line and continue the code. Don't forget to put the code inside a markdown code tag." + this.separatorTemplate() + this.aiFullHeader;
|
|
response = await this.generate(continuePrompt, {
|
|
n_predict: n_predict,
|
|
temperature: temperature,
|
|
top_k: top_k,
|
|
top_p: top_p,
|
|
repeat_penalty: repeat_penalty,
|
|
repeat_last_n: repeat_last_n,
|
|
callback: streamingCallback
|
|
});
|
|
const codes = this.extractCodeBlocks(response);
|
|
if (codes.length === 0) break;
|
|
|
|
if (!codes[0].is_complete) {
|
|
code += '\n' + codes[0].content.split('\n').slice(0, -1).join('\n');
|
|
} else {
|
|
code += '\n' + codes[0].content;
|
|
}
|
|
}
|
|
} else {
|
|
code = codes[0].content;
|
|
}
|
|
return code;
|
|
} else {
|
|
return null;
|
|
}
|
|
}
|
|
async generateCodes(prompt, images = [], {
|
|
n_predict = null,
|
|
stream = false,
|
|
temperature = 0.1,
|
|
top_k = 50,
|
|
top_p = 0.95,
|
|
repeat_penalty = 0.8,
|
|
repeat_last_n = 40,
|
|
seed = null,
|
|
n_threads = 8,
|
|
service_key = "",
|
|
streamingCallback = null
|
|
} = {}) {
|
|
let response;
|
|
const systemHeader = this.custom_message("Generation infos");
|
|
const codeInstructions = "Generated code must be put inside the adequate markdown code tag. Use this template:\n```language name\nCode\n```\n";
|
|
const fullPrompt = systemHeader + codeInstructions + this.separatorTemplate() + prompt;
|
|
|
|
if (images.length > 0) {
|
|
response = await this.generate_with_images(fullPrompt, images, {
|
|
n_predict,
|
|
temperature,
|
|
top_k,
|
|
top_p,
|
|
repeat_penalty,
|
|
repeat_last_n,
|
|
callback: streamingCallback
|
|
});
|
|
} else {
|
|
response = await this.generate(fullPrompt, {
|
|
n_predict,
|
|
temperature,
|
|
top_k,
|
|
top_p,
|
|
repeat_penalty,
|
|
repeat_last_n,
|
|
callback: streamingCallback
|
|
});
|
|
}
|
|
|
|
let codes = this.extractCodeBlocks(response);
|
|
let completeCodes = [];
|
|
|
|
while (codes.length > 0) {
|
|
let currentCode = codes.shift();
|
|
let codeContent = currentCode.content;
|
|
|
|
while (!currentCode.is_complete) {
|
|
console.warn("The AI did not finish the code, let's ask it to continue");
|
|
const continuePrompt = prompt + codeContent + this.userFullHeader + "continue the code. Rewrite last line and continue the code." + this.separatorTemplate() + this.aiFullHeader;
|
|
|
|
response = await this.generate(continuePrompt, {
|
|
n_predict,
|
|
temperature,
|
|
top_k,
|
|
top_p,
|
|
repeat_penalty,
|
|
repeat_last_n,
|
|
callback: streamingCallback
|
|
});
|
|
|
|
const newCodes = this.extractCodeBlocks(response);
|
|
if (newCodes.length === 0) break;
|
|
|
|
// Append the content of the first new code block
|
|
codeContent += '\n' + newCodes[0].content;
|
|
currentCode = newCodes[0];
|
|
|
|
// If there are more code blocks, add them to the codes array
|
|
if (newCodes.length > 1) {
|
|
codes = [...newCodes.slice(1), ...codes];
|
|
}
|
|
}
|
|
|
|
completeCodes.push({
|
|
language: currentCode.language,
|
|
content: codeContent
|
|
});
|
|
}
|
|
|
|
return completeCodes;
|
|
}
|
|
|
|
extractCodeBlocks(text) {
|
|
const codeBlocks = [];
|
|
let remaining = text;
|
|
let blocIndex = 0;
|
|
let firstIndex = 0;
|
|
const indices = [];
|
|
|
|
// Find all code block delimiters
|
|
while (remaining.length > 0) {
|
|
const index = remaining.indexOf("```");
|
|
if (index === -1) {
|
|
if (blocIndex % 2 === 1) {
|
|
indices.push(remaining.length + firstIndex);
|
|
}
|
|
break;
|
|
}
|
|
indices.push(index + firstIndex);
|
|
remaining = remaining.slice(index + 3);
|
|
firstIndex += index + 3;
|
|
blocIndex++;
|
|
}
|
|
|
|
let isStart = true;
|
|
for (let i = 0; i < indices.length; i++) {
|
|
if (isStart) {
|
|
const blockInfo = {
|
|
index: i,
|
|
file_name: "",
|
|
section: "",
|
|
content: "",
|
|
type: "",
|
|
is_complete: false
|
|
};
|
|
|
|
// Check for file name in preceding line
|
|
const precedingText = text.slice(0, indices[i]).trim().split('\n');
|
|
if (precedingText.length > 0) {
|
|
const lastLine = precedingText[precedingText.length - 1].trim();
|
|
if (lastLine.startsWith("<file_name>") && lastLine.endsWith("</file_name>")) {
|
|
blockInfo.file_name = lastLine.slice("<file_name>".length, -"</file_name>".length).trim();
|
|
} else if (lastLine.startsWith("## filename:")) {
|
|
blockInfo.file_name = lastLine.slice("## filename:".length).trim();
|
|
}
|
|
if (lastLine.startsWith("<section>") && lastLine.endsWith("</section>")) {
|
|
blockInfo.section = lastLine.slice("<section>".length, -"</section>".length).trim();
|
|
}
|
|
}
|
|
|
|
const subText = text.slice(indices[i] + 3);
|
|
if (subText.length > 0) {
|
|
const findSpace = subText.indexOf(" ");
|
|
const findReturn = subText.indexOf("\n");
|
|
let nextIndex = Math.min(findSpace === -1 ? Infinity : findSpace, findReturn === -1 ? Infinity : findReturn);
|
|
if (subText.slice(0, nextIndex).includes('{')) {
|
|
nextIndex = 0;
|
|
}
|
|
const startPos = nextIndex;
|
|
|
|
if (text[indices[i] + 3] === "\n" || text[indices[i] + 3] === " " || text[indices[i] + 3] === "\t") {
|
|
blockInfo.type = 'language-specific';
|
|
} else {
|
|
blockInfo.type = subText.slice(0, nextIndex);
|
|
}
|
|
|
|
if (i + 1 < indices.length) {
|
|
const nextPos = indices[i + 1] - indices[i];
|
|
if (nextPos - 3 < subText.length && subText[nextPos - 3] === "`") {
|
|
blockInfo.content = subText.slice(startPos, nextPos - 3).trim();
|
|
blockInfo.is_complete = true;
|
|
} else {
|
|
blockInfo.content = subText.slice(startPos, nextPos).trim();
|
|
blockInfo.is_complete = false;
|
|
}
|
|
} else {
|
|
blockInfo.content = subText.slice(startPos).trim();
|
|
blockInfo.is_complete = false;
|
|
}
|
|
|
|
codeBlocks.push(blockInfo);
|
|
}
|
|
isStart = false;
|
|
} else {
|
|
isStart = true;
|
|
}
|
|
}
|
|
|
|
return codeBlocks;
|
|
}
|
|
|
|
|
|
|
|
async listMountedPersonalities(host_address = this.host_address) {
|
|
const url = `${host_address}/list_mounted_personalities`;
|
|
|
|
try {
|
|
const response = await fetch(url);
|
|
return await response.json();
|
|
} catch (error) {
|
|
console.error(error);
|
|
return null;
|
|
}
|
|
}
|
|
|
|
async listModels(host_address = this.host_address) {
|
|
const url = `${host_address}/list_models`;
|
|
|
|
try {
|
|
const response = await fetch(url);
|
|
return await response.json();
|
|
} catch (error) {
|
|
console.error(error);
|
|
return null;
|
|
}
|
|
}
|
|
}
|
|
|
|
class TextChunker {
|
|
constructor(chunkSize = 512, overlap = 0, tokenizer = null, model = null) {
|
|
this.chunkSize = chunkSize;
|
|
this.overlap = overlap;
|
|
this.tokenizer = tokenizer || new TikTokenTokenizer();
|
|
this.model = model;
|
|
}
|
|
|
|
async getTextChunks(text, doc, cleanChunk = true, minNbTokensInChunk = 10) {
|
|
const paragraphs = text.split('\n\n');
|
|
const chunks = [];
|
|
let currentChunk = [];
|
|
let currentTokens = 0;
|
|
let chunkId = 0;
|
|
|
|
for (const paragraph of paragraphs) {
|
|
const cleanedParagraph = cleanChunk ? paragraph.trim() : paragraph;
|
|
const paragraphTokens = (await this.tokenizer.tokenize(cleanedParagraph)).length;
|
|
|
|
if (currentTokens + paragraphTokens > this.chunkSize) {
|
|
if (currentTokens > minNbTokensInChunk) {
|
|
let chunkText = currentChunk.join('\n\n');
|
|
if (cleanChunk) {
|
|
chunkText = TextChunker.removeUnnecessaryReturns(chunkText);
|
|
}
|
|
const chunk = new Chunk(doc, '', chunkText, currentTokens, chunkId);
|
|
chunkId++;
|
|
chunks.push(chunk);
|
|
}
|
|
|
|
if (this.overlap > 0) {
|
|
currentChunk = [...currentChunk.slice(-this.overlap), cleanedParagraph];
|
|
} else {
|
|
currentChunk = [cleanedParagraph];
|
|
}
|
|
currentTokens = currentChunk.reduce(async (sum, p) => sum + await this.tokenizer.tokenize(p).length, 0);
|
|
} else {
|
|
currentChunk.push(cleanedParagraph);
|
|
currentTokens += paragraphTokens;
|
|
}
|
|
}
|
|
|
|
if (currentChunk.length > 0 && currentTokens > minNbTokensInChunk) {
|
|
let chunkText = currentChunk.join('\n\n');
|
|
if (cleanChunk) {
|
|
chunkText = TextChunker.removeUnnecessaryReturns(chunkText);
|
|
}
|
|
const chunk = new Chunk(doc, '', chunkText, currentTokens, chunkId);
|
|
chunks.push(chunk);
|
|
}
|
|
|
|
return chunks;
|
|
}
|
|
|
|
static removeUnnecessaryReturns(paragraph) {
|
|
const lines = paragraph.split('\n');
|
|
return lines.filter(line => line.trim()).join('\n');
|
|
}
|
|
|
|
static async chunkText(text, tokenizer, chunkSize = 512, overlap = 0, cleanChunk = true, minNbTokensInChunk = 10) {
|
|
const paragraphs = text.split('\n\n');
|
|
const chunks = [];
|
|
let currentChunk = [];
|
|
let currentTokens = 0;
|
|
for (const paragraph of paragraphs) {
|
|
const cleanedParagraph = cleanChunk ? paragraph.trim() : paragraph;
|
|
const paragraphTokens = (await tokenizer.tokenize(cleanedParagraph)).length;
|
|
|
|
if (currentTokens + paragraphTokens > chunkSize) {
|
|
if (currentTokens > minNbTokensInChunk) {
|
|
let chunkText = currentChunk.join('\n\n');
|
|
if (cleanChunk) {
|
|
chunkText = TextChunker.removeUnnecessaryReturns(chunkText);
|
|
}
|
|
chunks.push(chunkText);
|
|
}
|
|
|
|
if (overlap > 0) {
|
|
currentChunk = [...currentChunk.slice(-overlap), cleanedParagraph];
|
|
} else {
|
|
currentChunk = [cleanedParagraph];
|
|
}
|
|
currentTokens = currentChunk.reduce(async (sum, p) => sum + await tokenizer.tokenize(p).length, 0);
|
|
} else {
|
|
currentChunk.push(cleanedParagraph);
|
|
currentTokens += paragraphTokens;
|
|
}
|
|
}
|
|
if (currentChunk.length > 0 && currentTokens > minNbTokensInChunk) {
|
|
let chunkText = currentChunk.join('\n\n');
|
|
if (cleanChunk) {
|
|
chunkText = TextChunker.removeUnnecessaryReturns(chunkText);
|
|
}
|
|
chunks.push(chunkText);
|
|
}
|
|
|
|
return chunks;
|
|
}
|
|
}
|
|
|
|
class TasksLibrary {
|
|
constructor(lollms) {
|
|
this.lollms = lollms;
|
|
}
|
|
|
|
async translateTextChunk(textChunk, outputLanguage = "french", host_address = null, model_name = null, temperature = 0.1, maxGenerationSize = 3000) {
|
|
const translationPrompt = [
|
|
`${this.lollms.system_message()}`,
|
|
`Translate the following text to ${outputLanguage}.`,
|
|
`Be faithful to the original text and do not add or remove any information.`,
|
|
`Respond only with the translated text.`,
|
|
`Do not add comments or explanations.`,
|
|
`${this.lollms.user_message("text to translate")}`,
|
|
`${textChunk}`,
|
|
`${this.lollms.ai_message("translation")}`,
|
|
].join("\n");
|
|
|
|
const translated = await this.lollms.generateText(
|
|
translationPrompt,
|
|
host_address,
|
|
model_name,
|
|
-1, // personality
|
|
maxGenerationSize, // n_predict
|
|
false, // stream
|
|
temperature, // temperature
|
|
undefined, // top_k, using undefined to fallback on LollmsClient's default
|
|
undefined, // top_p, using undefined to fallback on LollmsClient's default
|
|
undefined, // repeat_penalty, using undefined to fallback on LollmsClient's default
|
|
undefined, // repeat_last_n, using undefined to fallback on LollmsClient's default
|
|
undefined, // seed, using undefined to fallback on LollmsClient's default
|
|
undefined, // n_threads, using undefined to fallback on LollmsClient's default
|
|
undefined // service_key, using undefined to fallback on LollmsClient's default
|
|
);
|
|
|
|
return translated;
|
|
}
|
|
|
|
async tokenize(text) {
|
|
// Assuming the LollmsClient has a method to tokenize text
|
|
return await this.lollms.tokenize(text);
|
|
}
|
|
|
|
async summarizeText(
|
|
text,
|
|
summaryInstruction = "summarize",
|
|
docName = "chunk",
|
|
answerStart = "",
|
|
maxGenerationSize = 3000,
|
|
maxSummarySize = 512,
|
|
callback = null,
|
|
chunkSummaryPostProcessing = null,
|
|
summaryMode = "SEQUENTIAL",
|
|
reformat=false
|
|
) {
|
|
console.log("Tokenizing:")
|
|
console.log(text)
|
|
|
|
let tk = await this.tokenize(text);
|
|
let prevLen = tk.length;
|
|
let documentChunks = null;
|
|
console.log(`Text size: ${prevLen}`)
|
|
|
|
while (tk.length > maxSummarySize && (documentChunks === null || documentChunks.length > 1)) {
|
|
this.stepStart(`Compressing ${docName}...`);
|
|
let chunkSize = Math.floor(this.lollms.ctxSize * 0.6);
|
|
documentChunks = await TextChunker.chunkText(text, this.lollms, chunkSize, 0, true);
|
|
console.log(`documentChunks: ${documentChunks}`)
|
|
text = await this.summarizeChunks({
|
|
chunks: documentChunks,
|
|
summaryInstruction,
|
|
docName,
|
|
answerStart,
|
|
maxGenerationSize,
|
|
callback,
|
|
chunkSummaryPostProcessing,
|
|
summaryMode
|
|
});
|
|
tk = await this.tokenize(text);
|
|
let dtkLn = prevLen - tk.length;
|
|
prevLen = tk.length;
|
|
this.step(`Current text size: ${prevLen}, max summary size: ${maxSummarySize}`);
|
|
this.stepEnd(`Compressing ${docName}...`);
|
|
if (dtkLn <= 10) break; // it is not summarizing
|
|
}
|
|
if(reformat){
|
|
text = await this.lollms.generate(
|
|
[
|
|
this.lollms.system_message(),
|
|
`${text}`,
|
|
this.lollms.system_message(),
|
|
summaryInstruction,
|
|
"Do not add any extra comments.",
|
|
this.lollms.system_message() + answerStart
|
|
].join("\n"),
|
|
);
|
|
}
|
|
return text;
|
|
}
|
|
|
|
async smartDataExtraction({
|
|
text,
|
|
dataExtractionInstruction = "summarize the current chunk.",
|
|
finalTaskInstruction = "reformulate with better wording",
|
|
docName = "chunk",
|
|
answerStart = "",
|
|
maxGenerationSize = 3000,
|
|
maxSummarySize = 512,
|
|
callback = null,
|
|
chunkSummaryPostProcessing = null,
|
|
summaryMode = "SEQUENTIAL"
|
|
}) {
|
|
let tk = await this.tokenize(text);
|
|
let prevLen = tk.length;
|
|
|
|
while (tk.length > maxSummarySize) {
|
|
let chunkSize = Math.floor(this.lollms.ctxSize * 0.6);
|
|
let documentChunks = await TextChunker.chunkText(text, this.lollms, chunkSize, 0, true);
|
|
text = await this.summarizeChunks({
|
|
chunks: documentChunks,
|
|
summaryInstruction: dataExtractionInstruction,
|
|
docName,
|
|
answerStart,
|
|
maxGenerationSize,
|
|
callback,
|
|
chunkSummaryPostProcessing,
|
|
summaryMode
|
|
});
|
|
tk = await this.tokenize(text);
|
|
let dtkLn = prevLen - tk.length;
|
|
prevLen = tk.length;
|
|
this.step(`Current text size: ${prevLen}, max summary size: ${maxSummarySize}`);
|
|
if (dtkLn <= 10) break; // it is not summarizing
|
|
}
|
|
|
|
this.stepStart("Rewriting ...");
|
|
text = await this.summarizeChunks({
|
|
chunks: [text],
|
|
summaryInstruction: finalTaskInstruction,
|
|
docName,
|
|
answerStart,
|
|
maxGenerationSize,
|
|
callback,
|
|
chunkSummaryPostProcessing
|
|
});
|
|
this.stepEnd("Rewriting ...");
|
|
|
|
return text;
|
|
}
|
|
|
|
async summarizeChunks({
|
|
chunks,
|
|
summaryInstruction = "summarize the current chunk.",
|
|
docName = "chunk",
|
|
answerStart = "",
|
|
maxGenerationSize = 3000,
|
|
callback = null,
|
|
chunkSummaryPostProcessing = null,
|
|
summaryMode = "SEQUENTIAL"
|
|
}) {
|
|
if (summaryMode === "SEQUENTIAL") {
|
|
let summary = "";
|
|
for (let i = 0; i < chunks.length; i++) {
|
|
this.stepStart(`Summary of ${docName} - Processing chunk: ${i + 1}/${chunks.length}`);
|
|
summary = `${answerStart}` + await this.fastGen(
|
|
[
|
|
this.lollms.system_message(),
|
|
`${summary}`,
|
|
this.lollms.system_message(),
|
|
`${chunks[i]}`,
|
|
this.lollms.system_message(),
|
|
summaryInstruction,
|
|
"Keep only information relevant to the context",
|
|
"The output must keep information from the previous chunk analysis and add the current chunk extracted information.",
|
|
"Be precise and do not invent information that does not exist in the previous chunks analysis or the current chunk.",
|
|
"Do not add any extra comments.",
|
|
this.lollms.system_message() + answerStart
|
|
].join("\n"),
|
|
maxGenerationSize,
|
|
callback
|
|
);
|
|
if (chunkSummaryPostProcessing) {
|
|
summary = chunkSummaryPostProcessing(summary);
|
|
}
|
|
this.stepEnd(`Summary of ${docName} - Processing chunk: ${i + 1}/${chunks.length}`);
|
|
}
|
|
return summary;
|
|
} else {
|
|
let summaries = [];
|
|
for (let i = 0; i < chunks.length; i++) {
|
|
this.stepStart(`Summary of ${docName} - Processing chunk: ${i + 1}/${chunks.length}`);
|
|
let summary = `${answerStart}` + await this.fastGen(
|
|
[
|
|
`${this.lollms.system_message()}Document_chunk [${docName}]${this.lollms.system_message()}`,
|
|
`${chunks[i]}`,
|
|
`${this.lollms.system_message()}${summaryInstruction}`,
|
|
"Answer directly with the summary with no extra comments.",
|
|
`${this.lollms.system_message()}summary${this.lollms.system_message()}`,
|
|
`${answerStart}`
|
|
].join("\n"),
|
|
maxGenerationSize,
|
|
callback
|
|
);
|
|
if (chunkSummaryPostProcessing) {
|
|
summary = chunkSummaryPostProcessing(summary);
|
|
}
|
|
summaries.push(summary);
|
|
this.stepEnd(`Summary of ${docName} - Processing chunk: ${i + 1}/${chunks.length}`);
|
|
}
|
|
return summaries.join("\n");
|
|
}
|
|
}
|
|
|
|
async sequentialChunksSummary({
|
|
chunks,
|
|
summaryInstruction = "summarize",
|
|
docName = "chunk",
|
|
answerStart = "",
|
|
maxGenerationSize = 3000,
|
|
callback = null,
|
|
chunkSummaryPostProcessing = null
|
|
}) {
|
|
let summaries = [];
|
|
|
|
for (let i = 0; i < chunks.length; i++) {
|
|
let chunk1 = i < chunks.length - 1 ? chunks[i + 1] : "";
|
|
let chunk = i > 0 ? summaries[summaries.length - 1] : chunks[i];
|
|
|
|
this.stepStart(`Summary of ${docName} - Processing chunk: ${i + 1}/${chunks.length}`);
|
|
let summary = `${answerStart}` + await this.fastGen(
|
|
[
|
|
`${this.lollms.system_message()}Document_chunk: ${docName}${this.lollms.system_message()}`,
|
|
"Block1:",
|
|
`${chunk}`,
|
|
"Block2:",
|
|
`${chunk1}`,
|
|
`${this.lollms.system_message()}${summaryInstruction}`,
|
|
"Answer directly with the summary with no extra comments.",
|
|
`${this.lollms.system_message()}summary${this.lollms.system_message()}`,
|
|
`${answerStart}`
|
|
].join("\n"),
|
|
maxGenerationSize,
|
|
callback
|
|
);
|
|
if (chunkSummaryPostProcessing) {
|
|
summary = chunkSummaryPostProcessing(summary);
|
|
}
|
|
summaries.push(summary);
|
|
this.stepEnd(`Summary of ${docName} - Processing chunk: ${i + 1}/${chunks.length}`);
|
|
}
|
|
return summaries.join("\n");
|
|
}
|
|
|
|
// Placeholder methods for step stepStart, stepEnd, fastGen
|
|
step(message) {
|
|
console.log(message);
|
|
}
|
|
|
|
stepStart(message) {
|
|
console.log(message);
|
|
}
|
|
|
|
stepEnd(message) {
|
|
console.log(message);
|
|
}
|
|
|
|
async fastGen(prompt, maxGenerationSize, callback) {
|
|
// Use the LollmsClient to generate text
|
|
const response = await this.lollms.generateText(prompt);
|
|
if (callback) callback(response);
|
|
return response;
|
|
}
|
|
|
|
async yesNo(question, context = "", maxAnswerLength = 50, conditioning = "") {
|
|
/**
|
|
* Analyzes the user prompt and answers whether it is asking to generate an image.
|
|
*
|
|
* @param {string} question - The user's message.
|
|
* @param {string} context - The context for the question.
|
|
* @param {number} maxAnswerLength - The maximum length of the generated answer.
|
|
* @param {string} conditioning - An optional system message to put at the beginning of the prompt.
|
|
* @returns {boolean} True if the user prompt is asking to generate an image, False otherwise.
|
|
*/
|
|
return this.multichoiceQuestion(question, ["no", "yes"], context, maxAnswerLength, conditioning) > 0;
|
|
}
|
|
|
|
printPrompt(prompt){
|
|
console.log(prompt);
|
|
|
|
}
|
|
|
|
async multichoiceQuestion(question, possibleAnswers, context = "", maxAnswerLength = 50, conditioning = "") {
|
|
/**
|
|
* Interprets a multi-choice question from a user's response. This function expects only one choice as true.
|
|
* All other choices are considered false. If none are correct, returns -1.
|
|
*
|
|
* @param {string} question - The multi-choice question posed by the user.
|
|
* @param {Array} possibleAnswers - A list containing all valid options for the chosen value.
|
|
* @param {string} context - The context for the question.
|
|
* @param {number} maxAnswerLength - Maximum string length allowed while interpreting the users' responses.
|
|
* @param {string} conditioning - An optional system message to put at the beginning of the prompt.
|
|
* @returns {number} Index of the selected option within the possibleAnswers list. Or -1 if there was no match found among any of them.
|
|
*/
|
|
|
|
const choices = possibleAnswers.map((answer, index) => `${index}. ${answer}`).join("\n");
|
|
const elements = conditioning ? [conditioning] : [];
|
|
elements.push(
|
|
this.lollms.system_message(),
|
|
"Answer this multi choices question.",
|
|
"Answer with an id from the possible answers.",
|
|
"Do not answer with an id outside this possible answers."
|
|
);
|
|
|
|
if (context) {
|
|
elements.push(
|
|
this.lollms.custom_message("context"),
|
|
context
|
|
);
|
|
}
|
|
|
|
elements.push(
|
|
this.lollms.custom_message(`question`)+`$${question}`,
|
|
this.lollms.custom_message(`possible answers`),
|
|
choices,
|
|
this.lollms.custom_message("answer")
|
|
);
|
|
|
|
const prompt = this.buildPrompt(elements);
|
|
let gen = await this.lollms.generate(prompt, {
|
|
n_predict: maxAnswerLength,
|
|
temperature: 0.1,
|
|
top_k: 50,
|
|
top_p: 0.9,
|
|
repeat_penalty: 1.0,
|
|
repeat_last_n: 50,
|
|
callback: this.sink
|
|
})
|
|
gen = gen.trim().replace("", "").replace("", "");
|
|
|
|
const selection = gen.trim().split(" ")[0].replace(",", "").replace(".", "");
|
|
this.printPrompt("Multi choice selection", prompt + gen);
|
|
|
|
try {
|
|
return parseInt(selection, 10);
|
|
} catch (error) {
|
|
console.log("Model failed to answer the question");
|
|
return -1;
|
|
}
|
|
}
|
|
|
|
buildPrompt(promptParts, sacrificeId = -1, contextSize = null, minimumSpareContextSize = null) {
|
|
/**
|
|
* Builds the prompt for code generation.
|
|
*
|
|
* @param {Array<string>} promptParts - A list of strings representing the parts of the prompt.
|
|
* @param {number} sacrificeId - The ID of the part to sacrifice.
|
|
* @param {number} contextSize - The size of the context.
|
|
* @param {number} minimumSpareContextSize - The minimum spare context size.
|
|
* @returns {string} - The built prompt.
|
|
*/
|
|
if (contextSize === null) {
|
|
contextSize = this.lollms.ctxSize;
|
|
}
|
|
if (minimumSpareContextSize === null) {
|
|
minimumSpareContextSize = this.lollms.minNPredict;
|
|
}
|
|
|
|
if (sacrificeId === -1 || promptParts[sacrificeId].length < 50) {
|
|
return promptParts.filter(s => s !== "").join("\n");
|
|
} else {
|
|
const partTokens = [];
|
|
let nbTokens = 0;
|
|
|
|
for (let i = 0; i < promptParts.length; i++) {
|
|
const part = promptParts[i];
|
|
const tk = this.lollms.tokenize(part);
|
|
partTokens.push(tk);
|
|
if (i !== sacrificeId) {
|
|
nbTokens += tk.length;
|
|
}
|
|
}
|
|
|
|
let sacrificeText = "";
|
|
if (partTokens[sacrificeId].length > 0) {
|
|
const sacrificeTk = partTokens[sacrificeId];
|
|
const tokensToKeep = sacrificeTk.slice(-(contextSize - nbTokens - minimumSpareContextSize));
|
|
sacrificeText = this.lollms.detokenize(tokensToKeep);
|
|
}
|
|
|
|
promptParts[sacrificeId] = sacrificeText;
|
|
return promptParts.filter(s => s !== "").join("\n");
|
|
}
|
|
}
|
|
|
|
extractCodeBlocks(text) {
|
|
const codeBlocks = [];
|
|
let remaining = text;
|
|
let blocIndex = 0;
|
|
let firstIndex = 0;
|
|
const indices = [];
|
|
|
|
// Find all code block delimiters
|
|
while (remaining.length > 0) {
|
|
const index = remaining.indexOf("```");
|
|
if (index === -1) {
|
|
if (blocIndex % 2 === 1) {
|
|
indices.push(remaining.length + firstIndex);
|
|
}
|
|
break;
|
|
}
|
|
indices.push(index + firstIndex);
|
|
remaining = remaining.slice(index + 3);
|
|
firstIndex += index + 3;
|
|
blocIndex++;
|
|
}
|
|
|
|
let isStart = true;
|
|
for (let i = 0; i < indices.length; i++) {
|
|
if (isStart) {
|
|
const blockInfo = {
|
|
index: i,
|
|
file_name: "",
|
|
section: "",
|
|
content: "",
|
|
type: "",
|
|
is_complete: false
|
|
};
|
|
|
|
// Check for file name in preceding line
|
|
const precedingText = text.slice(0, indices[i]).trim().split('\n');
|
|
if (precedingText.length > 0) {
|
|
const lastLine = precedingText[precedingText.length - 1].trim();
|
|
if (lastLine.startsWith("<file_name>") && lastLine.endsWith("</file_name>")) {
|
|
blockInfo.file_name = lastLine.slice("<file_name>".length, -"</file_name>".length).trim();
|
|
} else if (lastLine.startsWith("## filename:")) {
|
|
blockInfo.file_name = lastLine.slice("## filename:".length).trim();
|
|
}
|
|
if (lastLine.startsWith("<section>") && lastLine.endsWith("</section>")) {
|
|
blockInfo.section = lastLine.slice("<section>".length, -"</section>".length).trim();
|
|
}
|
|
}
|
|
|
|
const subText = text.slice(indices[i] + 3);
|
|
if (subText.length > 0) {
|
|
const findSpace = subText.indexOf(" ");
|
|
const findReturn = subText.indexOf("\n");
|
|
let nextIndex = Math.min(findSpace === -1 ? Infinity : findSpace, findReturn === -1 ? Infinity : findReturn);
|
|
if (subText.slice(0, nextIndex).includes('{')) {
|
|
nextIndex = 0;
|
|
}
|
|
const startPos = nextIndex;
|
|
|
|
if (text[indices[i] + 3] === "\n" || text[indices[i] + 3] === " " || text[indices[i] + 3] === "\t") {
|
|
blockInfo.type = 'language-specific';
|
|
} else {
|
|
blockInfo.type = subText.slice(0, nextIndex);
|
|
}
|
|
|
|
if (i + 1 < indices.length) {
|
|
const nextPos = indices[i + 1] - indices[i];
|
|
if (nextPos - 3 < subText.length && subText[nextPos - 3] === "`") {
|
|
blockInfo.content = subText.slice(startPos, nextPos - 3).trim();
|
|
blockInfo.is_complete = true;
|
|
} else {
|
|
blockInfo.content = subText.slice(startPos, nextPos).trim();
|
|
blockInfo.is_complete = false;
|
|
}
|
|
} else {
|
|
blockInfo.content = subText.slice(startPos).trim();
|
|
blockInfo.is_complete = false;
|
|
}
|
|
|
|
codeBlocks.push(blockInfo);
|
|
}
|
|
isStart = false;
|
|
} else {
|
|
isStart = true;
|
|
}
|
|
}
|
|
|
|
return codeBlocks;
|
|
}
|
|
|
|
|
|
/**
|
|
* Updates the given code based on the provided query string.
|
|
* The query string can contain two types of modifications:
|
|
* 1. FULL_REWRITE: Completely replaces the original code with the new code.
|
|
* 2. REPLACE: Replaces specific code snippets within the original code.
|
|
*
|
|
* @param {string} originalCode - The original code to be updated.
|
|
* @param {string} queryString - The string containing the update instructions.
|
|
* @returns {object} - An object with the following properties:
|
|
* - updatedCode: The updated code.
|
|
* - modifications: An array of objects representing the changes made, each with properties 'oldCode' and 'newCode'.
|
|
* - hasQuery: A boolean indicating whether the queryString contained any valid queries.
|
|
*/
|
|
updateCode(originalCode, queryString) {
|
|
const queries = queryString.split('# REPLACE\n');
|
|
let updatedCode = originalCode;
|
|
const modifications = [];
|
|
|
|
// Check if there's a FULL_REWRITE first
|
|
const fullRewriteStart = queryString.indexOf('# FULL_REWRITE');
|
|
if (fullRewriteStart !== -1) {
|
|
const newCode = queryString.slice(fullRewriteStart + 14).trim();
|
|
updatedCode = newCode;
|
|
modifications.push({
|
|
oldCode: originalCode,
|
|
newCode
|
|
});
|
|
return {
|
|
updatedCode,
|
|
modifications,
|
|
hasQuery: true
|
|
};
|
|
}
|
|
|
|
if (queries.length === 1 && queries[0].trim() === '') {
|
|
console.log("No queries detected");
|
|
return {
|
|
updatedCode,
|
|
modifications: [],
|
|
hasQuery: false
|
|
};
|
|
}
|
|
|
|
for (const query of queries) {
|
|
if (query.trim() === '') continue;
|
|
|
|
const originalCodeStart = query.indexOf('# ORIGINAL\n') + 11;
|
|
const originalCodeEnd = query.indexOf('\n# SET\n');
|
|
let oldCode = query.slice(originalCodeStart, originalCodeEnd);
|
|
|
|
const newCodeStart = query.indexOf('# SET\n') + 6;
|
|
const newCode = query.slice(newCodeStart);
|
|
|
|
const modification = {
|
|
oldCode: oldCode.trim(),
|
|
newCode: newCode.trim()
|
|
};
|
|
if(oldCode =="<old_code>"){
|
|
oldCode = originalCode
|
|
}
|
|
console.log("oldCode:")
|
|
console.log(oldCode)
|
|
console.log("newCode:")
|
|
console.log(newCode)
|
|
console.log("Before update", updatedCode);
|
|
if(oldCode===updatedCode){
|
|
console.log("Changing the whole content")
|
|
updatedCode = newCode
|
|
}
|
|
else{
|
|
updatedCode = updatedCode.replace(oldCode, newCode.trim());
|
|
}
|
|
console.log("After update", updatedCode);
|
|
modifications.push(modification);
|
|
|
|
}
|
|
|
|
return {
|
|
updatedCode,
|
|
modifications,
|
|
hasQuery: true
|
|
};
|
|
}
|
|
}
|
|
class LOLLMSRAGClient {
|
|
constructor(lc) {
|
|
this.lc = lc;
|
|
this.key = lc.service_key || this.generateRandomKey();
|
|
console.log("Connecting to server with key:", this.key);
|
|
}
|
|
|
|
generateRandomKey() {
|
|
// Generate a random key (UUID v4)
|
|
return 'xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx'.replace(/[xy]/g, function(c) {
|
|
const r = Math.random() * 16 | 0, v = c === 'x' ? r : (r & 0x3 | 0x8);
|
|
return v.toString(16);
|
|
});
|
|
}
|
|
|
|
async request(endpoint, method = 'GET', body = null) {
|
|
try {
|
|
const options = {
|
|
method,
|
|
headers: {
|
|
'Content-Type': 'application/json',
|
|
},
|
|
body: body ? JSON.stringify(body) : null,
|
|
};
|
|
|
|
const response = await fetch(`${this.lc.host_address}${endpoint}`, options);
|
|
const data = await response.json();
|
|
|
|
if (!response.ok) {
|
|
throw new Error(`Error: ${data.detail || response.statusText}`);
|
|
}
|
|
|
|
return data;
|
|
} catch (error) {
|
|
console.error("Request failed:", error);
|
|
throw new Error(`Error: ${error.message}`);
|
|
}
|
|
}
|
|
|
|
async addDocument(title, content, path = "unknown") {
|
|
const document = { title, content, path, key: this.key };
|
|
return this.request('/add_document', 'POST', document);
|
|
}
|
|
|
|
async removeDocument(documentId) {
|
|
const body = { key: this.key };
|
|
return this.request(`/remove_document/${documentId}`, 'POST', body);
|
|
}
|
|
|
|
async indexDatabase() {
|
|
const body = { key: this.key };
|
|
console.log("Sending request to index database with body:", body);
|
|
|
|
try {
|
|
const response = await this.request('/index_database', 'POST', body, {
|
|
'Content-Type': 'application/json'
|
|
});
|
|
console.log("Index database response:", response);
|
|
return response;
|
|
} catch (error) {
|
|
console.error("Error indexing database:", error);
|
|
throw error;
|
|
}
|
|
}
|
|
async search(query) {
|
|
const searchQuery = { query, key: this.key };
|
|
return this.request('/search', 'POST', searchQuery);
|
|
}
|
|
|
|
async wipeDatabase() {
|
|
const body = { key: this.key };
|
|
return this.request('/wipe_database', 'DELETE', body);
|
|
}
|
|
}
|
|
|
|
|
|
// Example usage:
|
|
// const ragClient = new RAGClient('http://localhost:8000', 'your_bearer_token');
|
|
// ragClient.addDocument('My Title', 'This is the content of the document.')
|
|
// .then(response => console.log(response))
|
|
// .catch(error => console.error(error));
|