veekaybee

Anti-hype LLM reading list

Goals: Add links that are reasonable and good explanations of how stuff works. No hype and no vendor content if possible. Practical first-hand accounts of models in prod eagerly sought.

Foundational Concepts

Pre-Transformer Models

abhishekkrthakur

import torch
from datasets import load_dataset
from peft import LoraConfig, get_peft_model, prepare_model_for_int8_training
from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments
from trl import SFTTrainer


def train():
    train_dataset = load_dataset("tatsu-lab/alpaca", split="train")
    tokenizer = AutoTokenizer.from_pretrained("Salesforce/xgen-7b-8k-base", trust_remote_code=True)

kevinkub

// Licence: Robert Koch-Institut (RKI), dl-de/by-2-0

class IncidenceWidget {

  constructor() {
    this.previousDaysToShow = 31;
    this.apiUrlDistricts = (location) => `https://services7.arcgis.com/mOBPykOjAyBO2ZKk/arcgis/rest/services/RKI_Landkreisdaten/FeatureServer/0/query?where=1%3D1&outFields=RS,GEN,cases7_bl_per_100k,cases7_per_100k,BL&geometry=${location.longitude.toFixed(3)}%2C${location.latitude.toFixed(3)}&geometryType=esriGeometryPoint&inSR=4326&spatialRel=esriSpatialRelWithin&returnGeometry=false&outSR=4326&f=json`
    this.apiUrlDistrictsHistory = (districtId) => `https://services7.arcgis.com/mOBPykOjAyBO2ZKk/ArcGIS/rest/services/Covid19_hubv/FeatureServer/0/query?where=IdLandkreis%20%3D%20%27${districtId}%27%20AND%20Meldedatum%20%3E%3D%20TIMESTAMP%20%27${this.getDateString(-this.previousDaysToShow)}%2000%3A00%3A00%27%20AND%20Meldedatum%20%3C%3D%20TIMESTAMP%20%27${this.getDateString(1)}%2000%3A00%3A00%27&outFields=Landkreis,Meldedatum,AnzahlFall&outSR=4326&f=json`
    this.stateToAbbr = {