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

dayyass

dayyass

dayyass

import torch
from torch.nn.utils.rnn import pack_sequence, pad_packed_sequence


def unpack_sequence(packed_sequences):
    """Unpacks PackedSequence into a list of variable length Tensors"""
    
    unpacked_sequences = []
    
    padded_sequences, lengths = pad_packed_sequence(packed_sequences, batch_first=True)

dayyass

from itertools import permutations

import numpy as np
from sklearn.metrics import accuracy_score

np.random.seed(42)

y_true = np.random.randint(low=0, high=3, size=100)

noize_mapper = {0: 1, 1: 2, 2: 0}

dayyass

from sklearn.feature_extraction.text import TfidfVectorizer

# pymorphy2 lemmatizer
import pymorphy2

class Lemmatizer:
    def __init__(self):
        self.morph = pymorphy2.MorphAnalyzer()
        
    def __call__(self, x: str) -> str:

dayyass

from sklearn.feature_extraction.text import TfidfVectorizer

# data
corpus = [
    'This is the first document.',
    'This document is the second document.',
    'And this is the third one.',
    'Is this the first document?',
]

dayyass

import pymorphy2


class Lemmatizer:
    """
    Pymorphy2 lemmatizer class.
    """
    
    def __init__(self):
        """

dayyass

def humanize_bytes(bytes: int, suffix: str = "B") -> str:
    """
    Convert bytes to human readable format.

    :param int bytes: number of bytes.
    :param str suffix: bytes suffix.
    :return: human readable size.
    :rtype: str
    """

dayyass

FROM cr.msk.sbercloud.ru/aicloud-jupyter/jupyter-cuda10.1-tf2.2.0-mlspace:latest
MAINTAINER Dani El-Ayyass <[email protected]>

USER root

# Docker
# Set up the repository
RUN apt-get update
RUN apt-get -y install apt-transport-https ca-certificates curl gnupg lsb-release