iamlemec

#!/usr/bin/env bash

# Step 1: put this file in your path and make executable
# Step 2: add the following to your .gitattributes file
#   *.ipynb diff=nb2md
# Step 3: add the following to your .git/config
#   [diff "nb2md"]
#       textconv = nb2md
# or to it globally with
#   git config --global diff.nb2md.textconv nb2md

iamlemec

@import url(//fonts.googleapis.com/css?family=Open+Sans:400,700,400italic,700italic);

body {
	background-color: white;
	font-family: 'Open Sans', sans-serif;
}

#content {
	width: 700px;
	margin-top: 50px;

iamlemec

#!/bin/sh

# Use this instead of diff[1] to get colored[2] word-based diffs.
# Useful for text documents that have reflowed paragraphs.
# Requires that wdiff is installed in your $PATH.
#
# [1] All diff options are ignored. Only replaces simplest usage.
# [2] Colors are always emitted. If piping into less, use "-R" or set LESS=-R

# Iain Murray, February 2009, Tweaked in June 2011

iamlemec

def check_tokenizer(mod_ll, mod_hf, data, max_rows=None):
    from llama_cpp import Llama
    from transformers import AutoTokenizer
    from Levenshtein import editops
    from termcolor import cprint

    # load models
    if type(mod_ll) is str:
        mod_ll = Llama(mod_ll, verbose=False)
    if type(mod_hf) is str:

iamlemec

import torch
from transformers.models.roberta import RobertaConfig, RobertaModel, RobertaTokenizer

# load model and tokenizer
tokenizer = RobertaTokenizer.from_pretrained('FacebookAI/roberta-base')
model = RobertaModel.from_pretrained('FacebookAI/roberta-base', is_decoder=True).to('cuda')

# tokenize inputs
text = 'hello world, this is a test'
inputs = tokenizer(text, return_tensors='pt').to('cuda')