gautierdag

import openai

text_to_embeddings_cache = {}

def get_embeddings_for_instructions(instructions: list[str]):
        response = openai.Embedding.create(
            input=instructions,
            model="text-embedding-ada-002"
        )
        return response['data']

gautierdag

import pandas as pd
from datasets import Dataset
from transformers import (
    AutoModelForMaskedLM,
    AutoTokenizer,
    DataCollatorForLanguageModeling,
    Trainer,
    TrainingArguments,
)

gautierdag

import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from transformers import AutoModel, AutoTokenizer
from transformers.optimization import get_cosine_schedule_with_warmup
import pytorch_lightning as pl

# from deepspeed.ops.adam import FusedAdam - get different error (expected tensor on cuda but got cpu) with FusedAdam

class BoringDataset(torch.utils.data.Dataset):

gautierdag

import torch
from torch.optim import AdamW
from transformers import AutoModel

def get_bert_layerwise_lr_groups(bert_model, learning_rate=1e-5, layer_decay=0.9):
    """
    Gets parameter groups with decayed learning rate based on depth in network
    Layers closer to output will have higher learning rate
    
    Args:

gautierdag

for name in $(aws dynamodb list-tables | jq .TableNames)
do
    l=${#name} 
    if [ ${#name} -le 3 ]; then echo "Skipping name!"
    else
       short=${name:1:l-2} # clean json string representation of dataset
       if [ "${short: -1}" == '"' ]
         then short=${name:1:l-3}
       fi
       echo "Updating Table: ${short}"

gautierdag

import torch
import torch.nn as nn
import pytorch_lightning as pl

class BoringModel(pl.LightningModule):
    def __init__(
        self
    ):
        super(BoringModel, self).__init__()
        self.automatic_optimization = False

gautierdag

import torch
import torch.nn as nn
import pytorch_lightning as pl

class NCE(pl.LightningModule):
    """
    This implementation is taken from https://github.com/Spijkervet/SimCLR/blob/master/simclr/modules/nt_xent.py
    The mask_correlated_samples funtion has been modified to be much faster to compute
    and therefore be able to be called at train time without a predifined batch size.
    """

gautierdag

import torch
import pytorch_lightning as pl

class MyTransformer(pl.LightningModule):
    def __init__(
        self,
        learning_rate=0.001,
        warmup=4000,
    ):
      self.learning_rate = learning_rate

gautierdag

# https://apple.stackexchange.com/questions/175215/how-do-i-assign-a-keyboard-shortcut-to-an-applescript-i-wrote

on run {input, parameters}
	
	tell application "System Events"
		key code 123 using command down
		keystroke "print("
		key code 124 using command down
		keystroke ")"
	end tell

gautierdag

import numpy as np

#cast as object to be able to set np.nan and handle different types
arr = df[cols].values.astype(object)

# build shifted array
shifted = np.roll(arr, 1)
shifted[0] = np.nan

# choose indexes based on shift comparaison