#30daysofllms - Twitter Hashtag

2 days ago

day 21 of #30DaysOfLLMs continued my Transformer deep dive - realized tinyGPT uses a decoder-only architecture (modern GPT-style), not the original encoder-decoder design focusin on the original arch first, will cross-ref what's changed later when I have clarity #30daysOfCode

coding_robin's tweet photo. day 21 of #30DaysOfLLMs

continued my Transformer deep dive

- realized tinyGPT uses a decoder-only architecture (modern GPT-style), not the original encoder-decoder design

focusin on the original arch first, will cross-ref what's changed later when I have clarity

#30daysOfCode https://t.co/f80Njz306D

0

8

Adewole @coding_robin

3 days ago

day 20 of #30DaysOfLLMs continued my deep dive on Transformers - tried to read & dissect 'Attention Is All You Need' - the 2017 paper that introduced the Transformer architecture fell asleep halfway, so much technical jargon lol try again tomorrow #30daysOfCode

coding_robin's tweet photo. day 20 of #30DaysOfLLMs

continued my deep dive on Transformers

- tried to read & dissect 'Attention Is All You Need' - the 2017 paper that introduced the Transformer architecture

fell asleep halfway, so much technical jargon lol

try again tomorrow

#30daysOfCode https://t.co/sRCpj0IOZm

0

8

Adewole @coding_robin

4 days ago

Day 19 of #30DaysOfLLMs circled back to Transformers - the beating heart of modern LLMs - learned its origins: original transformer architecture introduced by google in 2017 - had encoder&decoder layers tho modern LLMs use diff approaches (eg GPT = decoder-only) #30daysOfCode

coding_robin's tweet photo. Day 19 of #30DaysOfLLMs

circled back to Transformers - the beating heart of modern LLMs

- learned its origins: original transformer architecture introduced by google in 2017

- had encoder&decoder layers tho modern LLMs use diff approaches (eg GPT = decoder-only)

#30daysOfCode https://t.co/oeL2LlonY5

0

1

0

16

Adewole @coding_robin

5 days ago

day 18 of #30DaysOfLLMs still on tokenization - learnd about the BPE (Byte Pair Encoding) algo and why it's commonly used for subword tokenization - attempted a BPE subword tokenizer frm scratch, trained on my dataset got my hands dirty, learned a few things... #30DaysOfCode

coding_robin's tweet photo. day 18 of #30DaysOfLLMs

still on tokenization

- learnd about the BPE (Byte Pair Encoding) algo and why it's commonly used for subword tokenization

- attempted a BPE subword tokenizer frm scratch, trained on my dataset

got my hands dirty, learned a few things...

#30DaysOfCode https://t.co/rKNnlRqja5

0

12

Adewole @coding_robin

6 days ago

day 17 of #30DaysOfLLMs went down the rabbit hole of tokenization - explored tokenization libraries (hugging face, SentencePiece) and how to integrate them with tinyGPT - learned the difference btwn trainin a tokenizer on ur own dataset vs usin a pretrained one #30DaysOfCode

coding_robin's tweet photo. day 17 of #30DaysOfLLMs

went down the rabbit hole of tokenization

- explored tokenization libraries (hugging face, SentencePiece) and how to integrate them with tinyGPT

- learned the difference btwn trainin a tokenizer on ur own dataset vs usin a pretrained one

#30DaysOfCode https://t.co/WBpq5Y4P5L

0

64

Adewole @coding_robin

7 days ago

day 16 of #30DaysOfLLMs more text generation(inference) tests w tinyGPT - learned to scale logits with temperature before samplin - controls how risky vs safe next token predictions are 1 = neutral, below 1 = predictable, above 1 = riskier/random fun experiment #30DaysOfCode

coding_robin's tweet photo. day 16 of #30DaysOfLLMs

more text generation(inference) tests w tinyGPT

- learned to scale logits with temperature before samplin - controls how risky vs safe next token predictions are

1 = neutral, below 1 = predictable, above 1 = riskier/random

fun experiment

#30DaysOfCode https://t.co/wefi64Nigp

0

14

pdawg

@prathamgrv

over 1 year ago

I think #100DaysOfCode is kind of overrated should I do #30DaysOfLLMs instead - sharing cool, informative blogs that even people with minimal ML background can understand

31

344

6

88

15K

Top Tweets for #30daysofllms

Last Seen Hashtags on Sotwe

Trends for you

Most Popular Users