filipg/aitech-moj-2023
1
5
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix

Browse Source
This commit is contained in:
Filip Gralinski 2022-06-11 08:36:47 +02:00
parent 33a7a1f83e
commit 6da452bc4f
1 changed files with 18 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
wyk/11_Transformer.org
View File

@ -90,25 +90,23 @@ Dokonajmy najpierw tokenizacji:
#+BEGIN_SRC python :session mysession :exports both :results raw drawer #+BEGIN_SRC python :session mysession :exports both :results raw drawer
from transformers import AutoTokenizer from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("gpt2") tokenizer = AutoTokenizer.from_pretrained("gpt2")
text = "The World War III will begin in" text = "The World War III will begin in 2028 in"
encoded_input = tokenizer(text, return_tensors='pt') encoded_input = tokenizer(text, return_tensors='pt')
encoded_input encoded_input
#+END_SRC #+END_SRC
#+RESULTS: #+RESULTS:
:results: :results:
{'input_ids': tensor([[ 464, 2159, 1810, 6711, 481, 2221, 287]]), 'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1]])} {'input_ids': tensor([[ 464, 2159, 1810, 6711, 481, 2221, 287, 1160, 2078, 287]]), 'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])}
:end: :end:
Możemy podejrzeć uzyskane tokeny:
#+BEGIN_SRC python :session mysession :exports both :results raw drawer #+BEGIN_SRC python :session mysession :exports both :results raw drawer
[tokenizer.decode(i) for i in encoded_input.input_ids[0]] [tokenizer.decode(i) for i in encoded_input.input_ids[0]]
#+END_SRC #+END_SRC
#+RESULTS: #+RESULTS:
:results: :results:
['The', ' World', ' War', ' III', ' will', ' begin', ' in'] ['The', ' World', ' War', ' III', ' will', ' begin', ' in', ' 20', '28', ' in']
:end: :end:
Zwróćmy uwagę, że w GPT-2 tokeny obejmują spacje! Zwróćmy uwagę, że w GPT-2 tokeny obejmują spacje!
@ -125,6 +123,15 @@ Teraz uruchommy zasadniczy model:
:results: :results:
:end: :end:
#+BEGIN_SRC python :session mysession :exports both :results raw drawer
softmax(outputs[0][0][-1])
#+END_SRC
#+RESULTS:
:results:
:end:
Z modelu GPT-2 otrzymamy rozkład prawdopodobieństwa kolejnego wyrazu, najpierw w postaci Z modelu GPT-2 otrzymamy rozkład prawdopodobieństwa kolejnego wyrazu, najpierw w postaci
nieznormalizowanych *logitów*: nieznormalizowanych *logitów*: