-
responsive web using rem
Using rem (default 16px), one can write web page containing html elements and texts in harmony. Default root element font size is 16px, but many users override it for better readability. Use of rem respects that. https://www.sitepoint.com/understanding-and-using-rem-units-in-css/ Setting default font size and browser zoom level are two different controls. Browser zoom scales up everything by…
Tags:
-
Polyak averaging and gradient accumulation in the Keras
So I discovered them in the tweets. As the paper Polyak Parameter Ensemble: Exponential Parameter Growth Leads to Better Generalization shows, Polyak Ensemble is for better generalization. Another in the tweet is gradient accumulation. I don’t know if I’ll need it since I often makes the batch size as large as possible.
Tags:
-
저성과자와 일하는 고민
성과가 아주 좋은 개발자는 그렇지 못한 개발자에 비해 100배 더 뛰어날까? 나는 그런 개발자가 팀을 이끌고 조직을 이끌면 그럴 수 있다고 본다. 그러나 1대 1로 붙었을 때 평균적인 개발자보다 100배가 뛰어나는 것은 불가능하다고 본다. 예를들어 워드 프레스 테마를 초기화하는 문제에대해 뛰어난 개발자는 누구보다 훨씬 빠르게 코드에서 답을 찾아 낼 수는 있겠지만 어떻게 초기화하는게 정석인지 까지…
Tags:
-
How to erase customization in a wordpress template
It was so painful to find out this. Deleting and reinstalling templates don’t reset the templates to the original one. I tried plugins like wp reset to reset theme, and tried database cleaner to find database table containing the customziation. I connected to the database in a command line and checked all the tables, but…
Tags:
-
How to debug user warning in pydantic
When pydantic comes across serialization issues, it outputs: But that doesn’t leave much trace to debug. To fix that, change warning filter. Then now you’ll see stack trace was the warning becomes an exception.
Tags:
-
How to wait for ProcessPoolExecutor and ThreadPoolExecutor
If your wait() doesn’t work, it’s probably due to exception that wasn’t caught. Catch it.
Tags:
-
How do I compute sha256 for arbitrary python data
Computing sha256 for python data is useful as ID or as filename. See this test.
Tags:
-
How to print intermediate values in the langchain chain
I don’t know why this isn’t documented anywhere. Let’s say there are two prompts. And then you add RunnableLambda.
Tags:
-
LLM의 한계에 대하여
LLM이 인간의 모든 지식을 담을 수 없다는 흥미로운 글을 읽었다. 인공지능이 인간의 지식을 다 표현할 수 있는가 아닌가는 상당히 오래된 주제이지만, 이 글의 열정적인 톤과 풍부한 사례는 무척 재미있었다. 예를들어 글에서는 performer (예를들어 피겨 스케이팅 선수) 의 perform 하는 기술은 글로 기술되지 않으며 글로 기술된다 하였다고 해도 그것이 그대로 그 퍼포먼스를 따라할 수 있단 이야기는…
-
Positional encoding for timeseries data
This is positional encoding for timseries data inspired by A Transformer-based Framework for Multivariate Time Series Representation Learning. Assuming that the input is (batch_size, sequence_length, feature_dim) and each feature is a float, the way we add positional encoding is: This is applied like: See PositionalEncoding of keras_nlp here for generic implementation.
Tags: