mcyc

6 ms·

1.
by mcyc 3mo ago
Lichess has a checkmate captcha that I think is cute. It requires you to solve a mate-in-one puzzle to, e.g., post on the forums. (Sorry, don't have a better link, there wasn't any non-technical I could find about it). https:&#x2
2.
by mcyc 3mo ago
Yeah, it is interesting to me that it is coming from the _city_'s government. I've seen sovereign AI things at the country level, but this is the first municipal one I have seen.
3.
Rio 3.5 Open 397B – from Rio de Janeiro's city government(huggingface.co)
17 points by mcyc 3mo ago|7 comments
4.
Finding Optimal Tokenizers(blog.aqnichol.com)
29 points by mcyc 3mo ago|2 comments
5.
by mcyc 4mo ago
You are right about most tokenizers being heavily biased towards English, but the situation is not so bad for Portuguese. Here are some results on the Goldfish corpus [1] with a few different tokenizers. This measures #characters in corpus
6.
Show HN: blunder.clinic, realistic daily chess puzzles(blunder.clinic)
4 points by mcyc 6mo ago|0 comments
7.
US science after a year of Trump(nature.com)
18 points by mcyc 8mo ago|0 comments
8.
by mcyc 8mo ago
This is a fantastic guide! I did a lot of work on structured generation for my PhD. Here are a few other pointers for people who might be interested: Some libraries: - Outlines, a nice library for structured generation - https:/&#x
9.
Bolmo: Byteifying the Next Generation of Language Models(arxiv.org)
2 points by mcyc 9mo ago|0 comments
10.
Pebble Index(repebble.com)
1 points by mcyc 9mo ago|1 comments
11.
Whisper Leak: A novel side-channel attack on remote language models(microsoft.com)
3 points by mcyc 10mo ago|1 comments
12.
There is no such thing as a tokenizer-free lunch(huggingface.co)
2 points by mcyc 10mo ago|0 comments
13.
The Chinese Computer: A Global History of the Information Age(foreignaffairs.com)
2 points by mcyc 10mo ago|0 comments
14.
Masked Softmax Layers in PyTorch(mcognetta.github.io)
4 points by mcyc 11mo ago|0 comments
15.
by mcyc 1y ago
This is a nice attitude. I think HN is overall pretty nice for geeking out and also hearing other people geek out, but there is still a strain of elitism (not like StackExchange thankfully) and so I'm happy to see comments like this.
16.
Minimal Boolean Formulas (2011)(research.swtch.com)
103 points by mcyc 1y ago|19 comments
17.
Fulbright Board Resigns After Accusing Trump Aides of Political Interference(nytimes.com)
6 points by mcyc 1y ago|1 comments
18.
by mcyc 1y ago
You can cross whitespace boundaries by setting flag `--split-on-whitespace` to false (it's true by default). https://github.com/google/sentencepiece/blob/master/doc/opti...
19.
by mcyc 1y ago
Just a minor nit: SentencePiece is a library, not a tokenization algorithm. It implements two tokenization algorithms, Unigram and BPE. BPE builds vocabularies from the base up so I assume you are talking about Unigram which starts with a b
20.
by mcyc 2y ago
People may also be interested in Pynini [1], a python wrapper (+ a lot of additional ease-of-use functionality) of OpenFst [2] (a really great library for transducers). There are some good tutorials in the form of homework assignments (from
21.
Adapting nnue-PyTorch's binary position format for Lichess(lichess.org)
1 points by mcyc 2y ago|0 comments
22.
Fine-tune classifier with ModernBERT in 2025(philschmid.de)
19 points by mcyc 2y ago|3 comments
23.
by mcyc 2y ago
You might be interested in the Recurse Center ( https://www.recurse.com/ ) and the experiences of people who have gone through it (they heavily encourage blogging about your time there so there is lots to read). Note: I am no
24.
by mcyc 2y ago
Thanks! Our paper [1] is kind of a goofy adversarial thing where we thought "here's this cool metric, how can we break it?". The tokenizers we propose are definitely not tokenizers you should use in practice. The original pap
25.
by mcyc 2y ago
It's two problems: 1) the sequence length increases too much. Idk what the average token length is for Llama, but imagine it's like 5+ bytes. Using individual bytes as tokens immediately makes the context 5x longer which is super
26.
by mcyc 2y ago
NB: Can't edit my original reply. Sorry actually I misread part of your comment in relation to the paper and confused δ and another parameter, K. To clarify, δ is the number of tokens in the tokenized corpus and K is the size of the vo
27.
by mcyc 2y ago
Hi, I'm Cognetta from the above Cognetta et al. I can't answer all of your questions (and I can't speak for the authors of this paper ofc), but I will try to answer some. > Is a tokenizer that maximizes the compression of
28.
The Lichess Game Compressor's Analysis of Game 1 of the World Championships(mcognetta.github.io)
5 points by mcyc 2y ago|0 comments
29.
by mcyc 2y ago
It's just a convention for SIMD functions and types.
30.
Call for Developer Projects(github.com)
91 points by mcyc 2y ago|35 comments

More ›