News for hackers. Stuff that matters.

convexstrictly

convexstrictly on Hacker News

Re:Gigatoken: Fastest Tokenizer on Wednesday July 22, 2026 @06:33AM (#49004530)
Tokenization is done on the CPU. Models never see the raw characters. That's why you get trick questions like the number of r's in strawberry.

There are many research papers on models using characters directly. One challenge is that effective context length is smaller.

Re:Gigatoken: Fastest Tokenizer on Tuesday July 21, 2026 @08:14PM (#49000160)
GitHub: https://github.com/marcelroed/gigatoken
Re:Gemini Flash 2.0 Thinking Experimental on Thursday December 19, 2024 @12:05PM (#42463446)
Video Demo:

https://x.com/OfficialLoganK/status/1869789822384255300

Re:Gemini Flash 2.0 Thinking Experimental on Thursday December 19, 2024 @12:04PM (#42463438)
"Just when you thought it was over... we’re introducing Gemini 2.0 Flash Thinking, a new experimental model that unlocks stronger reasoning capabilities and shows its thoughts.

The model plans (with thoughts visible), can solve complex problems with Flash speeds, and more ..."

- Logan Kilpatrick

https://x.com/OfficialLoganK/status/1869789822384255300

Re:M4 MacBook Pro on Wednesday October 30, 2024 @05:19PM (#42000457)
GB per second
Re:ThunderKittens: Simple, Fast, and Adorable AI Kernels on Wednesday October 30, 2024 @04:59PM (#42000236)
Simran Arora: "Join us for a livestream this Thursday, Halloween/Diwali, and join our channel on the GPU Mode Discord server to hang out with us/get involved:"

https://discord.com/login?redirect_to=%2Fchannels%2F11894982...

Re:ThunderKittens: Simple, Fast, and Adorable AI Kernels on Wednesday October 30, 2024 @04:48PM (#42000100)
CUDA + ThunderKittens 4.5 hour tutorial

https://www.youtube.com/watch?v=xcpEl0cGCC4

Re:Building GPT2o – Part 1: Audio on Friday June 14, 2024 @07:35AM (#40679800)
Results https://x.com/sbeastwindy/status/1801525876267372874
Re:LSP-AI: open-source language server serving as back end for AI code... on Saturday June 08, 2024 @05:54PM (#40620423)
Aider uses Treesitter to improve code generation. https://aider.chat/2023/10/22/repomap.html

Aider: https://github.com/paul-gauthier/aider

It is state of the art on SWE-Bench and SWE-Bench Lite. https://aider.chat/2024/06/02/main-swe-bench.html

Re:Llm.c – LLM training in simple, pure C/CUDA on Monday April 08, 2024 @10:40PM (#39975812)
Candle is a minimalist ML framework for Rust with a focus on performance (including GPU support) and ease of use

https://github.com/huggingface/candle

Re:LISA: Layerwise Importance Sampling for Memory-Efficient LLM Fine-T... on Wednesday March 27, 2024 @01:19PM (#39841943)
An author claims better performance than LoRA in 50% of the time.

https://twitter.com/Rui45898440/status/1772996453557997924

Re:NTIA AI Open Model Weights RFC on Tuesday March 26, 2024 @08:33PM (#39834487)
The federal government requests comments on regulation of AI models with openly available weights. The deadline is March 27, 2024.

Earlier thread. https://news.ycombinator.com/item?id=39494760

Re:Show HN: GPU Price list inspired by diskprices.com on Friday March 08, 2024 @05:03PM (#39647068)
It would be good to know on which listings Amazon is the seller. Filtering by that criterion may also be useful.

For professional cards, I've noticed dihuni.com has good prices. I have never purchased from them and have no idea what dealing with them is like.

Re:Star Trek prompt optimal for grade school math on Llama-70B on Monday February 26, 2024 @10:31AM (#39512401)
The results are from the paper

The Unreasonable Effectiveness of Eccentric Automatic Prompts

https://arxiv.org/abs/2402.10949

Older »