Show HN: DeepSeek V3 in ~200 lines of pure Jax
Source: Show HN ·
DeepSeek V3 (MLA, DeepSeekMoE, aux-loss-free balancing, MTP) in 190 lines of pure JAX, trained on a MacBook Air. Write-up: vaibhaav.in/writing/deepseek-v3-jax - Vaibhaav-Tiwari/deepseek-v3-jax
Frequently Asked Questions
What is Show HN: DeepSeek V3 in ~200 lines of pure Jax?
Show HN: DeepSeek V3 in ~200 lines of pure Jax is a tool discovery curated from Show HN. DeepSeek V3 (MLA, DeepSeekMoE, aux-loss-free balancing, MTP) in 190 lines of pure JAX, trained on a MacBook Air. Write-up: vaibhaav.in/writing/deepseek-v3-jax …
How can I access or experience this?
You can experience this content directly on its original host (Show HN) by clicking the "Open Original" button.
Where does this content come from?
Curated from Show HN by the OnSuggest team to surface the most interesting and serendipitous content on the web.
More like this
All Developer tools discoveries →Show HN: Hello Triangle, a puzzle game where you build a GPU
I've been working on a game Hello Triangle for a while. In the game, you build a GPU one part at a time by wiring circuits, and renders a picture on the screen with what your circuit computes. I have a free demo available that goes up to drawing a triangle, while I'm still working on the rest of the game. You don't start from logic gates. The game starts with adders and comparators. Every wire just carries a fixed point number, and there are no bit widths to think about. The reason of using fixed point numbers is that veteran players of Turing Complete can skip the bits part they already playe
Show HN: ShadowDiff – Shadow testing and SQL diffing for AI-migrated back ends
Dual-Traffic Shadow Testing Engine for AI-Refactored Legacy Code | Real-time traffic teeing, semantic diffing & automated LLM regression test generation - vishalrajofficial/ShadowDiff
Show HN: Cshoal – A school of fish, hunted by predators, in your terminal
The movement, behavior, and interactions of the fish and predators come from published studies, cited in the readme. The idea and direction are mine, but Claude wrote the code. I just wanted it to exist. Comments URL: https://news.ycombinator.com/item?id=50504822 Points: 1 # Comments: 0
Show HN: IndexFast – see which of your pages search engines actually know about
Add a domain, host one key file, and IndexFast submits new and changed URLs to IndexNow, Bing, Yandex, Baidu and Search Console. MCP server and REST API on every plan.
Show HN: Dynaura – Dynamic Island toasts for the web
dynaura is an open-source (MIT-licensed), zero-dependency UI toast notification library for the web, designed around an Apple-style "Dynamic Island" pill that spring-morphs between compact and expanded states. Comments URL: https://news.ycombinator.com/item?id=50504910 Points: 1 # Comments: 0
Show HN:Learn Chinese Through Mythology and HSK Graded Stories
Master Chinese with HSK 1–7 graded mythology stories, pinyin, instant word lookup, an interactive deity graph, and FSRS flashcards. Start reading free.