Footer

Your go-to destination for discovering trending open-source projects and uncovering the insights that matter

Legal

  • Terms and conditions
  • Privacy policy

More

  • Pricing
  • Advertise
  • onedump
  • gosnakego
  • Contact us

Social

  • X (Twitter)
  • GitHub
Trendshift owl
What they build speaks louder than words
© 2026 Trendshift, created by Julian Li with ❤️
SignalAdvertise
Trendshift
SignalAdvertise
Sign in
Trendshift owl

Discover trends that matter

SignalAdvertise
DailyWeeklyMonthlyYearly
Live mentionsTopics
RepositoriesDevelopers
InsightsStats

Trina AX
Trina AX
Bifrost
Bifrost
It's a plan
It's a plan
OfinTech Desk
OfinTech Desk
Busbar
Busbar
Omnigraph
Omnigraph

Vaibhaav-Tiwari/deepseek-v3-jax

DeepSeek V3 (MLA, DeepSeekMoE, aux-loss-free balancing, MTP) in ~200 lines of pure JAX, trained on a MacBook Air.

#LLM training
Visit GitHub
Python
260
20
1 contributors
MIT License
website
FeaturedAdvertise to 180K+ monthly visitors
Trina AX

A self-hosted team AI workspace for company files, with per-person folder permissions

Bifrost

Fastest Enterprise AI Gateway for Routing, Governing, and Securing all your AI Traffic

It's a plan

Open-source Linear alternative

An issue tracker where AI agents take the work you assign them

OfinTech Desk

An open-source AI workspace for finance and accounting teams, in which every figure is traced to its source

Busbar

Control what AI can do before it acts

Omnigraph

Versioned graph database on object storage. Git-style workflows and multimodal retrieval

Momentum

How fast Vaibhaav-Tiwari/deepseek-v3-jax is gaining stars, and where it lands on Trendshift and GitHub Trending

Activity and trending

Stars, forks and merged pull requests each day or month, lined up with the repository's rank on Trendshift Trending

Nov 13 – Nov 22, 2026

Stars +260 in 10 days

050100150
108

Forks +20 in 10 days

0510
10

Merged PRs 1 in 10 days

01
1

Trendshift Trendingwhere the repository ranked each day it made the top 25

Not in the Trendshift top 25 in this period

Nov 13Nov 15Nov 17Nov 19Nov 21
The left and right arrow keys step from day to day; Home and End go to the first and the last

Stars, forks, merged pull requests and Trendshift Trending rank, by day
DayStars gainedForks gainedMerged PRs
Fri, Nov 13, 2026+0+00
Sat, Nov 14, 2026+88+100
Sun, Nov 15, 2026+108+71
Mon, Nov 16, 2026+36+30
Tue, Nov 17, 2026+15+00
Wed, Nov 18, 2026+10+00
Thu, Nov 19, 2026+0+00
Fri, Nov 20, 2026+2+00
Sat, Nov 21, 2026+1+00
Sun, Nov 22, 2026+0+00
Total+260+201
  • 1#1 on the list
  • 7Rank that day
  • Not trending

Social mentions

Recent discussions about this repository across the web
  1. Nov 165 posts6 posts
    • I've written React components longer than this. Someone recreated DeepSeek V3's core architecture in 190 lines of JAX. MLA, MoE, multi-token prediction, all in readable code. And they trained a tiny…

      Read full on X, posted by @dongshenqnx1

      @dongshenqnx1 · x.com

    • 约 200 行纯 Jax,写一个 DeepSeek V3。这是 VaibhaavTiwari 今天在 Show HN 上发的项目。 200 行,一口气能读完。真上生产,分布式、显存、算子这些会把结构埋掉。压到 200 行,页上就剩结构。 先看能不能加载官方权重、跑出输出。能,就拿官方输出逐层对照,哪一步写错了会直接暴露;不能,就当结构笔记翻,学的是模块怎么拼起来,不能当成能直接跑的代码。

      Read full on X, posted by @iluciddreaming

      @iluciddreaming · x.com

    • 有人用 190 行纯 JAX 写了 DeepSeek V3 架构。 MLA、DeepSeekMoE、无辅助损失均衡和 MTP 都在 deepseek_v3.py 这一个文件里。 总参数 36.2M,每个 token 激活 13.6M。 作者在 24GB 的 MacBook Air M5 上从头训了一遍。 跑了 12,000 步,大概花了 2.6 小时。 验证 loss 从 5.02 降到…

      Read full on X, posted by @Jason7ole

      @Jason7ole · x.com

    • One file. 190 lines. DeepSeek V3's whole architecture, in JAX, no PyTorch. MLA, the MoE routing, aux-loss-free balancing, multi-token prediction. He trained it on a MacBook Air: 36.2M params, 207…

      Read full on X, posted by @codvynn

      @codvynn · x.com

    • 开源模型这个周末有点热闹,两件事都值得本地部署党看一眼。 第一件:Unsloth 给 Qwen3.8-27B 换了 Dynamic V3.1 量化,同体积精度涨 5-10%(Terminal Bench、DeepSWE 这些 agent 基准)。4-bit 版 16-19GB 就能跑,4090、5080 或者 24GB 内存的 Mac 都行;2-bit 甚至 9-11GB 就够。HN 上有人用…

      Read full on X, posted by @anti_default_99

      @anti_default_99 · x.com

    • DeepSeek V3 fits in one file. ~200 lines of pure JAX, no training framework. Runs in ~2.6 hrs on a MacBook Air. The full architecture is in there: MLA, DeepSeekMoE, MTP. Readable in one sitting.…

      Read full on X, posted by @AIAgentUseCase

      @AIAgentUseCase · x.com

  2. Nov 154 posts
    • someone rebuilt deepseek v3 in 190 lines of pure jax (on hn). counted: 190 code lines, 265 in the file. the checkpoint is a pickle, and a pickle can run code on load. so before running it on my mac i…

      Read full on X, posted by @daniiie7

      @daniiie7 · x.com

    • Someone rebuilt DeepSeek V3 in ~190 lines of pure JAX and trained it on a MacBook Air. One number matters for anyone serving MoE. Without the routing bias, expert load ranged from 0.09% to 18.6%.…

      Read full on X, posted by @thehenrylabs

      @thehenrylabs · x.com

    • DeepSeek V3 fits in 190 lines of code. Someone rebuilt it in plain JAX and trained a tiny version on a MacBook Air in 2.6 hours. - 36M params, no PyTorch - weights and data included - free & open…

      Read full on X, posted by @zyvre_ic

      @zyvre_ic · x.com

    • Lovable apps expose passport details and more than 200K private messages (

      Read full on X, posted by @betterhn50

      @betterhn50 · x.com

You might also like

Repositories close to Vaibhaav-Tiwari/deepseek-v3-jax
FeaturedAdvertise to 180K+ monthly visitors
OfinTech Desk

An open-source AI workspace for finance and accounting teams, in which every figure is traced to its source

Trina AX

A self-hosted team AI workspace for company files, with per-person folder permissions

It's a plan

Open-source Linear alternative

An issue tracker where AI agents take the work you assign them

Busbar

Control what AI can do before it acts

Bifrost

Fastest Enterprise AI Gateway for Routing, Governing, and Securing all your AI Traffic

Omnigraph

Versioned graph database on object storage. Git-style workflows and multimodal retrieval

jax-ml/jax-llm-examples

Python

Minimal yet performant LLM examples in pure JAX

29040

AakashKumarNain/nanoGPTJAX

Python

Implementing scalable LLMs in pure JAX (no third-party libraries)

568

google-deepmind/simply

Python

Minimal and scalable research codebase in JAX, designed for rapid iteration on frontier research in LLM and other autoregressive models.

59074

xai-org/grok-1

Python

Grok open release

52.2k8.7k

AI-Hypercomputer/maxtext

Python

A simple, performant, and scalable Jax LLM!

2.5k651

deepseek-ai/DeepSeek-V3.2-Exp

Python
1.7k200

NX-AI/xlstm-jax

Python

Official JAX implementation of xLSTM including fast and efficient training and inference code. 7B model available at https://huggingface.co/NX-AI/xLSTM-7b.

1057

state-spaces/mamba

Python

Mamba SSM architecture

19k1.9k

deepseek-ai/DeepSeek-V3

Python
104.9k17.3k