Jackmin (@jackminong) / X

Jackmin

779 posts

Jackmin

@jackminong

making sand smarter @PrimeIntellect 🇺🇸 Previously @JinaAI_ 🇩🇪 @MoneyLion 🇲🇾

San Francisco, CA

Joined October 2021

Pinned
Jackmin
@jackminong
May 1, 2025
excited to share that TOPLOC has been accepted at ICML 2025! see you in Vancouver 🇨🇦
Prime Intellect
@PrimeIntellect
Jan 28, 2025
Today, we release TOPLOC: A Locality Sensitive Hashing Scheme for Verifiable Inference - Detects modifications to models, prompts, or precision - Robust across GPU types, tensor parallel configurations and attention kernels - Up to 100× faster validation than generation -
45K
Jackmin
@jackminong
May 21, 2025
people probably underestimate the importance of fault tolerance at scale. if a gpu fails once a month, a 100k gpu cluster will have a gpu fail every 26 seconds
Prime Intellect
@PrimeIntellect
May 21, 2025
Introducing PCCL, the Prime Collective Communications Library — a low-level communication library built for decentralized training over the public internet, with fault tolerance as a core design principle. In testing, PCCL achieves up to 45 Gbit/s of bandwidth across datacenters
00:00
17K
Jackmin
@jackminong
Apr 17, 2025
80% of RL training compute time is spent generating rollouts which is such an embarrassingly parallel task ripe for decentralisation🍎
Prime Intellect
@PrimeIntellect
Apr 15, 2025
Today we’re launching INTELLECT-2: The first decentralized 32B-parameter RL training run open to join for anyone with compute — fully permissionless. Scaling towards frontier reasoning across coding, math and science.
00:00
12K
Jackmin
@jackminong
Jul 15, 2025
Toploc Poster session tomorrow (Wed) at 4:30 PM East Hall E-1106 I’ll be around through Saturday; if you’re into decentralized training & inference, lets chat!
7.8K
Jackmin
@jackminong
May 12, 2025
it seems clear as day to me that decentralized RL post training will become the source of improvements in AI capabilities in the next few years
Prime Intellect
@PrimeIntellect
May 12, 2025
Releasing INTELLECT-2: We’re open-sourcing the first 32B parameter model trained via globally distributed reinforcement learning: • Detailed Technical Report • INTELLECT-2 model checkpoint primeintellect.ai/blog/intellect…
5.6K
Jackmin
@jackminong
Oct 14, 2023
Replying to @StasBekman
Remoting into GPU machines isnt that bad as most IDEs have a remote feature which makes the experience pretty much the same as local development.
18K
Jackmin
@jackminong
Feb 6, 2025
Synthetic data generation is such an embarrassingly parallel task that it seems obvious this is how the big synthetic datasets will be created in the next few years. To SYNTHETIC-1 and beyond 🫡🚀
Prime Intellect
@PrimeIntellect
Feb 6, 2025
Introducing SYNTHETIC-1: Collaboratively generating the largest synthetic dataset of verified reasoning traces for math, coding and science using DeepSeek-R1. Join us to contribute compute towards state-of-the-art open reasoning models.
00:00
5.4K
Jackmin
@jackminong
Aug 12, 2025
Replying to @soldni @jxmnop and @yoavartzi
torchtitan has an MoE impl that supports grouped mm and composes with FSDP: github.com/pytorch/torcht… needs the latest torch version though (2.8) which flash-attn doesnt have a wheel for yet :(
17K
Jackmin
@jackminong
Jul 28, 2025
we should go back to fp32. we have strayed too far from god for more “AI TOPS”
Nathan Lambert
@natolambert
Jul 27, 2025
Lol OSS being behind on "numerical correctness" is the stuff of nightmares
6.1K
Jackmin
@jackminong
Jan 29, 2025
Replying to @ZeyuanAllenZhu
I dont know how well IOI gold maps to ability to train the best ML models though. Sure, it may mean these people have a raw baseline problem solving ability. But the IOI medalists I know all went into academia working on fundamental math / CS research that they probably wont be
25K
Jackmin
@jackminong
Oct 11, 2024
In order to support dynamically on and off-boarding compute to the run, we introduced a new distributed abstraction `ElasticDeviceMesh` which manages the resizing of process groups without requiring a cold restart. More info on this, along with some other interesting innovations
Prime Intellect
@PrimeIntellect
Oct 11, 2024
Announcing INTELLECT-1: the first-ever decentralized training of a 10B model Scaling decentralized training 10x beyond prior efforts. Anyone can join us to build open-source AGI 🦋
00:00
12K
Jackmin
@jackminong
Jan 28, 2025
One of the key capabilities required for a decentralized compute system is the ability to verify that computations were performed correctly by untrusted parties. Happy to be releasing our solution to the problem. Big thanks to my colleagues who helped out with the work
Prime Intellect
@PrimeIntellect
Jan 28, 2025
Today, we release TOPLOC: A Locality Sensitive Hashing Scheme for Verifiable Inference - Detects modifications to models, prompts, or precision - Robust across GPU types, tensor parallel configurations and attention kernels - Up to 100× faster validation than generation -
5.2K
Jackmin
@jackminong
Jun 22, 2025
sometimes i forget that i made my initial calculation assuming the cows are spherical and get surprised in prod when the cows stop being nice to work with
6K
Jackmin
@jackminong
Jun 25, 2025
havent slept in days 😿 can we hire some filipinos next time?
Johannes Hagemann
@johannes_hage
Jun 25, 2025
Replying to @johannes_hage
TOPLOC verifiable inference isn't real btw, it’s just @jackminong looking at all the reasoning chains and brutally slashing misbehaving computers
2.9K