Pritam
@Pritamstudyai
Followers
255
Following
12K
Media
97
Statuses
1K
add turbodiffusion
what a week for flopheads 🚀: - BLASST: 1.62x prefill & 1.48x decode speed that folds naturally into FA - SonicMoE: 1.86x faster MoE - Nemotron 3: Mamba2 SSM hybrid transformers going almost 2x on tok/sec over non-hybrid. - Nanomoe: RDEP & nvfp4 on blackwell
1
0
1
Another resource stacked up
0
0
1
If you want to do ZK in 2026, here are the courses I'd take: 1 - A linear algebra course. This is the foundation of almost all non-trivial fields of programming. 2 - A discrete math course (especially one that includes elementary number theory) 3 - A proofs course (as a
How do you actually audit ZK circuits? At DSS 101, @Jeyffre, founder of @rareskills_io, breaks down the core building blocks of ZK auditing: ✅ What ZK proofs really verify ✅ How constraints encode real-world logic ✅ Common pitfalls (missing constraints, compilers, memory
28
78
532
today we’re open-sourcing nmoe: https://t.co/iq6HliUqpq i started this because training deepseek-shaped ultra-sparse moes should be straightforward at research scale, but in practice it’s painful: - expert flops get stranded (router shatters your batch → tiny per-expert
github.com
MoE training for Me and You and maybe other people - GitHub - Noumena-Network/nmoe: MoE training for Me and You and maybe other people
24
71
584
if you wanna learn inference optimisation:
How long have you been "planning to understand" how modern LLM inference works? We just gave you a readable version of SGLang you can finish over the weekend. Introducing mini-SGLang ⚡ We distilled SGLang from 300K into 5,000 lines. Kept the core design, cut the complexity.
0
0
3
they did it🥳, congrats team
TurboDiffusion: 100–205× faster video generation on a single RTX 5090 🚀 Only takes 1.8s to generate a high-quality 5-second video. The key to both high speed and high quality? 😍SageAttention + Sparse-Linear Attention (SLA) + rCM Github: https://t.co/ybbNBjgHFP Technical
0
0
3
What if you could speed up a cutting-edge AI image generator without retraining it from scratch? Apple presents a new method that unifies two major AI image models (diffusion & flow matching) under a simple statistical rule. This lets them directly apply a powerful
3
24
168
My team is looking for interns! Please reach out with your resume if you want to contribute to the state of art deep learning frameworks and compilers, powering frontier models on pre/post training and inference!
58
41
621
so if that's the case one research can alone train and release sota models
I estimate that each frontier lab could have more research compute than all academic institutions in the US combined. Even if it’s not quite true yet, it’s telling.
0
0
0
This is the DeepSeek moment for Voice AI. Today we’re releasing Chatterbox Turbo — our state-of-the-art MIT licensed voice model that beats ElevenLabs Turbo and Cartesia Sonic 3! We’re finally removing the trade-offs that have held voice AI back. Fast models sound robotic.
161
381
4K
Microsoft just dropped VITRA-VLA, a new Vision-Language-Action model for robotics on Hugging Face. It learns dexterous manipulation from over 1 million real-life human hand activity videos.
4
52
304
Insecurity is profit. Human insecurity is the prime psychological driver of consumerism and a systemic driver of modern capitalism. [in current economics and attention economics]
0
0
1
Are my thoughts mine? or are they simply influenced by everyone around me?
1
0
2
this is really cool way to learn about LLMs and Rust.
🦀 Building an LLM From Scratch in Rust Want to really understand how LLMs work under the hood? Start at the foundation: a complete GPT-2 style transformer. https://t.co/qpsi9QN6RV
#rust #rustlang
0
0
6
We’re opening applications for the next two rounds of the Anthropic Fellows Program, beginning in May and July 2026. We provide funding, compute, and direct mentorship to researchers and engineers to work on real safety and security projects for four months.
90
283
3K
#KostasKeynoteLessons: Curious about the "Keynote magic" behind my slides? I’m releasing the full Keynote source file for my recent Gaussian Splatting lecture, all 10 GIGAbytes of it! Grab the files in the thread and feel free to remix.
16
43
387