Levi Zolyomi

ML PhD at Johannes Kepler University Linz | Researcher at NXAI

[Levi Zolyomi @zlymi_levi]

Great new model from our team! Go check out the paper and try out tirex 2, weights are now available on huggingface!

[Sepp Hochreiter @HochreiterSepp]

1/5🚀 Introducing TiRex-2— our next-generation time series foundation model. Time series forecasting in the real world is streaming:
-new observations continuously arrive,
-variables interact,
-some covariates are known for the future,
-models must update predictions efficiently.

[Lilian Weng @lilianweng]

A super long overdue (3+ years?) post on scaling laws.

Compute is expensive. Scaling laws are a way to help us reason about the optimal compute allocation between data and model size before committing to a large run.

The post covers what scaling laws predict, how compute-optimal allocation works, why Kaplan et al. and Chinchilla disagree, and how data limits + fitting details make extrapolation tricky.

https://t.co/HP26eJvjHB

[DurstewitzLab @DurstewitzLab]

In a #ICML2026 position paper we argue a dynamical systems perspective is needed to drive time series models forward: https://t.co/4xESG8umP0
We need to move away from transformers that do not respect a system’s dynamical structure, esp. for out-of-domain generalization & insight

[Tri Dao @tri_dao]

As hybrid models (Qwen 3.5 / Nemotron Ultra) run agents with massive context, Gated-DeltaNet / Mamba states become a bottleneck. A simple insight to make this 2x faster: load the states, compute, but don't store them. This recompute trick finally unlocks spec decoding for SSMs

[Sepp Hochreiter @HochreiterSepp]

Comparison of sub-quadratic architectures xLSTM, Mamba-2, and Gated DeltaNet: https://t.co/ZzV9ORD8iH
Comparison of xLSTM, Mamba-2, and Gated DeltaNet on code pre-training, distillation, and time-series.

xLSTM outperforms the others due to its gating scheme and state tracking.

[Machine Learning (ML) Papers @Memoirs]

On Subquadratic Architectures: From Applications to Principles

Anamaria-Roberta Hartl, Levente Zólyomi, David Stap, Pieter-Jan Hoedt, Niklas Schmidinger, Lukas Hauzenberger, Sebastian Böck, Günter Klambauer, Sepp Hochreiter
https://t.co/81bMHixQta [𝚌𝚜.𝙻𝙶]

[Levi Zolyomi @zlymi_levi]

New paper from our lab🚀 Nonlinear RNNs are still very much important in complex data domains, as we show it in code pretraining, code model linearization and time series foundation model pretraining!

[Anamaria-Roberta Hartl @anamariarp17]

New paper "On Subquadratic Architectures: From Applications to Principles" 🙌

On commonsense & reasoning benchmarks, xLSTM, Mamba-2 & Gated DeltaNet perform nearly indistinguishably.
Therefore, we analyse them where structure genuinely matters: on code & time series.👇

[Anamaria-Roberta Hartl @anamariarp17]

New paper "On Subquadratic Architectures: From Applications to Principles" 🙌

On commonsense & reasoning benchmarks, xLSTM, Mamba-2 & Gated DeltaNet perform nearly indistinguishably.
Therefore, we analyse them where structure genuinely matters: on code & time series.👇

[Jianyang Gao @gaoj0017]

The TurboQuant paper (ICLR 2026) contains serious issues in how it describes RaBitQ, including incorrect technical claims and misleading theory/experiment comparisons.

We flagged these issues to the authors before submission. They acknowledged them, but chose not to fix them. The paper was later accepted and widely promoted by Google, reaching tens of millions of views.

We’re speaking up now because once a misleading narrative spreads, it becomes much harder to correct. We’ve written a public comment on openreview (https://t.co/nDVjmNhATM).

We would greatly appreciate your attention and help in sharing it.

[Levi Zolyomi @zlymi_levi]

@marcinuio Whoa! Vibe coding is real, making this in just a few hours, amazing possibilities!

[Linus Tech Tips @LinusTech]

NEW VIDEO! I have been betrayed - AMD $5,000 Ultimate Tech Upgrade

https://t.co/cf6SK4cfzc

Enter the AMD giveaway for TWO Steam Decks below!

https://t.co/saf7bJ73Xc

#sponsoredby @AMD

[Levi Zolyomi @zlymi_levi]

@LinusEkenstam @memdotai mem it

[Levi Zolyomi @zlymi_levi]

@FluentInFinance @memdotai mem it

[Levi Zolyomi @zlymi_levi]

@svpino @memdotai mem it

[Levi Zolyomi @zlymi_levi]

@omarsar0 @memdotai mem it

[Marques Brownlee @MKBHD]

It's like they had a meeting and thought: What can we do to break into the US market? (they didn't have to go this hard)

[Andrej Karpathy @karpathy]

The most dramatic optimization to nanoGPT so far (~25% speedup) is to simply increase vocab size from 50257 to 50304 (nearest multiple of 64). This calculates added useless dimensions but goes down a different kernel path with much higher occupancy. Careful with your Powers of 2.

[Levi Zolyomi @zlymi_levi]

@MKBHD spotify 1.0

[Levi Zolyomi @zlymi_levi]

@Jeande_d @memdotai mem it

[Levi Zolyomi @zlymi_levi]

@AlphaSignalAI @memdotai mem it