Artem

kabachuha

·

https://scholar.google.com/citations?user=_kUfYFUAAAAJ

kabachuha

AI & ML interests

deep learning, natural language processing, text2image, text2video, computer vision

Recent Activity

new activity 4 days ago

Lightricks/LTX-2.5:Any chance for a bigger text encoder (e.g. Gemma 4 26b) for the next version?

new activity 6 days ago

kandinskylab/KVAE-Audio:Comparison with LTX-2.3 Audio VAE

new activity 7 days ago

Lightricks/LTX-2.5:Atrocious model - object permanence/consistency is nonexistent

View all activity

Organizations

upvoted a collection 12 days ago

LTX-2.5

LTX-2.5 base models, quantized models and accompanying LoRAs and IC-LoRAs • 4 items • Updated 13 days ago • 46

upvoted a paper 17 days ago

KVAE: Family of Tokenizers for Multimodal Generative Models

Paper • 2608.05798 • Published 18 days ago • 29

upvoted a collection 20 days ago

Kandinsky WM 1.0

Image-to-Video models for Physical AI: autonomous driving, robotics, general physics. • 3 items • Updated 20 days ago • 4

upvoted a collection 21 days ago

Laguna S 2.1

Our most capable model to date, designed for long-horizon work. • 13 items • Updated 21 days ago • 45

upvoted 2 papers about 2 months ago

PointDiT: Pixel-Space Diffusion for Monocular Geometry Estimation

Paper • 2607.02515 • Published Jul 2 • 20

Gemma 4 Technical Report

Paper • 2607.02770 • Published Jul 2 • 77

upvoted a paper 2 months ago

Interpreting and Steering a Text-to-Speech Language Model with Sparse Autoencoders

Paper • 2606.10029 • Published Jun 8 • 12

upvoted 2 papers 3 months ago

A Geometric Account of Activation Steering through Angle-Norm Decomposition

Paper • 2606.06735 • Published Jun 4 • 27

Whisper Hallucination Detection and Mitigation via Hidden Representation Steering and Sparse AutoEncoders

Paper • 2606.07473 • Published Jun 5 • 15

upvoted an article 3 months ago

Article

Welcome NVIDIA Cosmos 3: The First Open Omni-model for Physical AI Reasoning and Action

nvidia

•

Jun 1

• 88

upvoted a collection 4 months ago

KVAE 2.0

KVAE 2.0 is a family of image and video tokenizers with a time compression ratio of 4 and spacial compression ratio of 8 and 16 • 3 items • Updated 18 days ago • 5

upvoted a paper 5 months ago

Interpreting CLIP with Hierarchical Sparse Autoencoders

Paper • 2502.20578 • Published Feb 27, 2025 • 1

upvoted a collection 5 months ago

My LTX-2 Loras

9 items • Updated Apr 3 • 7

upvoted 3 papers 6 months ago

SOM Directions are Better than One: Multi-Directional Refusal Suppression in Language Models

Paper • 2511.08379 • Published Nov 11, 2025 • 5

Effective Reasoning Chains Reduce Intrinsic Dimensionality

Paper • 2602.09276 • Published Feb 9 • 12

LLaDA2.1: Speeding Up Text Diffusion via Token Editing

Paper • 2602.08676 • Published Feb 9 • 71

upvoted 3 papers 7 months ago

AudioSAE: Towards Understanding of Audio-Processing Models with Sparse AutoEncoders

Paper • 2602.05027 • Published Feb 4 • 63

GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization

Paper • 2601.05242 • Published Jan 8 • 234

Cross-Frame Representation Alignment for Fine-Tuning Video Diffusion Models

Paper • 2506.09229 • Published Jun 10, 2025 • 7

upvoted a paper 8 months ago

LTX-2: Efficient Joint Audio-Visual Foundation Model

Paper • 2601.03233 • Published Jan 6 • 189