MiSS is a novel PEFT method that features a low-rank structure but introduces a new update mechanism distinct from LoRA, achieving an excellent balance between performance and efficiency.

My Codex Skills

dTRPO: Trajectory Reduction in Policy Optimization of Diffusion Large Language Models

Code for “Enhancing Diffusion-Based Sampling with Molecular Collective Variables"

A Lean formalisation of Maryna Viazovska's Fields Medal-winning solution to the sphere packing problem in dimension 8 and 24.

Terminal Velocity Matching

Course website for 6.S184/6.S975: Generative AI with Stochastic Differential Equations

Code for the paper “Four Over Six: More Accurate NVFP4 Quantization with Adaptive Block Scaling”

Course on Flash-attention in Triton

Generic building-block toolbox for training neural networks with adaptive and recursive execution. It provides reusable components to control iteration, stopping, and unrolling during training, ena…

Official Implementation of pMF https://arxiv.org/abs/2601.22158

Official Implementation of "Meta Flow Maps enable scalable reward alignment"

The Student's Guide to @lintool

Simple EEG tokenizer with PyTorch datasets.

[ICLR 2026] Official code for TraceRL: Revolutionizing post-training for Diffusion LLMs, powering the SOTA TraDo series.

[ICLR 2026] Discrete Diffusion Divergence Instruct (DiDi-Instruct)

Hybrid ML + physics model of the Earth's atmosphere

High-performance CUDA kernels for equivariant graph neural networks (MACE, NequIP, Allegro). 10-20x faster than e3nn.

d3LLM: Ultra-Fast Diffusion LLM 🚀

[ICLR 2026] Improving Reasoning for Diffusion Language Models via Group Diffusion Policy Optimization

Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞

[ICLR 2026] SoFlow: Solution Flow Models for One-Step Generative Modeling

MoE training for Me and You and maybe other people