Official code for paper: [CLS] Attention is All You Need for Training-Free Visual Token Pruning: Make VLM Inference Faster.

[NeurIPS 2025] Official code for paper: Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs.

[TCSVT] Official repository of the paper "A Glimpse to Compress: Dynamic Visual Token Pruning for Large Vision-Language Models"

This is a collection of our research on efficient AI, covering hardware-aware NAS and model compression.

[ICCV 2025] Official code for paper: Beyond Text-Visual Attention: Exploiting Visual Cues for Effective Token Pruning in VLMs

A tutorial of model quantization using TensorFlow

DWARF v2 is a research prototype that expands on the original DWARF.

O(N) attention with a bounded inference KV cache. D4 Daubechies wavelet field + content-gated Q·K gather at dyadic offsets.

Substrate-agnostic, structure-complete memory fabric — read the whole book from any address; forward-compatible across classical, photonic, and quantum hardware (rare-earth quantum memory). Reproducible O(N^2)->O(N) LLM efficiency.

Reproducibility materials for causal whitespace patching in Korean byte-latent LMs: a 2.5–2.9% matched-quality latency result without scale amplification.

Official implementation of recall-controlled early abort for LLM agent episodes