🔍 Visualize attention patterns in transformer models to better understand how LLMs process text inputs with interactive heatmaps and comparisons.

💥 Optimize linear attention models with efficient Triton-based implementations in PyTorch, compatible across NVIDIA, AMD, and Intel platforms.

🐙 Implements Flash Attention with sink for gpt-oss-20b; includes test.py. WIP backward pass, varlen support, and community sync to return softmax_lse only.

Advanced sparse modern Hopfield models delivering fast associative memory with energy-efficient inference.

Keras beit,caformer,CMT,CoAtNet,convnext,davit,dino,efficientdet,edgenext,efficientformer,efficientnet,eva,fasternet,fastervit,fastvit,flexivit,gcvit,ghostnet,gpvit,hornet,hiera,iformer,inceptionnext,lcnet,levit,maxvit,mobilevit,moganet,nat,nfnets,pvt,swin,tinynet,tinyvit,uniformer,volo,vanillanet,yolor,yolov7,yolov8,yolox,gpt2,llama2, alias kecam

Implementation of Danijar's latest iteration for his Dreamer line of work

🧑‍🏫 60+ Implementations/tutorials of deep learning papers with side-by-side notes 📝; including transformers (original, xl, switch, feedback, vit, ...), optimizers (adam, adabelief, sophia, ...), gans(cyclegan, stylegan2, ...), 🎮 reinforcement learning (ppo, dqn), capsnet, distillation, ... 🧠

This repository contains the code to reproduce the experiments performed in the Dynamical Mean-Field Theory of Self-Attention Neural Networks article.

Train a GPT from scratch on your laptop

a JAX + NNX implementation of Google's Titan NN. Transformer short-term attention + neural long-term memory with test-time updates

A code deep-dive on one of the key innovations from Deepseek - Multihead Latent Attention (MLA)

A complete implementation of the "Attention Is All You Need" Transformer model from scratch using PyTorch. This project focuses on building and training a Transformer for neural machine translation (English-to-Italian) on the OpusBooks dataset.

Scenic: A Jax Library for Computer Vision Research and Beyond

Linear-time sequence modeling that replaces attention's O(n²d) complexity with O(nd) summation-based aggregation. Demonstrates constraint-driven emergence: how functional representations can develop from optimization pressure and architectural constraints alone, without explicit pairwise interactions.

sd-webui-forge-classic/tree/neo working missing extentions from Forge

Vision Transformers for image classification, image segmentation, and object detection.

A simple and minimal open source implementation of "Introducing LFM2: The Fastest On-Device Foundation Models on the Market" from Liquid AI in Pytorch

Implementation of the ScreenAI model from the paper: "A Vision-Language Model for UI and Infographics Understanding"

We introduce VLM-Mamba, the first Vision-Language Model built entirely on State Space Models (SSMs), specifically leveraging the Mamba architecture.

Helps preserving your attention when watching long videos by skipping hummings, "hum"s. Using the model HUMAwareVAD2025