Search Results for "activation code finder"

Claude Cognitive is an advanced memory and context-management extension designed to address the stateless limitations of Claude Code by giving the model a form of persistent “working memory” and multi-instance coordination. It introduces an attention-based context router that prioritizes files and content relevant to the current development discussion — tagging them as HOT, WARM, or COLD based on recency and keyword activation — so Claude Code doesn’t waste token budget rereading irrelevant code.

...It is structured similarly to competitive programming platforms like LeetCode but focuses specifically on tensor operations and deep learning concepts. The platform provides a collection of curated problems that cover fundamental topics such as activation functions, normalization layers, attention mechanisms, and full transformer architectures. It runs in a Jupyter-based environment, allowing users to write, test, and debug their code interactively while receiving immediate feedback. An automated judging system evaluates correctness, gradient flow, and numerical stability, helping users understand both functional and theoretical aspects of their implementations.

Pika Skills is an open-source framework designed to extend the capabilities of AI coding agents by introducing modular, reusable “skills” that can be dynamically invoked during development workflows. Each skill acts as a self-contained unit composed of structured instructions, executable scripts, and dependency definitions, enabling agents to autonomously perform complex tasks without requiring manual configuration or orchestration. The system is tightly integrated with the Pika Developer...

The automated-interpretability repository implements tools and pipelines for automatically generating, simulating, and scoring explanations of neuron (or latent feature) behavior in neural networks. Instead of relying purely on manual, ad hoc interpretability probing, this repo aims to scale interpretability by using algorithmic methods that produce candidate explanations and assess their quality. It includes a “neuron explainer” component that, given a target neuron or latent feature,...

...The project implements techniques that allow model components to be dynamically moved between CPU memory and GPU memory during inference, significantly reducing the amount of GPU VRAM required to run the model. This approach takes advantage of the sparse activation properties of mixture-of-experts architectures, where only a subset of expert networks are used for each token during generation. By selectively loading and caching the required experts, the system avoids keeping the entire model in GPU memory at once. The repository includes notebooks and code examples that demonstrate how to run large language models on consumer hardware such as personal GPUs or cloud notebook environments.

ConvNeXt is a modernized convolutional neural network (CNN) architecture designed to rival Vision Transformers (ViTs) in accuracy and scalability while retaining the simplicity and efficiency of CNNs. It revisits classic ResNet-style backbones through the lens of transformer design trends—large kernel sizes, inverted bottlenecks, layer normalization, and GELU activations—to bridge the performance gap between convolutions and attention-based models. ConvNeXt’s clean, hierarchical structure...

This repository implements Class Activation Mapping (CAM), a technique to expose the implicit attention of convolutional neural networks by generating heatmaps that highlight the most discriminative image regions influencing a network’s class prediction. The method involves modifying a CNN model slightly (e.g., using global average pooling before the final layer) to produce a weighted combination of feature maps as the class activation map.