Luce-Org/luce-megakernel

Single-kernel fusion for efficient hybrid LLMs on consumer GPUs

View on index · View in 3D Map

Related repos

• run local LLMs fast on consumer GPUs with flexible quantization
• Memory system for LLM agents
• Runs LLMs over a cluster of GPUs
• Automates GPU kernel generation with LLMs and agents
• Access multiple LLMs through a unified interface
• Runs LLMs on Ryzen NPUs without GPUs
• Harness for training LLMs on a single GPU node
• LLocalSearch: local search aggregator using LLM agents, runs without external API keys
• Run huge LLMs with 100k context on an 8GB GPU using SSD offload, no quantization
• Resources for running LLMs on local hardware