Tensormesh

LMCache is an open source Knowledge Delivery Network (KDN) designed as a caching layer for large language model serving that accelerates inference by reusing KV (key-value) caches across repeated or overlapping computations. It enables fast prompt caching, allowing LLMs to “prefill” recurring text only once and then reuse those stored KV caches, even in non-prefix positions, across multiple serving instances. This approach reduces time to first token, saves GPU cycles, and increases throughput in scenarios such as multi-round question answering or retrieval augmented generation. LMCache supports KV cache offloading (moving cache from GPU to CPU or disk), cache sharing across instances, and disaggregated prefill, which separates the prefill and decoding phases for resource efficiency. It is compatible with inference engines like vLLM and TGI and supports compressed storage, blending techniques to merge caches, and multiple backend storage options.

Luminal is a machine-learning framework built for speed, simplicity, and composability, focusing on static graphs and compiler-based optimization to deliver high performance even for complex neural networks. It compiles models into minimal “primops” (only 12 primitive operations) and then applies compiler passes to replace those with device-specific optimized kernels, enabling efficient execution on GPU or other backends. It supports modules (building blocks of networks with a standard forward API) and the GraphTensor interface (typed tensors and graphs at compile time) for model definition and execution. Luminal’s core remains intentionally small and hackable, with extensibility via external compilers for datatypes, devices, training, quantization, and more. Quick-start guidance shows how to clone the repo, build a “Hello World” example, or run a larger model like LLaMA 3 using GPU features.