ovg-project / kvcachedLinks
kvcached: Elastic KV cache for dynamic GPU sharing and efficient multi-LLM inference.
☆86Updated this week
Alternatives and similar repositories for kvcached
Users that are interested in kvcached are comparing it to the libraries listed below
Sorting:
- ☆69Updated last year
- Stateful LLM Serving☆81Updated 6 months ago
- A lightweight design for computation-communication overlap.☆165Updated this week
- High performance Transformer implementation in C++.☆132Updated 7 months ago
- FlexFlow Serve: Low-Latency, High-Performance LLM Serving☆56Updated last week
- [OSDI'24] Serving LLM-based Applications Efficiently with Semantic Variable☆181Updated 11 months ago
- SpotServe: Serving Generative Large Language Models on Preemptible Instances☆128Updated last year
- NVSHMEM‑Tutorial: Build a DeepEP‑like GPU Buffer☆60Updated this week
- Artifact of OSDI '24 paper, ”Llumnix: Dynamic Scheduling for Large Language Model Serving“☆62Updated last year
- ☆120Updated 10 months ago
- DeeperGEMM: crazy optimized version☆70Updated 4 months ago
- ☆84Updated 5 months ago
- gLLM: Global Balanced Pipeline Parallelism System for Distributed LLM Serving with Token Throttling☆39Updated this week
- A framework for generating realistic LLM serving workloads☆55Updated 3 months ago
- A prefill & decode disaggregated LLM serving framework with shared GPU memory and fine-grained compute isolation.☆108Updated 3 months ago
- An interference-aware scheduler for fine-grained GPU sharing☆145Updated 7 months ago
- Artifact for "Marconi: Prefix Caching for the Era of Hybrid LLMs" [MLSys '25 Outstanding Paper Award, Honorable Mention]☆21Updated 6 months ago
- [NeurIPS 2024] Efficient LLM Scheduling by Learning to Rank☆59Updated 10 months ago
- ☆63Updated 4 months ago
- Official repository for the paper DynaPipe: Optimizing Multi-task Training through Dynamic Pipelines☆20Updated last year
- ☆126Updated 11 months ago
- A GPU-optimized system for efficient long-context LLMs decoding with low-bit KV cache.☆58Updated last week
- Debug print operator for cudagraph debugging☆13Updated last year
- nnScaler: Compiling DNN models for Parallel Training☆118Updated last week
- A resilient distributed training framework☆94Updated last year
- Research prototype of PRISM — a cost-efficient multi-LLM serving system with flexible time- and space-based GPU sharing.☆24Updated 3 weeks ago
- ☆76Updated 3 years ago
- ☆290Updated last week
- PyTorch library for cost-effective, fast and easy serving of MoE models.☆231Updated 2 months ago
- Automated Parallelization System and Infrastructure for Multiple Ecosystems☆79Updated 9 months ago