FlexFusion / FlexFusionLinks
The official implementation for the intra-stage fusion technique introduced in https://arxiv.org/abs/2409.13221
☆27Updated 8 months ago
Alternatives and similar repositories for FlexFusion
Users that are interested in FlexFusion are comparing it to the libraries listed below
Sorting:
- Accelerating Large-Scale Reasoning Model Inference with Sparse Self-Speculative Decoding☆74Updated last month
- NVSHMEM‑Tutorial: Build a DeepEP‑like GPU Buffer☆151Updated 3 months ago
- Nex Venus Communication Library☆70Updated last month
- ☆81Updated 2 months ago
- [NeurIPS 2024] Efficient LLM Scheduling by Learning to Rank☆66Updated last year
- ☆70Updated 2 months ago
- ☆39Updated 2 months ago
- [HPCA 2026] A GPU-optimized system for efficient long-context LLMs decoding with low-bit KV cache.☆76Updated 3 weeks ago
- [DAC'25] Official implement of "HybriMoE: Hybrid CPU-GPU Scheduling and Cache Management for Efficient MoE Inference"☆95Updated 3 weeks ago
- Scalable long-context LLM decoding that leverages sparsity—by treating the KV cache as a vector storage system.☆110Updated last week
- Artifact for "Marconi: Prefix Caching for the Era of Hybrid LLMs" [MLSys '25 Outstanding Paper Award, Honorable Mention]☆48Updated 10 months ago
- ☆65Updated 8 months ago
- ☆27Updated 9 months ago
- [NeurIPS 2025] ClusterFusion: Expanding Operator Fusion Scope for LLM Inference via Cluster-Level Collective Primitive☆60Updated 3 weeks ago
- A lightweight design for computation-communication overlap.☆207Updated 2 weeks ago
- ☆59Updated last year
- gLLM: Global Balanced Pipeline Parallelism System for Distributed LLM Serving with Token Throttling☆52Updated this week
- ☆49Updated 8 months ago
- Stateful LLM Serving☆93Updated 9 months ago
- Multi-Level Triton Runner supporting Python, IR, PTX, and cubin.☆80Updated 2 weeks ago
- High performance Transformer implementation in C++.☆148Updated 11 months ago
- ☆130Updated last year
- A framework for generating realistic LLM serving workloads☆94Updated 3 months ago
- DeeperGEMM: crazy optimized version☆74Updated 8 months ago
- NEO is a LLM inference engine built to save the GPU memory crisis by CPU offloading☆77Updated 6 months ago
- An experimental communicating attention kernel based on DeepEP.☆35Updated 5 months ago
- Building the Virtuous Cycle for AI-driven LLM Systems☆112Updated this week
- DeepSeek-V3.2-Exp DSA Warmup Lightning Indexer training operator based on tilelang☆39Updated last month
- Tile-based language built for AI computation across all scales☆115Updated 2 weeks ago
- DLSlime: Flexible & Efficient Heterogeneous Transfer Toolkit☆91Updated this week