ScalingIntelligence / hydragenLinks
Hydragen: High-Throughput LLM Inference with Shared Prefixes
☆41Updated last year
Alternatives and similar repositories for hydragen
Users that are interested in hydragen are comparing it to the libraries listed below
Sorting:
- [ICLR2025] Breaking Throughput-Latency Trade-off for Long Sequences with Speculative Decoding☆125Updated 8 months ago
- GEAR: An Efficient KV Cache Compression Recipefor Near-Lossless Generative Inference of LLM☆165Updated last year
- ☆92Updated 3 months ago
- Boosting 4-bit inference kernels with 2:4 Sparsity☆80Updated 11 months ago
- A minimal implementation of vllm.☆51Updated last year
- ☆50Updated 3 months ago
- ☆92Updated 5 months ago
- Framework to reduce autotune overhead to zero for well known deployments.☆81Updated last week
- Code for paper: [ICLR2025 Oral] FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inference☆136Updated 3 months ago
- ☆63Updated 4 months ago
- 16-fold memory access reduction with nearly no loss☆104Updated 5 months ago
- PyTorch bindings for CUTLASS grouped GEMM.☆110Updated 3 months ago
- ☆88Updated 9 months ago
- ☆74Updated 5 months ago
- Accelerate LLM preference tuning via prefix sharing with a single line of code☆43Updated last month
- DeeperGEMM: crazy optimized version☆70Updated 3 months ago
- A Suite for Parallel Inference of Diffusion Transformers (DiTs) on multi-GPU Clusters☆47Updated last year
- Triton-based Symmetric Memory operators and examples☆23Updated this week
- Example of applying CUDA graphs to LLaMA-v2☆12Updated 2 years ago
- extensible collectives library in triton☆88Updated 4 months ago
- [ICLR'25] Fast Inference of MoE Models with CPU-GPU Orchestration☆226Updated 9 months ago
- Odysseus: Playground of LLM Sequence Parallelism☆76Updated last year
- ☆78Updated 4 months ago
- TileFusion is an experimental C++ macro kernel template library that elevates the abstraction level in CUDA C for tile processing.☆95Updated 2 months ago
- Quantized Attention on GPU☆44Updated 9 months ago
- [ICLR 2025] TidalDecode: A Fast and Accurate LLM Decoding with Position Persistent Sparse Attention☆47Updated 3 weeks ago
- An experimental communicating attention kernel based on DeepEP.☆34Updated last month
- ☆140Updated 6 months ago
- [NeurIPS 2024] The official implementation of "Kangaroo: Lossless Self-Speculative Decoding for Accelerating LLMs via Double Early Exitin…☆59Updated last year
- [NeurIPS 2024] Efficient LLM Scheduling by Learning to Rank☆58Updated 9 months ago