ScalingIntelligence / hydragen
Hydragen: High-Throughput LLM Inference with Shared Prefixes
☆22Updated 6 months ago
Related projects ⓘ
Alternatives and complementary repositories for hydragen
- Code for Palu: Compressing KV-Cache with Low-Rank Projection☆56Updated this week
- Breaking Throughput-Latency Trade-off for Long Sequences with Speculative Decoding☆70Updated last week
- GEAR: An Efficient KV Cache Compression Recipefor Near-Lossless Generative Inference of LLM☆146Updated 4 months ago
- Boosting 4-bit inference kernels with 2:4 Sparsity☆51Updated 2 months ago
- The source code of our work "Prepacking: A Simple Method for Fast Prefilling and Increased Throughput in Large Language Models"☆56Updated last month
- ☆55Updated 5 months ago
- Ouroboros: Speculative Decoding with Large Model Enhanced Drafting (EMNLP 2024 main)☆76Updated last month
- CUDA and Triton implementations of Flash Attention with SoftmaxN.☆66Updated 5 months ago
- Implementation of Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting☆41Updated 4 months ago
- LLM Serving Performance Evaluation Harness☆55Updated 2 months ago
- NAACL '24 (Best Demo Paper RunnerUp) / MlSys @ NeurIPS '23 - RedCoast: A Lightweight Tool to Automate Distributed Training and Inference☆60Updated 3 weeks ago
- Transformers components but in Triton☆25Updated this week
- Repository for Sparse Finetuning of LLMs via modified version of the MosaicML llmfoundry☆38Updated 9 months ago
- [ACL 2024] RelayAttention for Efficient Large Language Model Serving with Long System Prompts☆34Updated 8 months ago
- ☆36Updated 4 months ago
- PyTorch bindings for CUTLASS grouped GEMM.☆53Updated last week
- Triton-based implementation of Sparse Mixture of Experts.☆184Updated last month
- Odysseus: Playground of LLM Sequence Parallelism☆55Updated 4 months ago
- Cataloging released Triton kernels.☆133Updated 2 months ago
- Materials for learning SGLang☆86Updated this week
- ring-attention experiments☆96Updated 3 weeks ago
- A safetensors extension to efficiently store sparse quantized tensors on disk☆46Updated this week
- ☆95Updated last month
- Fast Inference of MoE Models with CPU-GPU Orchestration☆170Updated 2 weeks ago
- extensible collectives library in triton☆65Updated last month
- 16-fold memory access reduction with nearly no loss☆57Updated this week
- FlexAttention w/ FlashAttention3 Support☆27Updated last month
- TidalDecode: A Fast and Accurate LLM Decoding with Position Persistent Sparse Attention☆21Updated last month
- The official repo for "LLoCo: Learning Long Contexts Offline"☆110Updated 4 months ago
- PyTorch library for cost-effective, fast and easy serving of MoE models.☆101Updated 2 months ago