feifeibear / Odysseus-Transformer
Odysseus: Playground of LLM Sequence Parallelism
☆57Updated 5 months ago
Related projects ⓘ
Alternatives and complementary repositories for Odysseus-Transformer
- PyTorch bindings for CUTLASS grouped GEMM.☆53Updated 3 weeks ago
- Boosting 4-bit inference kernels with 2:4 Sparsity☆51Updated 2 months ago
- [ACL 2024] RelayAttention for Efficient Large Language Model Serving with Long System Prompts☆34Updated 8 months ago
- GPTQ inference TVM kernel☆36Updated 6 months ago
- ☆88Updated 2 months ago
- Quantized Attention on GPU☆30Updated 2 weeks ago
- Summary of system papers/frameworks/codes/tools on training or serving large model☆56Updated 11 months ago
- A sparse attention kernel supporting mix sparse patterns☆55Updated last month
- A Suite for Parallel Inference of Diffusion Transformers (DiTs) on multi-GPU Clusters☆32Updated 3 months ago
- ☆46Updated 2 months ago
- ☆64Updated 3 months ago
- GEAR: An Efficient KV Cache Compression Recipefor Near-Lossless Generative Inference of LLM☆147Updated 4 months ago
- ☆79Updated 2 months ago
- [ICML 2024] Quest: Query-Aware Sparsity for Efficient Long-Context LLM Inference☆202Updated 2 weeks ago
- ☆42Updated 6 months ago
- ☆35Updated 2 weeks ago
- Ouroboros: Speculative Decoding with Large Model Enhanced Drafting (EMNLP 2024 main)☆76Updated last month
- Code for Palu: Compressing KV-Cache with Low-Rank Projection☆57Updated this week
- PyTorch bindings for CUTLASS grouped GEMM.☆68Updated 4 months ago
- ☆96Updated last month
- Breaking Throughput-Latency Trade-off for Long Sequences with Speculative Decoding☆78Updated this week
- Official PyTorch implementation of FlatQuant: Flatness Matters for LLM Quantization☆63Updated last week
- Implementation of Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting☆44Updated 4 months ago
- Official repository for LightSeq: Sequence Level Parallelism for Distributed Training of Long Context Transformers☆195Updated 3 months ago
- Official implementation of ICML 2024 paper "ExCP: Extreme LLM Checkpoint Compression via Weight-Momentum Joint Shrinking".☆41Updated 4 months ago
- 16-fold memory access reduction with nearly no loss☆59Updated last week
- Triton implementation of Flash Attention2.0☆22Updated last year
- High-speed GEMV kernels, at most 2.7x speedup compared to pytorch baseline.☆90Updated 4 months ago
- Standalone Flash Attention v2 kernel without libtorch dependency☆98Updated 2 months ago