Ulysses sequence-parallel all-to-all as a torch custom op, moved by the GPU copy engines into torch symmetric memory. Zero SM usage; 1.66-2.17x over torch.distributed on NVLink.
☆54Aug 25, 2026Updated 2 months ago
Alternatives and similar repositories for fast-ulysses
Users that are interested in fast-ulysses are comparing it to the libraries listed below. We may earn a commission when you buy through links labeled 'Ad' on this page.
Sorting:
- High performance RMSNorm Implement by using SM Core Storage(Registers and Shared Memory)☆31Jan 22, 2026Updated 9 months ago
- Efficient and unified implementations for TopK-based sparse attention☆41Jul 10, 2026Updated 3 months ago
- Autonomous GPU kernel optimization system driven by AI agents.☆32Mar 29, 2026Updated 7 months ago
- ☆11May 21, 2026Updated 5 months ago
- A cross-node expert-parallel MoE training library for computation-communication overlap.☆16Updated this week
- Managed Database hosting by DigitalOcean • AdPostgreSQL, MySQL, MongoDB, Kafka, Valkey, and OpenSearch available. Automatically scale up storage and focus on building your apps.
- FA4-based Relative Attention Kernel developed by TML and Colfax☆18Sep 11, 2026Updated last month
- ☆18Jun 19, 2026Updated 4 months ago
- Muon in Int8 Precision Made Possible☆20Updated this week
- Breakable CUDA graph capture for PyTorch - capture compatible regions as CUDA graphs while running incompatible operations eagerly betwee…☆30Oct 8, 2026Updated 3 weeks ago
- CUDA Graph runtime tensor debugging utilities for PyTorch☆18Jul 12, 2026Updated 3 months ago
- ☆20Aug 9, 2026Updated 2 months ago
- Skills for writing tilelang and debugging with CUDA toolkits.☆149Updated this week
- NVSHMEM‑Tutorial: Build a DeepEP‑like GPU Buffer☆208Feb 11, 2026Updated 8 months ago
- High-performance Accelerated LLM Operators☆24Updated this week
- Deploy on Railway without the complexity - Free Credits Offer • AdConnect your repo and Railway handles the rest with instant previews. Quickly provision container image services, databases, and storage volumes.
- ☆54May 19, 2025Updated last year
- Production-ready MoE load balancing via real-time expert replication☆263Oct 15, 2026Updated 3 weeks ago
- ☆15Jul 10, 2026Updated 3 months ago
- ☆29Apr 5, 2026Updated 7 months ago
- Understand your LLM before deploy☆36Updated this week
- Qwen-Image's DiT inference with TensorRT-10☆21Oct 13, 2025Updated last year
- NVIDIA cuTile learn☆168Dec 9, 2025Updated 10 months ago
- NVFP4 Flash-Attention 4 on BlackWell☆69Sep 26, 2026Updated last month
- A practical way of learning Swizzle☆45Feb 3, 2025Updated last year
- Managed Database hosting by DigitalOcean • AdPostgreSQL, MySQL, MongoDB, Kafka, Valkey, and OpenSearch available. Automatically scale up storage and focus on building your apps.
- ☆37Aug 7, 2025Updated last year
- ☆32Jul 2, 2025Updated last year
- a size profiler for cuda binary☆74Oct 3, 2026Updated last month
- FLA but cuTile☆27Apr 17, 2026Updated 6 months ago
- NCU-driven iterative optimization workflow for CUDA/CUTLASS/Triton/CuTe DSL kernels.☆25Apr 10, 2026Updated 6 months ago
- A Distributed Attention Towards Linear Scalability for Ultra-Long Context, Heterogeneous Data Training☆971Updated this week
- ☆275Oct 14, 2026Updated 3 weeks ago
- Easy recipes to speed up latency of Flux, QwenImage, and LTX-2 with NVFP4 and MXFP8 on Blackwell.☆14Apr 10, 2026Updated 6 months ago
- An agent harness that compiles a model into one provably-correct, self-retargeting CUDA megakernel and self-tunes it past cuBLAS at batch…☆162Sep 18, 2026Updated last month
- Deploy on Railway without the complexity - Free Credits Offer • AdConnect your repo and Railway handles the rest with instant previews. Quickly provision container image services, databases, and storage volumes.
- LLM Inference via Triton (Flexible & Modular): Focused on Kernel Optimization using CUBIN binaries, Starting from gpt-oss Model☆119Apr 28, 2026Updated 6 months ago
- Expert Specialization MoE Solution based on CUTLASS☆27Apr 14, 2026Updated 6 months ago
- An experimental communicating attention kernel based on DeepEP.☆34Jul 29, 2025Updated last year
- High-performance LLM operator library built on TileLang.☆202Updated this week
- ☆98Feb 5, 2026Updated 9 months ago
- Automated CUDA kernel performance diagnostics from NVIDIA Nsight Compute (NCU) CSV exports.☆36Mar 18, 2026Updated 7 months ago
- Official repository Flash Local Linear Attention☆40May 28, 2026Updated 5 months ago