Qwen3-0.6B megakernel: 527 tok/s decode on RTX 3090 (3.8x faster than PyTorch)
☆134Feb 10, 2026Updated 6 months ago
Alternatives and similar repositories for MegaQwen
Users that are interested in MegaQwen are comparing it to the libraries listed below. We may earn a commission when you buy through links labeled 'Ad' on this page.
Sorting:
- Aggressive decode optimizations for Qwen3-0.6B on RTX 5090☆59Feb 25, 2026Updated 5 months ago
- ☆22Jul 10, 2026Updated last month
- ☆20Apr 26, 2026Updated 3 months ago
- High-performance GPU kernels for LLM inference in OpenAI Triton. Fused RMSNorm, SwiGLU, INT8 GEMM with benchmarks and roofline analysis.☆40Jul 22, 2026Updated last month
- Writing FLUX in Triton☆42Sep 22, 2024Updated 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.
- GGML bindings that aim to be idiomatic Rust rather than directly corresponding to the C/C++ interface☆20Sep 25, 2023Updated 2 years ago
- ☆23Aug 1, 2026Updated 3 weeks ago
- Pure Triton kernels for Qwen3.5-27B inference on NVIDIA B200☆121Feb 28, 2026Updated 5 months ago
- LLM Inference via Triton (Flexible & Modular): Focused on Kernel Optimization using CUBIN binaries, Starting from gpt-oss Model☆119Apr 28, 2026Updated 3 months ago
- MetaAttention: A Unified and Performant Attention Framework Across Hardware Backends(PPoPP'26)☆17Aug 6, 2026Updated 2 weeks ago
- 存放一些 CUDA 编程相关的博客文件。☆22Oct 16, 2025Updated 10 months ago
- ☆203Updated this week
- This repositories contains the reference implementation for the Sparse Delta Memory paper.More precisely, it contains the model definitio…☆36Jul 9, 2026Updated last month
- Region-level profiling for CUDA kernels with trace, NVBit, CUPTI, NSys, and an interactive Explorer.☆129Apr 17, 2026Updated 4 months ago
- Wordpress hosting with auto-scaling - Free Trial Offer • AdFully Managed hosting for WordPress and WooCommerce businesses that need reliable, auto-scalable performance. Cloudways SafeUpdates now available.
- Optimize GEMM with tensorcore step by step☆40Dec 17, 2023Updated 2 years ago
- mKernel: fast multi-node, multi-GPU fused kernels☆268Updated this week
- High-performance LLM operator library built on TileLang.☆170Updated this week
- A high availability distributed filesystem built on FoundationDB and fuse.☆22Mar 13, 2023Updated 3 years ago
- Companion code for The Physics of LLM Inference book☆26Apr 21, 2026Updated 4 months ago
- MoE training for Me and You and maybe other people☆396Mar 15, 2026Updated 5 months ago
- TileFusion is an experimental C++ macro kernel template library that elevates the abstraction level in CUDA C for tile processing.☆117Aug 4, 2026Updated 2 weeks ago
- Repository for the CUDA H100 Course☆73Apr 12, 2026Updated 4 months ago
- Cuda implemenation of flash-kmeans, 2x faster☆23May 8, 2026Updated 3 months ago
- Deploy open-source AI quickly and easily - Special Bonus Offer • AdRunpod Hub is built for open source. One-click deployment and autoscaling endpoints without provisioning your own infrastructure.
- A Triton-only attention backend for vLLM☆28Jul 14, 2026Updated last month
- Fastest kernels written from scratch☆613Aug 15, 2026Updated last week
- A PyTorch implementation of the GPT-OSS-20B architecture. All components are coded from scratch: RoPE with YaRN, RMSNorm, SwiGLU with cla…☆238Dec 2, 2025Updated 8 months ago
- ☆55Dec 19, 2025Updated 8 months ago
- ☆400Aug 12, 2026Updated last week
- ☆66Updated this week
- ☆200Aug 9, 2026Updated 2 weeks ago
- Github mirror of trition-lang/triton repo.☆188Updated this week
- CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs☆247Updated this week
- Wordpress hosting with auto-scaling - Free Trial Offer • AdFully Managed hosting for WordPress and WooCommerce businesses that need reliable, auto-scalable performance. Cloudways SafeUpdates now available.
- Slow-Fast Inference: Training-free Inference Acceleration via Within-sentence Support Stability☆18Aug 13, 2026Updated last week
- A pure-Python implementation of the Nvidia CuTe layout algebra intended to be approachable and easy to learn.☆240Jun 29, 2026Updated last month
- FlashTile is a CUDA Tile IR compiler that is compatible with NVIDIA's tileiras, targeting SM70 through SM121 NVIDIA GPUs.☆61Feb 6, 2026Updated 6 months ago
- Winner 🏆 (Agent-only) MLSys 2026 - FlashInfer AI Kernel Generation Contest for the DeepSeek Sparse Attention (DSA) track with an average…☆156Updated this week
- My submission for the GPUMODE/AMD fp8 mm challenge☆29Jun 4, 2025Updated last year
- ⚡️Write HGEMM from scratch using Tensor Cores with WMMA, MMA and CuTe API, Achieve Peak⚡️ Performance.☆158May 10, 2025Updated last year
- A series of high-performance GEMM (General Matrix Multiply) implementations Iteratively optimised for H100 GPUs in Pure CUDA.☆81Feb 18, 2026Updated 6 months ago