⬢github Python · 8K ★ +52 since we first saw it · pushed 1 d ago · NOASSERTION
tile-ai/tilelang
Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels
TileLang is a Pythonic domain-specific language for writing high-performance kernels (GEMM, FlashAttention, quantized matmuls) that run on GPUs, CPUs, and NPUs. Built on the TVM compiler stack, it lets developers write concise, readable code while still getting low-level, hardware-specific optimizations — now targeting CUDA, ROCm, Metal, and Huawei Ascend hardware.
Why now: Recent momentum: v0.1.13 added a multi-backend dialect and Metal support, a new LSP improves the dev experience, and a fresh Ascend 950 NPU backend shows TileLang expanding beyond NVIDIA GPUs into new accelerator ecosystems.
Who it is for: ML systems and kernel engineers who need near-optimal GPU/NPU performance without writing raw CUDA or vendor-specific assembly.
Stars over our 19 snapshots: 8K to 8K, since 4 h ago.
Where people talked about it
- ⬢github tile-ai/tilelang 10 min ago
API: https://socialmediatrends-api.osmike.com/v1/repos/tile-ai/tilelang