Spec-driven GPU operator library for LLMs — designed for AI agents to build, evaluate, and optimize
Built on TileLang
Status: TileOPs is under active development. APIs may change.
TileOPs is a GPU operator library for LLM training and inference, built on TileLang. Beyond providing a growing collection of production-quality operators, TileOPs explores a spec-driven development model where AI agents can read declarative operator specifications, generate kernel implementations, and evaluate them against hardware-theoretical performance bounds — with minimal human scaffolding.
Every operator is split into two layers with a strict boundary:
- Op (L2) — stateless Python entry point. Handles validation, dtype casting, and memory layout. CUDA-Graph compatible;
torch.compile(fullgraph=True)support is declared per op in the manifest. - Kernel (L1) — TileLang GPU implementation with hardware-specific optimizations (Hopper).
This separation keeps user-facing behavior independent of GPU strategy, allowing agents and developers to modify either layer without side effects on the other.
- Spec-driven — each operator is declared in a machine-readable manifest (
tileops/manifest/) that specifies signatures, workloads, and roofline formulas, serving as the entry point for both agent code generation and automated validation - Roofline-evaluated — kernel performance is measured against Speed-of-Light hardware bounds, not relative baselines
- Auto-tuning — built-in search over tile sizes, pipelines, and scheduling parameters
- Lightweight — depends only on TileLang, PyTorch, and einops
TileOPs is under active development and is installed from source; PyPI releases will begin with the first stable release. A CUDA-capable GPU is required.
- Python >= 3.10
- PyTorch >= 2.1, < 2.11 (CI validates 2.10)
- CUDA Toolkit 12.x
- NVIDIA GPU: Hopper (SM_90)
- TileLang >= 0.1.9, < 0.2.0 (CI validates 0.1.11)
git clone https://github.com/tile-ai/TileOPs
cd TileOPs
make install # dev dependencies + pre-commit hooksNote
If CUDA and TileLang are already installed system-wide and you encounter build issues:
PIP_NO_BUILD_ISOLATION=1 pip install -e '.[dev]' -v && pre-commit install
Verify:
python -m pytest tests/ -q # requires a CUDA GPUimport torch
from tileops.ops import GemmOp
M, N, K = 1024, 1024, 512
dtype = torch.float16
gemm = GemmOp() # shapes and dtype are inferred at call time
a = torch.randn(M, K, device="cuda", dtype=dtype)
b = torch.randn(N, K, device="cuda", dtype=dtype) # trans_b=True by default
d = gemm(a, b) # equals a @ b.TDesign docs and development guides are in docs/. The full API reference and performance tables are published at TileOPs.github.io.
See docs/ for design docs. Branch and commit conventions are in .claude/conventions/types.sh.
TileOPs is released under the MIT License.