Write TT hardware kernels using a Python DSL with data movement primitives.
npx skills add https://github.com/tenstorrent/tt-forge --skill tt-lang
Please help me install this Agent Skill. Skill: tt-lang Source: https://github.com/tenstorrent/tt-forge/tree/main/skills/tt-lang Command: npx skills add https://github.com/tenstorrent/tt-forge --skill tt-lang
TT-Lang solves the challenge of writing low-level TT hardware kernels by providing a Python-based DSL that exposes explicit data movement and tile-level compute primitives.
Create a tiny kernel with ttl.kernel and ttl.compute and run it on the functional simulator to verify core TT-Lang syntax.
High-intent search queries and answers about installing and using this skill.
You can write TT hardware kernels in Python using a domain-specific language that exposes explicit data movement and tile-level compute primitives, enabling fine-grained control over single or multi-core TT chips.
Yes, you can fuse workloads like matmul, softmax, and attention into efficient kernels by utilizing grid, blocks, pipes, and streaming patterns to manage data movement and compute operations.
You can test TT kernels before physical deployment by running them on a functional simulator to verify core syntax and validate explicit data movement and tile-level compute primitives.
Yes, TT kernel development supports ttnn tensor handling alongside data movement through ttl.copy and ttl.make_dataflow_buffer_like for end-to-end kernel development.
The best way to manage data movement for multi-core TT chips is by using dataflow buffers and pipes within a Python-based DSL to handle on-demand data transfers and streaming loops.
Using a Python DSL for low-level hardware kernel development provides rapid prototyping capabilities for ML kernels while still exposing explicit data movement and tile-level compute primitives for fine-grained hardware control.
tenstorrentConvert PyTorch tensors and execute tensor operations on Tenstorrent MeshDevice topologies.
tenstorrentProfile and optimize TT-Lang kernels using TT-Metal tooling.
piemonteWrite Metal GPU code for compute and render pipelines on Apple platforms.
copyleftdevBuild FPGA-based trading systems with Verilog, Python, and C++.
