› relativity.cc / docs
Docs
Install libtt, run JAX on a Tenstorrent card, and debug compilation and execution.
Choose a starting point
- New machine: Getting started covers the system driver, Python plugin, and a device check.
- Working JAX environment: First experiment checks a matrix multiply against NumPy, times execution, and exports StableHLO.
- Model serving: Inference gives a Qwen3-8B launch configuration and an MMLU smoke check.
- Training: Train a tiny Qwen3 model walks through TorchTitan, TorchAX, SGD updates, and checking a saved checkpoint.
- Compiler work: Software stack explains the interfaces and what to collect when a program fails.
The hardware examples use a single Blackhole p150a. The training walkthrough uses a tiny model on one device; full-model and multi-card training remain under development.
01 Getting started Install the system driver and libtt plugin, then run a computation on the TT backend. → 02 Software stack How PyTorch via torchax and JAX reach Tenstorrent through libtt, and where TorchTPU fits in the roadmap. → 03 First experiment Check a matrix multiply against NumPy, measure warm execution, and save its StableHLO. → 04 Inference Serve Qwen3-8B with SGLang-JAX and libtt on one Blackhole card. → 05 Training Train a tiny TorchTitan Qwen3 model on a p150a, save its weights, and check the checkpoint with PyTorch. →