Run LLMs, Fast.
Run open models on your own GPUs. Parallel decoding and reusable prompts for Apple Silicon and supported NVIDIA hardware.
curl -fsSL https://tensorfold.dev/install.sh | sh 
Inference,
measured.
Real runs on named hardware.
Open the release. Inspect the conditions.
Nemotron 3.5 Lightning
RTX PRO 6000 Blackwell
Greedy code decode with drafting enabled, an MLX affine 4-bit checkpoint and the CUDA engine.
Separate workloads and machines, not a like-for-like comparison.
Community benchmarksOne stream.
More work per pass.
Drafted tokens, retained prompts and connected machines, working through the same engine.
Check the work together.
On supported models, a drafter proposes several tokens and the target checks them together, committing only the matching prefix.
Open models.
Your infrastructure.
Run qualified checkpoints on Apple Silicon or NVIDIA hardware, through APIs compatible with the tools your teams already use.
Discuss a deployment
Part of NVIDIA's startup program.
With thanks to Mia and her team for their patches, setup and testing.
Explore the collaboration ↗
Preparing
for orbit.
TensorFold is being prepared for a satellite deployment, where every hardware resource counts.
Concept illustration. Hardware results will follow qualification.