Software Engineer
Tenstorrent10 months ago
Tokyo, JapanMid Level
Responsibilities
- Develop tt-metal and implement DNN models in ttnn and tt-transformers.
- Analyze and optimize DNN model performance.
- Improve DNN model accuracy.
Requirements
- Experience launching current ML models on new hardware platforms.
- Ability to analyze performance bottlenecks, tune architectures, and improve model accuracy and robustness.
- Understanding and implementation experience with DNN optimization techniques such as quantization, sparsity, and attention optimization.
- Understanding of hardware characteristics including matrix/SIMD compute engines and memory caches.
- Ability to follow current ML research and translate it into practical improvements.
- Fluent English communication with development teams.
- Experience with NVIDIA GPUs such as H100 and A100, AMD Instinct, or Habana Gaudi is preferred.
- Experience tuning and optimizing large-scale AI workloads is preferred.
- DNN profiling and optimization experience and C++ performance-tuning experience are preferred.
- Knowledge of HPC clusters and distributed processing technologies such as MPI and NCCL is preferred.
- Experience implementing cluster-management software such as Slurm or Kubernetes is preferred.
- Open-source project participation or contribution experience is preferred.
- Applicant must be eligible to access U.S. export-controlled technology or obtain applicable license approval.
Benefits
- On-site role based in Tokyo, Japan.
Tech Stack
Categories
About Tenstorrent
Tenstorrent builds AI compute hardware and RISC‑V CPU technology for companies developing training and inference workloads in data centers and at the edge. The business sells accelerator cards and systems with a full software stack, and licenses RISC‑V CPU IP. Founded in 2016 and headquartered in Toronto, it is privately held and led by semiconductor architect Jim Keller.