Teams at SambaNova Systems

Train and evaluate machine learning models that improve large language model inference efficiency and capability. Responsibilities include speculative decoding, quantization, sparsity, distillation, pruning, parameter-efficient fine-tuning, agentic systems, benchmarking, research implementation, and reinforcement learning or post-training. The role collaborates with performance, compiler, runtime, and product teams to ship model improvements.

Develop and optimize compiler infrastructure for SambaNova’s RDU AI accelerator. Responsibilities include implementing IR passes, lowering and partitioning machine learning models, tensor tiling, on-chip memory management, fabric mapping, and low-level debugging in C++ and Python. The role collaborates with runtime and hardware teams and includes mentorship while building production compiler software.

Build and operate the machine learning inference stack for SambaNova’s RDU accelerators. Develop production Python features including speculative and constrained decoding, tool calling, prompt caching, long-context inference, and vLLM integration. Maintain public inference APIs, create accuracy and regression tests, and troubleshoot latency, throughput, and reliability issues. The role requires strong systems and programming fundamentals, experience with LLM serving frameworks and inference techniques, and PyTorch knowledge.