Design and implement novel algorithms that optimize the inference stack for LLMs to ensure the highest througput and lowest latency for billions of users including sampling techniques, speculative decoding advancements, model compression, knowledge distillation and quantization strategies. Work on development and optimization of the low-level software that runs AI models efficiently. Implement and optimize custom kernels for inference, focusing on improving latency, throughput, and memory usage, across different hardware and model architectures.
Co-design algorithms that exploit the unique architectectural advantages of custom hardware and accelerators. Collaborate with hardware architects on next-generation systems and use a deep understanding of both model architecture and hardware to inform algorithm selection. Amplify impact and influence the research ecosystem through publications and scientific dissemination.
Job description provided by Google.
View the original posting
