Inside NVIDIA Groq 3 LPX: The Low-Latency Inference Accelerator for the NVIDIA Vera Rubin Platform

Develop an abstraction layer that benchmarks various inference backends (Groq vs TensorRT-LLM) using a unified API. This helps developers swap hardware targets without rewriting inference pipelines.

Suggested repo: bench-serve

"Measure, compare, and switch between inference engines in seconds."

Estimated effort: 60h