Baidu Releases ERNIE Evolve Automated Research System, GPU Inference Throughput Up 33.7%
Baidu has released the ERNIE Evolve automated research system for improving AI training and inference: given a research goal, it selects research directions, modifies code, and runs experiments; only improvements that pass independent evaluation and review are retained, and the results of each round guide the next round. The system achieved SOTA on NanoChat AutoResearch and NanoGPT Speedrun, and in ERNIE 5.5 Flash inference tests it improved deterministic GPU inference throughput by 33.7% and single-request decoding speed by 14.9%. Baidu calls this a step toward recursive self-improvement (RSI), and its next step is to extend the method to training systems, model capabilities, and evaluation.
Why read this
SelectedAI score 60
Baidu has disclosed the operating mechanism of its automated research system and its measured inference-side figures, which can be used to assess how far automated research has progressed in engineering optimization.
Topics
Body
Introducing ERNIE Evolve, our automated research system for improving AI training and inference. ⚙️
Given a clear research goal, it selects research directions, modifies code and runs experiments. Only improvements that pass independent evaluation and review are kept, and each round’s findings guide the next.
From public benchmarks to ERNIE 5.5 Flash inference tests, here’s what it achieved: - SOTA on NanoChat AutoResearch and NanoGPT Speedrun - 33.7% higher deterministic GPU inference throughput - 14.9% faster single-request decoding on GPUs
This is a practical step toward recursive self-improvement (RSI). Next, we plan to extend this approach to training systems, model capabilities and evaluation.