Baidu ERNIE Discloses RSI Progress for the First Time; ERNIE Evolve Has Been Used in ERNIE 5.5 Flash R&D
Baidu ERNIE discloses RSI progress for the first time, unveiling the automated research system ERNIE Evolve that participated in the R&D of ERNIE 5.5 Flash. The system can autonomously propose research directions, conduct experiments, verify results, and accumulate experience. On the three evaluation metrics across the two tasks NanoChat AutoResearch and NanoGPT Speedrun, ERNIE Evolve achieved the current publicly verified SOTA performance, and has already been used for inference optimization of ERNIE 5.5 Flash, including performance optimization for low-latency scenarios on Kunlunxin P800, deterministic inference throughput optimization, and extreme decoding speed optimization. Baidu open-sourced ERNIE 5.5 Flash Base on November 5, and a person close to Baidu revealed that the new ERNIE model may be released soon.
Why read this
AI score 60
Baidu publicly discloses for the first time the R&D loop of its automated research system ERNIE Evolve, and presents its publicly verified results on two tasks.
Topics
Body · AI translation
On November 9, Baidu's ERNIE disclosed the latest progress on RSI for the first time, unveiling ERNIE Evolve, an automated research system involved in the development of ERNIE 5.5 Flash. It can, around goals and constraints, autonomously propose research directions, conduct experiments, verify results, and accumulate experience; it can both continuously optimize existing paths and explore new directions based on actual results. On the three evaluation metrics of the two tasks NanoChat AutoResearch and NanoGPT Speedrun, it achieved the current publicly verified SOTA performance. Currently, ERNIE Evolve has been applied to the inference optimization of the soon-to-be-released ERNIE 5.5 Flash, including performance optimization for low-latency scenarios on Kunlun Core P800, deterministic inference throughput optimization, and extreme decoding speed optimization.
The ERNIE tech blog pointed out that inference infrastructure optimization is the first landing point of RSI in real R&D. In the future, the same research loop will gradually extend to training system optimization, model capability research, and evaluation system iteration, driving AI to autonomously discover problems, propose hypotheses, design experiments, and verify results. The expansion process will be carried out within clear goals, permissions, and resource boundaries, and will be subject to continuous safety evaluation and human oversight.
On November 5, Baidu officially open-sourced the latest base model ERNIE 5.5 Flash Base, which achieved SOTA performance among models of the same size on multiple benchmarks including knowledge, reasoning, mathematics, code, multilingual, and long context. A person close to Baidu revealed that the new ERNIE model may be released soon.
This article is provided by Baidu ERNIE; QbitAI is authorized to repost it, and the views belong to the original author.