Responsibilities
1. Optimize the data enhancement algorithm, optimize the data synthesis algorithm in the pre-training/SFT/RLHF stage, and improve the accuracy, diversity and domain adaptability of the training data 2. Responsible for the research and development of large models, leading the training innovation of Code-LLM and logical reasoning large models, overcoming technical difficulties such as NL2Code and complex reasoning chain generation, and promoting LLM Test-Time Compute, Parameter-Efficient Finetuning and other technologies are put into practice 3. Promote the implementation of large model applications, explore the application value of RAG-QA robots, data insight robots, etc. in real scenarios, and achieve continuous iteration of application effects based on the evaluation system 4. Follow up on open source SOTA large models, explore large model scaling laws and practice Post-Training of open source large models in the field of data intelligence, and accumulate technology patents.
Qualifications
1. Master degree or above, computer science/artificial intelligence/applied mathematics and other related majors, top conference (ACL/NeurIPS/ICML/EMNLP/ICLR) paper or competition winners are preferred 2. Solid foundation of deep learning technology, familiar with deep learning training frameworks (PyTorch/Huggingface, etc.), understanding of common large language models, Embedding architectures and training methods are preferred 3. Excellent problem analysis and solving skills, as well as good communication and collaboration skills, able to explore new technologies with the team.
