🤖 AI 中文导读 · 由岗位 JD 原文生成
1. Participate in research on post-training algorithms for large language models, focusing on core capabilities such as code, tool calling, agents, instruction following,……
🈶
中文翻译 · AI 生成，以原文为准
1. Participate in research on post-training algorithms for large language models, focusing on core capabilities such as code, tool calling, agents, instruction following, knowledge utilization, complex reasoning, and long text, and build a complete post-training technical system from training data and training methods to evaluation and analysis, continuously improving the general capabilities and real user experience of foundation models.
2. Establish a closed loop of "data-training-evaluation-analysis-iteration," identify model capability bottlenecks through ablation experiments, error analysis, and attribution analysis, and drive continuous iteration of training data and model performance improvement.
3. Work closely with teams such as data engineers and Infra engineers to design and implement automated pipelines for complex code and agent trajectory data, systematically improving model performance in core scenarios such as code engineering, AI search, tool calling, and task planning.
4. Track the latest research progress in the field of large model post-training domestically and internationally, reproduce and validate cutting-edge methods, explore new training paradigms with long-term value, and research and build core capabilities such as long-horizon interaction, task planning, self-feedback and improvement, and anthropomorphism.
1. Graduated from a key university with a major in computer science, artificial intelligence, automation, electronic information, mathematics, or related fields, with a master's degree or above.
2. Strong interest in improving large model capabilities, solid foundation in large model algorithms and training, able to continuously track papers and open-source projects and translate research ideas into verifiable experimental plans. Project experience in large language model fine-tuning, alignment, data synthesis, model evaluation, or capability analysis.
3. Solid programming and computer fundamentals, proficiency in languages such as Python, excellent programming habits and engineering implementation skills, familiarity with the Linux development environment, and proficiency in Git and common R&D tools such as AICoding.
4. Strong experimental design and analysis skills, able to locate model issues through controlled experiments, ablation experiments, and Bad Case analysis.
5. Good communication and collaboration skills, self-drive, and sense of responsibility, able to independently advance research tasks.
This position is a unified recruitment position of iFlytek Group. Candidates will be hired after passing resume screening, initial interview, second interview, final review, and other stages.
Bonus points
1. Achieved excellent results in algorithmic programming competitions, mathematics competitions, or other related machine learning competitions.
2. Published high-level papers at machine learning or natural language processing conferences such as NeurIPS, ICML, ICLR, ACL, EMNLP, and AAAI.
3. Experience with models of more than ten billion parameters or distributed training with more than a thousand GPUs.