【About Us】
We are a leading quantitative trading firm and liquidity provider dedicated to delivering superior risk-adjusted returns. Our trading models have stood the test of time by combining comprehensive mathematical analysis, extensive financial market knowledge, and cutting-edge artificial intelligence technology solutions. We are pioneers in systematic decision-making, algorithmic execution, and active risk management. Our team consists of experienced professionals from top investment banks (such as Morgan Stanley/Merrill Lynch (Bank of America)/UBS/Macquarie) and graduates with outstanding academic backgrounds from institutions like London School of Economics/Oxford University/Nanyang Technological University/National University of Singapore.
【Responsibilities】
- Design and develop Large Language Models (LLMs), including model pre-training, efficient fine-tuning, and performance optimization
- Develop and optimize model training frameworks, implementing key technologies such as distributed training and Parameter-Efficient Fine-Tuning (PEFT)
- Build LLM evaluation systems and design domain-specific benchmarks
- Optimize model inference performance, implement model quantization, pruning, and deployment optimization
【Requirements】
- Master’s degree or above in Computer Science or related fields
- 2+ years of deep learning project development experience, including large-scale model training practice
- Solid foundation in machine learning algorithms
- Excellent experimental design and results analysis capabilities
- Strong coding standards and documentation skills
【Large Model Development】
- Expert in LLM training technologies (such as LoRA, QLoRA, Adapter, and other PEFT methods)
- Deep understanding of Transformer architecture and mainstream pre-training models (such as LLaMA, Mistral) principles and implementation
- Familiar with low-level optimization techniques like Flash Attention and random gradient compression
- Experience in model quantization and compression (such as INT4/INT8 quantization, model pruning, knowledge distillation)
- Experience in inference performance optimization, understanding of vLLM, TensorRT-LLM, and other inference acceleration frameworks
【Distributed Training】
- Expert in PyTorch, deep understanding of distributed training mechanisms like DistributedDataParallel and FSDP
- Familiar with large-scale training frameworks like DeepSpeed and Megatron-LM
- Mastery of 3D parallel (data parallel, tensor parallel, pipeline parallel) training technologies
- Experience in multi-GPU/multi-machine training system design and performance tuning
- Familiar with memory optimization methods like gradient checkpointing and mixed precision training
【System Optimization】
- Expert in Linux systems and CUDA programming
- Deep understanding of GPU architecture and memory management
- Capable of training and inference performance analysis and optimization
- Familiar with distributed storage systems (such as S3, HDFS)
【Model Evaluation】
- Expert in model performance and effectiveness evaluation methods
- Familiar with A/B testing and statistical analysis techniques
- Experience in model interpretability analysis
【Core Technologies】
- Expert in Python data processing (numpy, pandas, scikit-learn, etc.)
- Proficient in using PySpark for large-scale data processing
- Capable of designing and implementing custom loss functions
- Familiar with data visualization and experimental analysis tools
【Bonus Qualifications】
- Published papers or contributions to open-source projects related to large models
- Familiar with low-level implementation of core architectures like Transformer
- Experience in financial institutions or quantitative investment
- Understanding of financial market mechanisms and trading strategies
【Benefits】
- 2 days remote work per week, up to 25 days overseas remote work annually
- Competitive base salary and bonuses
- Flat organizational structure, positive team atmosphere
- Multiple company overseas trips annually
- Recreational activities including sports and board games
【Location】
Shanghai