Role Positioning
As a large model engineer at Amber Group, you will be at the forefront of the integration of Web3 and AI technology, focusing on the training and optimization of large language models (LLM). You will work with cross-functional teams to promote innovative applications of models in the Web3 field. If you are passionate about artificial intelligence, good at solving complex problems, and eager to grow in an environment full of creativity and technical leadership, this will be your ideal position!

Job Responsibilities

• Model Training and Optimization: Design and implement training strategies for large language models, including supervised fine-tuning (SFT), reinforcement learning (such as GRPO, PPO) and other methods to improve the intelligence of models in the Web3 field.
• Data Processing and Generation: Build high-quality training data sets, perform data distillation and Long & Short Chain of Thought (CoT) data generation to ensure that the model has strong reasoning capabilities.
• Model Architecture and Evaluation: Explore and apply advanced model architectures such as Mixture of Experts (MoE), develop model evaluation frameworks and indicators, and continuously optimize model performance.
• Distributed training and deployment: Develop and maintain distributed training solutions for models to ensure efficient training and stable deployment of models.
• Exploration of cutting-edge technology: Track the latest research trends in the field of AI, such as OpenAi GPT-4.5, DeepSeek-R1, etc., and promote the innovative application of technology in actual business.

Job requirements

• Educational background: Bachelor's degree in computer science, artificial intelligence, machine learning or related fields, master's or doctoral degree preferred.
• Proficient in Transformer architecture, proficient in Transformer Reinforcement Learning (TRL), PyTorch or TensorFlow deep learning/reinforcement learning frameworks, etc.
• Experience in fine-tuning large language models, familiar with technologies such as Reasoning-Oriented Reinforcement Learning (RORL).
• Familiar with distributed training frameworks, and practical experience in technologies such as model parallelism, Flash Attention, and LoRA.
• Proficient in programming languages ​​such as Python and Go, with good coding style and software engineering practical experience.
• Familiar with model servitization technologies, such as Triton, vLLM, TGI, etc., and those with experience in inference optimization are preferred.
• Able to read and implement cutting-edge papers, write technical reports or blogs.
• Those who have published papers or contributed to open source projects at top conferences (such as NeurIPS, ICLR, ICML, ACL) are preferred.
• Have excellent teamwork and communication skills, and be able to collaborate efficiently with cross-functional teams.
• Have a deep understanding of the open source AI community, and those who have contributed to related projects are preferred.