Machine Learning Expert

As a Machine Learning Engineer III specializing in LLM Training, your role will involve working on specialized LLMs that are designed to understand and reason over vast enterprise codebases. You will engage in real model training activities, including RL loops, PEFT, verifiable rewards, and long-context modeling, rather than focusing solely on API integration. Your responsibilities will include: - Building custom training loops using technologies such as PyTorch, HuggingFace, TRL, and Unsloth. - Running full-stack ML experiments involving data processing, training, evaluation, and iteration. - Developing scalable training infrastructure utilizing FSDP/DeepSpeed and distributed training methodologies. For this role, you are required to have: - 3+ years of hands-on experience in deep learning with a focus on actual model training. - Proven expertise in training transformers or large DNNs either from scratch or from existing checkpoints. - Experience in implementing RL algorithms or custom training pipelines. - Proficiency in debugging training issues including NaNs, KL drift, reward hacking, etc. - Comfort with fast-paced, on-site, high-performance execution. - Background in program analysis and AST tooling. - Previous experience in long-context modeling. - Ability to contribute to shaping the core training architecture with high-impact decisions. This position offers a challenging opportunity to work on cutting-edge technologies in the field of machine learning and contribute significantly to the development of advanced LLMs. As a Machine Learning Engineer III specializing in LLM Training, your role will involve working on specialized LLMs that are designed to understand and reason over vast enterprise codebases. You will engage in real model training activities, including RL loops, PEFT, verifiable rewards, and long-context modeling, rather than focusing solely on API integration. Your responsibilities will include: - Building custom training loops using technologies such as PyTorch, HuggingFace, TRL, and Unsloth. - Running full-stack ML experiments involving data processing, training, evaluation, and iteration. - Developing scalable training infrastructure utilizing FSDP/DeepSpeed and distributed training methodologies. For this role, you are required to have: - 3+ years of hands-on experience in deep learning with a focus on actual model training. - Proven expertise in training transformers or large DNNs either from scratch or from existing checkpoints. - Experience in implementing RL algorithms or custom training pipelines. - Proficiency in debugging training issues including NaNs, KL drift, reward hacking, etc. - Comfort with fast-paced, on-site, high-performance execution. - Background in program analysis and AST tooling. - Previous experience in long-context modeling. - Ability to contribute to shaping the core training architecture with high-impact decisions. This position offers a challenging opportunity to work on cutting-edge technologies in the field of machine learning and contribute significantly to the development of advanced LLMs.