Machine Learning Engineer, Convergence

• Direct experience training LLMs or VLMs with methods such as distillation, supervised fine-tuning, and policy optimisation
• Experience with large-scale distributed training and inference
• Experience debugging ML systems and codebases
• Proficiency in frameworks like PyTorch
• Strong general foundations in software engineering, with an interest in non-ML software too – doing whatever it takes to build incredible models as a small team

Desirable

• We are looking for talented ML engineers and researchers to join our team and focus on training models which power Proxy, our generalist agent
• You will work with a small team — equipped with lots of GPUs – to train models, including multi-modal vision LLMs and action models
• You will also be laying the foundations of machine learning engineering at Convergence, utilising tools and best practices to improve our ML workflows
• Your role will span the full stack of model training, including:
• Implementing and testing different fine-tuning and preference learning techniques like DPO
• Building datasets through scrappy methods, including synthetic data pipelines, data scrapers, combining open source datasets, and spinning up data annotations
• Conducting experiments to find good data mixes, regularisers, and hyperparameters
• Data collection and cleaning. Implementing scalable data pipelines
• Designing processes and software to facilitate ML experimentation
• Implementing and debugging new ML frameworks and approaches
• Training models
• Building tooling to evaluate and play with your models
• Improving a variety of things like data quality, data formatting, job startup speed, evaluation speed, ease of experimentation
• Adjusting our infrastructure for model inference, such as improving constrained generation for tool-use
• Working with engineering to integrate models into Proxy