About
Multiverse Computing is a deep-tech company that develops quantum software and AI solutions. Its products include CompactifAI, a compression tool for foundational AI models, and Singularity, a quantum and quantum-inspired optimization platform for complex industry problems.
Responsibilities
•Train and post-train LLMs and multimodal models on multi-node GPU clusters using SFT, preference optimisation, RL, and distillation
•Design model compression techniques and healing recipes to recover the quality of compressed models
•Build RL environments, reward functions, and verifiers for reasoning, coding, and agentic tasks; extend RL methods to world models and Physical AI
•Research and prototype post-transformer and hybrid architectures, from small-scale ablations to scaling recipes
•Build data pipelines for curation, filtering, deduplication, and large-scale synthetic data generation
•Design ablations and evaluations using current benchmark suites and internal evaluations
•Port and optimise models for inference on accelerators and edge chips; profile latency, memory, and throughput
•Document experiments and results in internal reports and, where relevant, papers and open-source releases
•Review code, mentor teammates, and follow and bring in the latest research
Requirements
•Master's or PhD in Computer Science, AI, Mathematics, Physics, or a related field, or equivalent industry experience
•At least 4 years of hands-on experience training or post-training transformer models
•Strong PyTorch skills and ability to work below the library level with model code, custom losses, and parallelism configuration
•Experience with distributed training frameworks and parallelism on multi-node clusters with Slurm/K8s
•Hands-on post-training experience with SFT, DPO, and at least one RL method
•Strong foundations in linear algebra, optimisation, and probability, and understanding of training dynamics
•Good understanding of GPU fundamentals, including memory hierarchy, mixed precision, profiling, and throughput
•Rigorous experimental practice, including ablation design, reproducibility, experiment tracking, and noise-aware evaluation
•Strong debugging and communication skills
•Fluent English
Nice to have
•NVIDIA stack or similar frameworks
•RL infrastructure
•Model compression
•Post-transformer architectures
•Physical AI
•Multimodal model training
•Inference on alternative hardware
•Large-scale data processing and synthetic data generation
•Evaluation tooling
•Familiarity with Artificial Analysis
•Publications at top venues
•Notable open-source contributions
•Models released on Hugging Face
•Spanish
Conditions
Not specified in the posting.
