A Garc a

Adhitya Charan, Adwaid Suresh, Anuj Kumar et al.

Progress in deep learning is, at scale, more a matter of systems engineering than of modelling: the behaviour of a model in training (its throughput, its memory footprint, and the numerical fidelity of the result) is determined less by the architecture itself than by how that architecture is expressed on the hardware. To achieve absolute control over this hardware expression while abstracting away systems complexity to make modelling seamless and eliminating the need for repetitive orchestration logic, BluTrain was architected from first principles as a robust, lightweight, and architecture-general training framework in standard C++ and the core CUDA programming model. Every layer is implemented natively: a typed tensor module with reverse-mode autograd, a linear-algebra library, a caching allocator, a multi-mode distributed-execution module, and an MLIR-based deep-learning compiler. In formal evaluations training a 124M-parameter GPT-2 baseline in FP32 on an 8-GPU 6000 Ada system, BluTrain outperforms industry-standard baselines in both throughput (sustaining an average of 407K tokens/s versus PyTorch's 395K tokens/s) and memory efficiency (achieving up to a 22% footprint reduction), while strictly preserving numerical fidelity and converging to a marginally lower final validation loss. With every layer explicitly open to native tuning, the performance ceiling is the framework's own to raise.

Jean Martins, Leonid Mokrushin, Marin Orlic et al.

Intent-based automation for 6G envisions networks steered by high-level goals rather than low-level configurations. Existing LLM-based approaches translate natural language into plausible intent representations but typically omit what production deployment requires: grounding in actual service catalogs, formal validation, and cross-layer decomposition. We address this with an agentic workflow comprising three coupled reasoning layers: (i) grounding the translation in a semantic service catalog that exposes TMF compliant service specifications; (ii) validation of the RDF intent via SHACL structural checking against the TMF Intent Ontology; and (iii) decomposition that selects a CFSS profile via constraint satisfaction over QoS capability envelopes, then covers its infrastructure requirements with RFSS profiles via weighted set cover. Across 930 benchmark runs over six GPT-4.1/5 models, the workflow achieves 97% success in structured mode and 90% on average across natural-language scenarios, with 100% correct rejection of infeasible requests. Grounding LLM context in catalog capability metadata reduces adversarial hallucinations by 26 percentage points; larger gains than scaling model size alone.