Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces RATrain, a resource-aware training runtime designed specifically for dense large language models on bandwidth-constrained heterogeneous supercomputing platforms like MT-3000. By treating the training process as a scheduling problem, RATrain optimizes gradient synchronization and data movement while adhering to strict memory constraints. The implementation demonstrates significant performance improvements, achieving up to 1.35x speedup and maintaining high scaling efficiency across multiple configurations of LLaMA-2 models.
RATrain achieves a remarkable 1.35x end-to-end speedup for LLM training on bandwidth-limited supercomputers, challenging the notion that high-bandwidth environments are necessary for efficiency.
Production heterogeneous supercomputing platforms are increasingly used to host large language model (LLM) training workloads. However, existing GPU-oriented training runtimes typically rely on high-bandwidth device memory, fast interconnects, and mature collective communication libraries, making them difficult to directly adapt to MT-3000, a platform with an explicit memory hierarchy, limited usable DDR capacity, and constrained inter-cluster communication. This paper presents RATrain, a resource-aware training runtime for dense LLMs on bandwidth-constrained heterogeneous supercomputing platforms. RATrain formulates standard non-interleaved 1F1B training as a training-state lifecycle scheduling problem, and schedules gradient synchronization, parameter update, parameter-view prefetching, and activation recovery at layer-level and stage-local granularity. RATrain further combines an MT-3000-aware execution backend for efficient and predictable FP16 GEMM, Attention Backward, and explicit data movement with a resource-aware planner that selects feasible training configurations under the 20GB usable-DDR constraint per compute cluster. We implement RATrain on a real MT-3000 platform and evaluate it using LLaMA-2-7B, Baichuan2-13B, Qwen2.5-32B, and LLaMA-2-70B configurations. Results show that RATrain achieves up to 1.35$\times$ end-to-end speedup over MT-3000-adapted GPU-style training strategies. For LLaMA-2-7B, RATrain scales to 1024 compute clusters, reaches 112,790.55 tokens/s, and achieves 97.0\% scaling efficiency. A further 1.028B-token correctness run shows that RATrain preserves the loss trajectory of a semantically equivalent Baseline-1F1B run, with a maximum relative loss deviation of 0.081\%.