PackMamba: Efficient Processing of Variable-Length Sequences in Mamba training
This addresses a bottleneck in training state-space models for generative AI, offering incremental improvements in efficiency for researchers and practitioners handling variable-length data.
The paper tackled inefficiency in training Mamba models with variable-length sequences by proposing PackMamba, which modifies operators to avoid inter-sequence information passing, achieving speedups of 3.06x on a 1.4B model and 2.62x on a 2.8B model compared to baseline single-sequence processing.
With the evolution of large language models, traditional Transformer models become computationally demanding for lengthy sequences due to the quadratic growth in computation with respect to the sequence length. Mamba, emerging as a groundbreaking architecture in the field of generative AI, demonstrates remarkable proficiency in handling elongated sequences with reduced computational and memory complexity. Nevertheless, the existing training framework of Mamba presents inefficiency with variable-length sequence inputs. Either single-sequence training results in low GPU utilization, or batched processing of variable-length sequences to a maximum length incurs considerable memory and computational overhead. To address this problem, we analyze the performance of bottleneck operators in Mamba under diverse tensor shapes and proposed PackMamba, a high-throughput Mamba that efficiently handles variable-length sequences. Diving deep into state-space models (SSMs), we modify the parallel operators to avoid passing information between individual sequences while maintaining high performance. Experimental results on an NVIDIA A100 GPU demonstrate throughput exceeding the baseline single-sequence processing scheme: 3.06x speedup on the 1.4B model and 2.62x on the 2.8B model.