TrimLLaMA: Progressive Layer Dropping for Domain-SpecificLLMs

Specializing large language models (LLMs) for local deployment in domainspecific use cases is necessary for strong performance while meeting latency and privacy constraints.However, conventional task-specific adaptation approaches do not show simultaneous memory saving and inference speedup at deployment time.Practical compression techniques like quantization and pruning require dedicated hardware or kernel support to achieve measured inference speedup.We develop TRIMLLAMA based on the layer-wise specialization phenomenon we empirically observed and verified on contemporary LLMs.TRIMLLAMA reduces the depth of LLMs via progressive layer dropping.We show it retains LLMs' capacity in specific domains and achieves inference speedup irrespective of hardware and deep learning frameworks.We evaluated TRIMLLAMA on LLMs of various sizes for inference; models adapted on medical, legal, and financial datasets all demonstrate 2.1 -5.7 inference speedup on consumer GPUs and up to 3.1 speedup on A100 when compared to state-of-the-art model compression algorithms, with no loss in accuracy at 5060% model compression ratio.