Search for: All records

We introduce SpareLLM, SelectingPassableAndResource-EfficientLLMs, a novel LLM framework designed to minimize the inference costs (i.e., resource-efficient) of large-scale NLP tasks while ensuring sufficient result quality (i.e., passable). It enables users to specify an equivalence constraint in terms of the equivalence of outputs to those of the most powerful LLM. SpareLLM then generates results that deviate from the outputs of this LLM only with a probability below a user-defined threshold. SpareLLM employs a profiling phase that evaluates the performance of multiple LLMs to identify those that meet the user-defined equivalence level. It optimizes the tradeoff between profiling overheads and the anticipated cost savings resulting from profiling. Moreover, SpareLLM further reduces inference costs by strategically leveraging a mix of LLMs. Our experiments on five real-world datasets show that SpareLLM achieves significant cost savings, up to 8.6x, while generating equivalent outputs in 90% of cases compared to GPT-4-Turbo. Compared to recent LLM cascading baselines, SpareLLM demonstrates a superior tradeoff between cost and accuracy, accounting for 91.1% and 83.8% of the points on the Pareto curve for OpenAI and Llama models. 

We introduce ThalamusDB, a novel approximate query processing system that processes complex SQL queries on multi-modal data. ThalamusDB supports SQL queries integrating natural language predicates on visual, audio, and text data. To answer such queries, ThalamusDB exploits a collection of zero-shot models in combination with relational processing. ThalamusDB utilizes deterministic approximate query processing, harnessing the relative efficiency of relational processing to mitigate the computational demands of machine learning inference. For evaluating a natural language predicate, ThalamusDB requests a small number of labels from users. User can specify their preferences on the performance objective regarding the three relevant metrics: approximation error, computation time, and labeling overheads. The ThalamusDB query optimizer chooses optimized plans according to user preferences, prioritizing data processing and requested labels to maximize impact. Experiments with several real-world data sets, taken from Craigslist, YouTube, and Netflix, show that ThalamusDB achieves an average speedup of 35.0x over MindsDB, an exact processing baseline, and outperforms ABAE, a sampling-based method, in 78.9% of cases. 

SkinnerDB uses reinforcement learning for reliable join ordering, exploiting an adaptive processing engine with specialized join algorithms and data structures. It maintains no data statistics and uses no cost or cardinality models. Also, it uses no training workloads nor does it try to link the current query to seemingly similar queries in the past. Instead, it uses reinforcement learning to learn optimal join orders from scratch during the execution of the current query. To that purpose, it divides the execution of a query into many small time slices. Different join orders are tried in different time slices. SkinnerDB merges result tuples generated according to different join orders until a complete query result is obtained. By measuring execution progress per time slice, it identifies promising join orders as execution proceeds. Along with SkinnerDB, we introduce a new quality criterion for query execution strategies. We upper-bound expected execution cost regret, i.e., the expected amount of execution cost wasted due to sub-optimal join order choices. SkinnerDB features multiple execution strategies that are optimized for that criterion. Some of them can be executed on top of existing database systems. For maximal performance, we introduce a customized execution engine, facilitating fast join order switching via specialized multi-way join algorithms and tuple representations. We experimentally compare SkinnerDB’s performance against various baselines, including MonetDB, Postgres, and adaptive processing methods. We consider various benchmarks, including the join order benchmark, TPC-H, and JCC-H, as well as benchmark variants with user-defined functions. Overall, the overheads of reliable join ordering are negligible compared to the performance impact of the occasional, catastrophic join order choice.