Search for: All records

Linear RNN architectures, like Mamba, can be competitive with Transformer models in language modeling while having advantageous deployment characteristics. Given the focus on training large-scale Transformer models, we consider the challenge of converting these pretrained models for deployment. We demonstrate that it is feasible to distill large Transformers into linear RNNs by reusing the linear projection weights from attention layers with academic GPU resources. The resulting hybrid model, which incorporates a quarter of the attention layers, achieves performance comparable to the original Transformer in chat benchmarks and outperforms open-source hybrid Mamba models trained from scratch with trillions of tokens in both chat benchmarks and general benchmarks. Moreover, we introduce a hardware-aware speculative decoding algorithm that accelerates the inference speed of Mamba and hybrid models. Overall we show how, with limited computation resources, we can remove many of the original attention layers and generate from the resulting model more efficiently. Our top-performing model, distilled from Llama3-8B-Instruct, achieves a 29.61 length-controlled win rate on AlpacaEval 2 against GPT-4 and 7.35 on MT-Bench, surpassing the best 8B scale instruction-tuned linear RNN model. We also find that the distilled model has natural length extrapolation, showing almost perfect accuracy in the needle-in-a-haystack test at 20x the distillation length. 

Token-free language models learn directly from raw bytes and remove the inductive bias of subword tokenization. Operating on bytes, however, results in significantly longer sequences. In this setting, standard autoregressive Transformers scale poorly as the effective memory required grows with sequence length. The recent development of the Mamba state space model (SSM) offers an appealing alternative approach with a fixed-sized memory state and efficient decoding. We propose MambaByte, a token-free adaptation of the Mamba SSM trained autoregressively on byte sequences. In terms of modeling, we show MambaByte to be competitive with, and even to outperform, state-of-the-art subword Transformers on language modeling tasks while maintaining the benefits of token-free language models, such as robustness to noise. In terms of efficiency, we develop an adaptation of speculative decoding with tokenized drafting and byte-level verification. This results in a 2.6× inference speedup to the standard MambaByte implementation, showing similar decoding efficiency as the subword Mamba. These findings establish the viability of SSMs in enabling token-free language modeling. 

In black-box optimization problems, we aim to maximize an unknown objective function, where the function is only accessible through feedbacks of an evaluation or simulation oracle. In real-life, the feedbacks of such oracles are often noisy and available after some unknown delay that may depend on the computation time of the oracle. Additionally, if the exact evaluations are expensive but coarse approximations are available at a lower cost, the feedbacks can have multi-fidelity. In order to address this problem, we propose a generic extension of hierarchical optimistic tree search (HOO), called ProCrastinated Tree Search (PCTS), that flexibly accommodates a delay and noise-tolerant bandit algorithm. We provide a generic proof technique to quantify regret of PCTS under delayed, noisy, and multi-fidelity feedbacks. Specifically, we derive regret bounds of PCTS enabled with delayed-UCB1 (DUCB1) and delayed-UCB-V (DUCBV) algorithms. Given a horizon T, PCTS retains the regret bound of non-delayed HOO for expected delay of O(log T), and worsens by T^((1-α)/(d+2)) for expected delays of O(T^(1-α)) for α ∈ (0,1]. We experimentally validate on multiple synthetic functions and hyperparameter tuning problems that PCTS outperforms the state-of-the-art black-box optimization methods for feedbacks with different noise levels, delays, and fidelity. 

UDO is a versatile tool for offline tuning of database systems for specific workloads. UDO can consider a variety of tuning choices, reaching from picking transaction code variants over index selections up to database system parameter tuning. UDO uses reinforcement learning to converge to near-optimal configurations, creating and evaluating different configurations via actual query executions (instead of relying on simplifying cost models). To cater to different parameter types, UDO distinguishes heavy parameters (which are expensive to change, e.g. physical design parameters) from light parameters. Specifically for optimizing heavy parameters, UDO uses reinforcement learning algorithms that allow delaying the point at which the reward feedback becomes available. This gives us the freedom to optimize the point in time and the order in which different configurations are created and evaluated (by benchmarking a workload sample). UDO uses a cost-based planner to minimize reconfiguration overheads. For instance, it aims to amortize the creation of expensive data structures by consecutively evaluating configurations using them. We evaluate UDO on Postgres as well as MySQL and on TPC-H as well as TPC-C, optimizing a variety of light and heavy parameters concurrently. 

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.