Large language models (LLMs) are increasingly used in applications where LLM inputs may span many different tasks. Recent work has found that the choice of LLM is consequential, and different LLMs may be good for different input samples. Prior approaches have thus explored how engineers might select an LLM to use for each sample (i.e.). While existing routing methods mostly require training auxiliary models on human-annotated data, our work explores whether it is possible to perform routing. We propose Smoothie, a weak supervision-inspired routing approach that requires no labeled data. Given a set of outputs from different LLMs, Smoothie constructs a latent variable graphical model over embedding representations of observable LLM outputs and unknown "true" outputs. Using this graphical model, we estimate sample-dependent quality scores for each LLM, and route each sample to the LLM with the highest corresponding score. We find that Smoothie's LLM quality-scores correlate with ground-truth model quality (correctly identifying the optimal model on 9/14 tasks), and that Smoothie outperforms baselines for routing by up to 10 points accuracy.

Large language models (LLMs) are increasingly used in applications where LLM inputs may span many different tasks. Recent work has found that the choice of LLM is consequential, and different LLMs may be good for different input samples. Prior approaches have thus explored how engineers might select an LLM to use for each sample (i.e. While existing routing methods mostly require training auxiliary models on human-annotated data, our work explores whether it is possible to perform unsupervised routing. We propose Smoothie, a weak supervision-inspired routing approach that requires no labeled data.

Hyper-parameter optimization is the black art of tuning a learner's control parameters. In software analytics, a repeated result is that such tuning can result in dramatic performance improvements. Despite this, hyper-parameter optimization is often applied rarely or poorly in software analytics--perhaps due to the CPU cost of exploring all those parameter options can be prohibitive. We theorize that learners generalize better when the loss landscape is ``smooth''. This theory is useful since the influence on ``smoothness'' of different hyper-parameter choices can be tested very quickly (e.g. for a deep learner, after just one epoch). To test this theory, this paper implements and tests SMOOTHIE, a novel hyper-parameter optimizer that guides its optimizations via considerations of ``smothness''. The experiments of this paper test SMOOTHIE on numerous SE tasks including (a) GitHub issue lifetime prediction; (b) detecting false alarms in static code warnings; (c) defect prediction, and (d) a set of standard ML datasets. In all these experiments, SMOOTHIE out-performed state-of-the-art optimizers. Better yet, SMOOTHIE ran 300% faster than the prior state-of-the art. We hence conclude that this theory (that hyper-parameter optimization is best viewed as a ``smoothing'' function for the decision landscape), is both theoretically interesting and practically very useful. To support open science and other researchers working in this area, all our scripts and datasets are available on-line at https://github.com/yrahul3910/smoothness-hpo/.

State-action value functions (i.e., Q-values) are ubiquitous in reinforcement learning (RL), giving rise to popular algorithms such as SARSA and Q-learning. We propose a new notion of action value defined by a Gaussian smoothed version of the expected Q-value. We show that such smoothed Q-values still satisfy a Bellman equation, making them learnable from experience sampled from an environment. Moreover, the gradients of expected reward with respect to the mean and covariance of a parameterized Gaussian policy can be recovered from the gradient and Hessian of the smoothed Q-value function. Based on these relationships, we develop new algorithms for training a Gaussian policy directly from a learned smoothed Q-value approximator. The approach is additionally amenable to proximal optimization by augmenting the objective with a penalty on KL-divergence from a previous policy. We find that the ability to learn both a mean and covariance during training leads to significantly improved results on standard continuous control benchmarks.