Matt Hoffman

Gemma Team, Morgane Riviere, Shreya Pathak et al. · deepmind

In this work, we introduce Gemma 2, a new addition to the Gemma family of lightweight, state-of-the-art open models, ranging in scale from 2 billion to 27 billion parameters. In this new version, we apply several known technical modifications to the Transformer architecture, such as interleaving local-global attentions (Beltagy et al., 2020a) and group-query attention (Ainslie et al., 2023). We also train the 2B and 9B models with knowledge distillation (Hinton et al., 2015) instead of next token prediction. The resulting models deliver the best performance for their size, and even offer competitive alternatives to models that are 2-3 times bigger. We release all our models to the community.

Caglar Gulcehre, Srivatsan Srinivasan, Jakub Sygnowski et al. · deepmind

Deep neural networks are the most commonly used function approximators in offline reinforcement learning. Prior works have shown that neural nets trained with TD-learning and gradient descent can exhibit implicit regularization that can be characterized by under-parameterization of these networks. Specifically, the rank of the penultimate feature layer, also called \textit{effective rank}, has been observed to drastically collapse during the training. In turn, this collapse has been argued to reduce the model's ability to further adapt in later stages of learning, leading to the diminished final performance. Such an association between the effective rank and performance makes effective rank compelling for offline RL, primarily for offline policy evaluation. In this work, we conduct a careful empirical study on the relation between effective rank and performance on three offline RL datasets : bsuite, Atari, and DeepMind lab. We observe that a direct association exists only in restricted settings and disappears in the more extensive hyperparameter sweeps. Also, we empirically identify three phases of learning that explain the impact of implicit regularization on the learning dynamics and found that bootstrapping alone is insufficient to explain the collapse of the effective rank. Further, we show that several other factors could confound the relationship between effective rank and performance and conclude that studying this association under simplistic assumptions could be highly misleading.

Pier Giuseppe Sessa, Robert Dadashi, Léonard Hussenot et al.

Reinforcement learning from human feedback (RLHF) is a key driver of quality and safety in state-of-the-art large language models. Yet, a surprisingly simple and strong inference-time strategy is Best-of-N sampling that selects the best generation among N candidates. In this paper, we propose Best-of-N Distillation (BOND), a novel RLHF algorithm that seeks to emulate Best-of-N but without its significant computational overhead at inference time. Specifically, BOND is a distribution matching algorithm that forces the distribution of generations from the policy to get closer to the Best-of-N distribution. We use the Jeffreys divergence (a linear combination of forward and backward KL) to balance between mode-covering and mode-seeking behavior, and derive an iterative formulation that utilizes a moving anchor for efficiency. We demonstrate the effectiveness of our approach and several design choices through experiments on abstractive summarization and Gemma models. Aligning Gemma policies with BOND outperforms other RLHF algorithms by improving results on several benchmarks.

Bobak Shahriari, Abbas Abdolmaleki, Arunkumar Byravan et al.

Actor-critic algorithms that make use of distributional policy evaluation have frequently been shown to outperform their non-distributional counterparts on many challenging control tasks. Examples of this behavior include the D4PG and DMPO algorithms as compared to DDPG and MPO, respectively [Barth-Maron et al., 2018; Hoffman et al., 2020]. However, both agents rely on the C51 critic for value estimation.One major drawback of the C51 approach is its requirement of prior knowledge about the minimum andmaximum values a policy can attain as well as the number of bins used, which fixes the resolution ofthe distributional estimate. While the DeepMind control suite of tasks utilizes standardized rewards and episode lengths, thus enabling the entire suite to be solved with a single setting of these hyperparameters, this is often not the case. This paper revisits a natural alternative that removes this requirement, namelya mixture of Gaussians, and a simple sample-based loss function to train it in an off-policy regime. We empirically evaluate its performance on a broad range of continuous control tasks and demonstrate that it eliminates the need for these distributional hyperparameters and achieves state-of-the-art performance on a variety of challenging tasks (e.g. the humanoid, dog, quadruped, and manipulator domains). Finallywe provide an implementation in the Acme agent repository.