We introduce OLMoE, a fully open, state-of-the-art language model leveraging sparse Mixture-of-Experts (MoE). OLMoE-1B-7B has 7 billion (B) parameters but uses only 1B per input token. We pretrain it on 5 trillion tokens and further adapt it to create OLMoE-1B-7B-Instruct. Our models outperform all available models with similar active parameters, even surpassing larger ones like Llama2-13B-Chat and DeepSeekMoE-16B. We present various experiments on MoE training, analyze routing in our model showing high specialization, and open-source all aspects of our work: model weights, training data, code, and logs.

In this paper, we propose a method to improve prediction accuracy of semantic segmentation methods as follows: (1) construct a neural network that has pre-processing layers based on a convolutional autoencoder ahead of a semantic segmentation network, and (2) train the entire network initialized by the weights of the pre-trained autoencoder. We applied this method to the fully convolutional network (FCN) and experimentally compared its prediction accuracy on the cityscapes dataset. The Mean IoU of the proposed target model with the He normal initialization is 18.7% higher than that of FCN with the He normal initialization. In addition, those of the modified models of the target model are significantly higher than that of FCN with the He normal initialization. The accuracy and loss curves during the training showed that these are resulting from the improvement of the generalization ability. All of these results provide strong evidence that the proposed method is significantly effective in improving the prediction accuracy of FCN. The proposed method has the following features: it is comparatively simple, whereas the effect on improving the generalization ability and prediction accuracy of FCN is significant; the increase in the number of parameters by using it is very small, and that in the computation time is substantially large. In principle, the proposed method can be applied to other semantic segmentation methods. For semantic segmentation, at present, there is no effective way to improve the prediction accuracy of existing methods. None have published a method which is the same as or similar to our method and none have used such a method in practice. Therefore, we believe that our method is useful in practice and worthy of being widely known and used.

In the realm of computer vision, Neural Fields have gained prominence as a contemporary tool harnessing neural networks for signal representation. Despite the remarkable progress in adapting these networks to solve a variety of problems, the field still lacks a comprehensive theoretical framework. This article aims to address this gap by delving into the intricate interplay between initialization and activation, providing a foundational basis for the robust optimization of Neural Fields. Our theoretical insights reveal a deep-seated connection among network initialization, architectural choices, and the optimization process, emphasizing the need for a holistic approach when designing cutting-edge Neural Fields.

Despite recent progress in generative open-domain dialogue, the issue of low response diversity persists. Prior works have addressed this issue via either novel objective functions, alternative learning approaches such as variational frameworks, or architectural extensions such as the Randomized Link (RL) Transformer. However, these approaches typically entail either additional difficulties during training/inference, or a significant increase in model size and complexity. Hence, we propose the \underline{Pa}rtially \underline{Ra}ndomized trans\underline{Former} (PaRaFormer), a simple extension of the transformer which involves freezing the weights of selected layers after random initialization. Experimental results reveal that the performance of the PaRaformer is comparable to that of the aforementioned approaches, despite not entailing any additional training difficulty or increase in model complexity.