Information-seeking agents have emerged as a powerful paradigm for knowledge-intensive tasks, yet today’s systems remain specialized for the open web, documents, or local knowledge bases, hindering scalable and cross-domain deployment. We present the first systematic empirical study of consolidating these information-seeking agents into a single foundation agentic model. We compare two paradigms—data-level mixing, which trains a unified model on a mixture of datasets, and parameter-level merging, which merges independently trained experts in parameter space—across 3 training scenarios, evaluating 26 representative parameter-level methods on 10 benchmarks. To compare across heterogeneous benchmarks, we introduce a geometric Composite Score and an Imbalance Score that describe overall performance and task skew. Our analysis shows that (i) well-designed parameter-level merging attains parity with data mixing at a fraction of its training cost and is order-agnostic; (ii) parameter-level merging structurally preserves out-of-domain capabilities that data mixing universally forgets; and (iii) cross-scenario stability is strongly tied to consolidation quality. We distil our observations into a method-selection guide and design principles for next-generation merging operators.

Previous studies have proved that cross-lingual knowledge distillation can significantly improve the performance of pre-trained models for cross-lingual similarity matching tasks. However, the student model needs to be large in this operation. Otherwise, its performance will drop sharply, thus making it impractical to be deployed to memory-limited devices. To address this issue, we delve into cross-lingual knowledge distillation and propose a multi-stage distillation framework for constructing a small-size but high-performance cross-lingual model. In our framework, contrastive learning, bottleneck, and parameter recurrent strategies are delicately combined to prevent performance from being compromised during the compression process. The experimental results demonstrate that our method can compress the size of XLM-R and MiniLM by more than 50%, while the performance is only reduced by about 1%.

As the multi-modal e-commerce is thriving, high-quality advertising product copywriting has gain more attentions, which plays a crucial role in the e-commerce recommender, advertising and even search platforms. The advertising product copywriting is able to enhance the user experience by highlighting the product’s characteristics with textual descriptions and thus to improve the likelihood of user click and purchase. Automatically generating product copywriting has attracted noticeable interests from both academic and industrial communities, where existing solutions merely make use of a product’s title and attribute information to generate its corresponding description. However, in addition to the product title and attributes, we observe that there are various auxiliary descriptions created by the shoppers or marketers in the e-commerce platforms (namely human knowledge), which contains valuable information for product copywriting generation, yet always accompanying lots of noises. In this work, we propose a novel solution to automatically generating product copywriting that involves all the title, attributes and denoised auxiliary knowledge. To be specific, we design an end-to-end generation framework equipped with two variational autoencoders that works interactively to select informative human knowledge and generate diverse copywriting.

Existing zero-shot cross-lingual transfer methods rely on parallel corpora or bilingual dictionaries, which are expensive and impractical for low-resource languages. To disengage from these dependencies, researchers have explored training multilingual models on English-only resources and transferring them to low-resource languages. However, its effect is limited by the gap between embedding clusters of different languages. To address this issue, we propose Embedding-Push, Attention-Pull, and Robust targets to transfer English embeddings to virtual multilingual embeddings without semantic loss, thereby improving cross-lingual transferability. Experimental results on mBERT and XLM-R demonstrate that our method significantly outperforms previous works on the zero-shot cross-lingual text classification task and can obtain a better multilingual alignment.