The paper investigates weight‑space merging of independently fine‑tuned multilingual machine translation models. Experiments show that merging is more successful when models share a target language, yet it still cannot match the peak performance of language‑specific checkpoints. When target languages differ, performance drops sharply, and analysis reveals that overlapping neuron activation and incompatible upper‑layer geometries cause these failures.
By Baban Gain, Trilok Nath Singh, Asif Ekbal
arXiv:2603.18908v5 Announce Type: replace
Abstract: Independently trained language models often learn compatible late-stage representations, despite differences in training objectives, architectures,...
By Matt Gorbett, Suman Jana
When a tool-using agent is given the same task in a different language, does it still take the same steps? Multilingual evaluation rarely asks: it compares final answers and discards the actions.
The study investigates whether monolingual language models, trained without joint multilingual objectives, develop cross-lingual alignment. By evaluating models such as Goldfish and independently built monolingual systems, the authors find that alignable representational geometry emerges across layers, strengthening with larger data, larger models, or closer linguistic proximity. A single Procrustes rotation on parallel sentences can map hidden states between models, and applying this rotation to a German model’s residuals swaps factual predictions to those of the donor English model, demonstrating functional transfer.
By Ej Zhou, Suchir Salhan, Catherine Arnett, Anna Korhonen
The paper introduces NinaXander, a method for composing frozen language models from different architecture families by inserting a trained shared‑latent adapter between their layers. By running the initial layers of one model, converting the intermediate representation with the adapter, and then continuing with the remaining layers of another model, multiple composed models can be created without retraining. Experiments with RWKV and Pythia show that while some compositions preserve syntactic quality and reduce memory usage, none match the parent model’s accuracy and language‑modeling performance drops on out‑of‑domain data.
By Takanori Kotama, Shun-ichiro Hayashi, Daichi Mukunoki, Tetsuya Hoshino, Takahiro Katagiri
arXiv:2610.19696v1 Announce Type: new
Abstract: Multi-agent systems allow language models to solve tasks collaboratively by sharing complementary information. These systems typically communicate thro...
By Xin Li, Mengbing Liu, Hao Jiang, Xin Gao, Annan Wang, Xingwei Qu, Yichi Zhang, Chau Yuen
arXiv:2607. 18026v1 Announce Type: new Abstract: Can large language models with substantially different parameter spaces be merged by direct weighted averaging, without training or semantic alignment?
By Jiahe Fan, Yinghao Hou, Si Chen, Aiyuan Zhang, Hong Xie, Defu Lian
arXiv:2609. 31181v1 Announce Type: new Abstract: Black-box model identification works by scoring a model's response to natural-language prompts.
By Nicol\'as Vera Z\'u\~niga
The study compares an English-only and a bilingual decoder-only model, each 310 M parameters, trained on eight diverse languages while controlling for English exposure, compute, and document overlap. After aligning on shared English vocabulary, the authors find that token embeddings appear similar, but the deeper hidden states used for prediction diverge across models. This hidden‑state mismatch grows through middle transformer layers and persists despite controls, indicating that contextual processing differs between the models.
"whyItMatters":"The findings show that embedding alignment can conceal significant internal representation differences, which is crucial for any downstream work that assumes aligned multilingual models are interchangeable."
By Anjishnu Mukherjee, Ziwei Zhu, Antonios Anastasopoulos
For Your Eyes Only: Evaluating Coordination Between Isolated Language Model Instances explores whether a language model can embed a signal in natural language that another independent instance can detect without shared memory or coordination training. The study introduces a cooperative signalling game where a Sender describes two words, one hidden, and a Receiver must identify the target. Seven contemporary models from four architectural families were tested on 300 word pairs, revealing that most struggle to coordinate when signals must be undetectable, though one frontier model performs near-perfectly even after filtering, and that models can also use this capability for deliberate misdirection.
By Alexander Shirnin, Aleksey Kudelya
arXiv:2606. 01060v1 Announce Type: cross Abstract: Preference alignment has substantially improved the observable behavior of large language models, yet it remains unclear what alignment changes internally.
By Partha Pratim Saha, Samarth Raina, Mayur Parvatikar, Amit Dhanda, Vinija Jain, Aman Chadha, Amitava Das
Manacá-1B is a 1.72‑billion‑parameter, open decoder‑only language model trained from scratch for Brazilian Portuguese, released with a fully containerized, reproducible training pipeline and complete logs. The authors evaluate it against nine open baselines on four Portuguese benchmarks, reporting standard errors and paired significance tests, and find that Manacá-1B outperforms smaller models on LAMBADA‑PT while remaining competitive on commonsense completion. They also uncover a tokenizer‑related evaluation pitfall that can drastically lower accuracy and provide a simple fix, releasing all code, logs, and corrected tokenizer for full reproducibility.
By Bruno Leonardo Santos Menezes, Carlos Leonardo Souza Cardoso, Fabio Andre Machado Porto