arXiv subjects

Tsz Chung Cheng

Publications and source records attributed to Tsz Chung Cheng.

3 recordsLinked to original sources

arXiv2026

CantoneseLLM v2: Reasoning in a Low-Resource Language

Cantonese is widely spoken but remains low-resource in written data, with no large corpus of native Cantonese reasoning traces available for model training. We develop and release CantoneseLLM v2, comprising models based on Qwen3 8B and 30B-A3B. The models are trained through CPT on 784 million Cantonese and Hong Kong-related tokens, chat-vector merging, SFT, DPO, and RLVR. Evaluation across the training stages shows that chat-vector merging transfers instruction following but preserves the donor model's reasoning language, while SFT with limited Cantonese reasoning data substantially shortens or removes reasoning traces and reduces benchmark performance. DPO restores the reasoning-block format, particularly for the 8B model, but recovers only part of the lost performance. The RLVR training with Cantonese language and Traditional Chinese scripts as multiplicative constraints introduced Cantonese language alignment and restored the lost performance. The 30B-A3B model reaches 73.16 on HKCanto-Eval, within 1.20 points of its merged checkpoint, while retaining the Cantonese reasoning behaviour absent from that checkpoint. We release the model checkpoints, the training environments, and a thirteen-year Traditional Chinese Common Crawl dataset. The models can be accessed at https://huggingface.co/collections/hon9kon9ize/cantonesellm-v20

cs.CL↗

arXiv2026

Magnum.np.distributed: Accelerating Finite Difference Micromagnetic Simulations with Multiple GPUs

Micromagnetic simulations are essential tools in nanomagnetism and spintronics research. Although widely adopted solvers like Mumax3 and the Python-native magnum.np use GPU acceleration to improve performance, these tools are limited to single-device computation. In this work, we present the first Python-native multi-GPU micromagnetic framework by extending magnum.np with PyTorch Distributed. This leverages high-speed communication and computation across multiple GPUs while retaining the benefits of ease of installation, platform-agnostic design, and compatibility with Python. For computationally intensive demagnetisation effective-field calculations, we achieve a 7.0x speedup across 8 GPUs connected via NVLink, whereas Halo exchange required for Heisenberg exchange shows limited scaling due to kernel dispatch latency. We also demonstrated the framework's versatility by achieving a 6.8x speedup in demagnetisation field computation on CPU with NUMA pinning via the MPI backend of PyTorch Distributed. Faster turnaround times will enable researchers to explore larger, more complex systems and accelerate the design cycle for novel spintronic devices.

cs.DC↗

arXiv2025

HKCanto-Eval: A Benchmark for Evaluating Cantonese Language Understanding and Cultural Comprehension in LLMs

The ability of language models to comprehend and interact in diverse linguistic and cultural landscapes is crucial. The Cantonese language used in Hong Kong presents unique challenges for natural language processing due to its rich cultural nuances and lack of dedicated evaluation datasets. The HKCanto-Eval benchmark addresses this gap by evaluating the performance of large language models (LLMs) on Cantonese language understanding tasks, extending to English and Written Chinese for cross-lingual evaluation. HKCanto-Eval integrates cultural and linguistic nuances intrinsic to Hong Kong, providing a robust framework for assessing language models in realistic scenarios. Additionally, the benchmark includes questions designed to tap into the underlying linguistic metaknowledge of the models. Our findings indicate that while proprietary models generally outperform open-weight models, significant limitations remain in handling Cantonese-specific linguistic and cultural knowledge, highlighting the need for more targeted training data and evaluation methods. The code can be accessed at https://github.com/hon9kon9ize/hkeval2025

cs.CL↗