Searcharxiv⌕ Search

arXiv subjects

Nguyen Van Vinh

Publications and source records attributed to Nguyen Van Vinh.

7 recordsLinked to original sources
arXiv2026

Bridging the Reasoning Gap in Vietnamese with Small Language Models via Test-Time Scaling

The democratization of ubiquitous AI hinges on deploying sophisticated reasoning capabilities on resource-constrained devices. However, Small Language Models (SLMs) often face a "reasoning gap", particularly in non-English languages like Vietnamese, where they struggle to maintain coherent chains of thought. This paper investigates Test-Time Scaling strategies for the Qwen3-1.7B architecture within the context of Vietnamese Elementary Mathematics. We introduce Vi-S1K, a high-fidelity reasoning dataset localized via a Gemini 2.5 Flash-Lite powered pipeline, and Vi-Elementary-Bench, a dual-resource benchmark for rigorous evaluation. Using an LLM-as-a-Judge protocol, we reveal that the base model possesses robust latent knowledge (Accuracy: 4.05/5.00) but suffers from a severe "formatting gap" in communication. Supervised Fine-Tuning (SFT) acts as a critical "reasoning unlocker", yielding a 77% improvement in Explanation Quality and bridging the gap between raw calculation and pedagogical coherence. Furthermore, our analysis of prompting strategies uncovers a significant trade-off: structured frameworks like ReAct impose a "cognitive tax" on the 1.7B parameter capacity, degrading performance relative to pure Chain-of-Thought (CoT) combined with Self-Consistency. These findings establish a deployment hierarchy for SLMs, demonstrating that SFT combined with simplified test-time scaling is superior to complex agentic workflows for edge-based reasoning.

cs.CL↗
arXiv2025

Auto-Prompting with Retrieval Guidance for Frame Detection in Logistics

Prompt engineering plays a critical role in adapting large language models (LLMs) to complex reasoning and labeling tasks without the need for extensive fine-tuning. In this paper, we propose a novel prompt optimization pipeline for frame detection in logistics texts, combining retrieval-augmented generation (RAG), few-shot prompting, chain-of-thought (CoT) reasoning, and automatic CoT synthesis (Auto-CoT) to generate highly effective task-specific prompts. Central to our approach is an LLM-based prompt optimizer agent that iteratively refines the prompts using retrieved examples, performance feedback, and internal self-evaluation. Our framework is evaluated on a real-world logistics text annotation task, where reasoning accuracy and labeling efficiency are critical. Experimental results show that the optimized prompts - particularly those enhanced via Auto-CoT and RAG - improve real-world inference accuracy by up to 15% compared to baseline zero-shot or static prompts. The system demonstrates consistent improvements across multiple LLMs, including GPT-4o, Qwen 2.5 (72B), and LLaMA 3.1 (70B), validating its generalizability and practical value. These findings suggest that structured prompt optimization is a viable alternative to full fine-tuning, offering scalable solutions for deploying LLMs in domain-specific NLP applications such as logistics.

cs.CL↗
arXiv2024

An Exactly Solvable Dual Bus Route Model

In 1998, O'Loan et al. introduced a simplified bus route model to illustrate bus dynamics. However, fluctuations in passenger numbers make it challenging to achieve an exact solution for the model's stationary state, as these fluctuations can impact bus behavior. In this study, we present an exactly solvable model for a dual bus route system that builds upon O'Loan et al.'s model. This dual model allows us to comprehensively analyze bus route dynamics. Our model introduces additional parameters not previously considered by O'Loan et al. to account for neighboring effects and align with an observation in the original model, which can influence the average stationary current and velocity of buses. When the neighboring effect is weak, our model behaves similarly to O'Loan et al.'s bus route model. However, with a strong neighboring effect, our model exhibits intriguing characteristics. Additionally, in some limiting cases, our model recovers well-known models such as the simple exclusion process, its generalization, the cooperative exclusion process, and the RNA polymerase model during the elongation stage.

cond-mat.stat-mech↗
arXiv2024

An Exactly Solvable Model for Single-Lane Unidirectional Ant Traffic

In 2002, Chowdhury et al. introduced a simplified model aimed at depicting the dynamics of single-lane unidirectional ant traffic. Despite efforts, an exact solution for the stationary state of this ant-trail model remains elusive. The primary challenge arises from inherent fluctuations in the total amount of pheromone along the ant-trail. These fluctuations significantly influence ant movement. Consequently, unlike conventional models involving driven interacting particles, the average velocity exhibits non-monotonic behavior with ant density. In this study, we propose an exactly solvable model for ant traffic whose dynamics is based on the one in Chowdhury et al.'s model, and then discuss the circumstances under which the non-monotonic trend of average velocity arises.

cond-mat.stat-mech↗
arXiv2022

ViNMT: Neural Machine Translation Toolkit

We present an open-source toolkit for neural machine translation (NMT). The new toolkit is mainly based on vaulted Transformer (Vaswani et al., 2017) along with many other improvements detailed below, in order to create a self-contained, simple to use, consistent and comprehensive framework for Machine Translation tasks of various domains. It is tooled to support both bilingual and multilingual translation tasks, starting from building the model from respective corpora, to inferring new predictions or packaging the model to serving-capable JIT format.

cs.CL↗