Saved in:
Bibliographic Details
Main Authors: Dao, Alan, Vu, Dinh Bach, Nguyen, Alex, Buppodom, Norapat
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.00360
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916875081875456
author Dao, Alan
Vu, Dinh Bach
Nguyen, Alex
Buppodom, Norapat
author_facet Dao, Alan
Vu, Dinh Bach
Nguyen, Alex
Buppodom, Norapat
contents Small language models (SLMs) are inherently limited in knowledge-intensive tasks due to their constrained capacity. While test-time computation offers a path to enhanced performance, most approaches treat reasoning as a fixed or heuristic process. In this work, we propose a new paradigm: viewing the model's internal reasoning, delimited by <think> and </think> tags, as a dynamic task vector machine. Rather than treating the content inside these tags as a mere trace of thought, we interpret the generation process itself as a mechanism through which the model \textbf{constructs and refines its own task vectors} on the fly. We developed a method to optimize this dynamic task vector machine through RLVR and successfully trained an agentic web-search model. We present Lucy, a 1.7B-parameter SLM that leverages this dynamic reasoning mechanism with MCP integration to achieve 78.3% accuracy on the SimpleQA benchmark, performing on par with much larger models such as DeepSeek-V3. This demonstrates that small models can rival large ones when equipped with structured, self-constructed task reasoning.
format Preprint
id arxiv_https___arxiv_org_abs_2508_00360
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Lucy: edgerunning agentic web search on mobile with machine generated task vectors
Dao, Alan
Vu, Dinh Bach
Nguyen, Alex
Buppodom, Norapat
Computation and Language
Small language models (SLMs) are inherently limited in knowledge-intensive tasks due to their constrained capacity. While test-time computation offers a path to enhanced performance, most approaches treat reasoning as a fixed or heuristic process. In this work, we propose a new paradigm: viewing the model's internal reasoning, delimited by <think> and </think> tags, as a dynamic task vector machine. Rather than treating the content inside these tags as a mere trace of thought, we interpret the generation process itself as a mechanism through which the model \textbf{constructs and refines its own task vectors} on the fly. We developed a method to optimize this dynamic task vector machine through RLVR and successfully trained an agentic web-search model. We present Lucy, a 1.7B-parameter SLM that leverages this dynamic reasoning mechanism with MCP integration to achieve 78.3% accuracy on the SimpleQA benchmark, performing on par with much larger models such as DeepSeek-V3. This demonstrates that small models can rival large ones when equipped with structured, self-constructed task reasoning.
title Lucy: edgerunning agentic web search on mobile with machine generated task vectors
topic Computation and Language
url https://arxiv.org/abs/2508.00360