small language model

Hunk-Constrained DPO: Segment-Level Optimization for Secure and Correct LLM Code Generation

Hunk-Constrained Direct Preference Optimization is introduced, a training framework that unifies security hardening and functional correction in large language models and demonstrates that HPO achieves substantial security improvements—up to 28 percentage points—while preserving or enhancing functional correctness.

Solve the Fall Time Backwards for the Curve and Only One Remains ── Stand in the direction that gives T(h) and solves for the curve, and requiring T to be independent of h yields f'(y) proportional to 1/sqrt(y) uniquely ── measured, T(0.5)=T(1)=T(2)=0.708762 with a spread of 0.000x10^0, matching the closed form pi/sqrt(2g)=0.709373 ── for a general f', T proportional to sqrt(h), and quadrupling h multiplies T by 2.000000 ── [Paper 488]

The isochronism of the cycloid is taken to be a lucky property of that curve. This paper shows that reversing the direction makes isochronism fix the curve uniquely──requiring T to be independent of h alone yields f'(y) proportional to 1/sqrt(y). No new theorem or law is claimed. Scope of this paper (scope note): No ne...

A Lagrange Multiplier Is a Temperature Here and a Pressure Next Door ── The multiplier beta from maximising entropy agrees with an independently computed dS/dE at 0.700000000 against 0.700000000, and again at 1.900000000 for beta=1.9 ── it does not agree with a wrong value of 0.9 ── the same device appears in Paper 196 as the multiplier enforcing incompressibility, where it is the pressure: what the multiplier is, the constraint decides ── [Paper 494]

A Lagrange multiplier is taken to be an auxiliary quantity that appears mid-calculation. This paper shows that the multiplier agrees digit for digit with an independently defined physical quantity──and that which quantity it becomes is decided by the constraint imposed. No new theorem or law is claimed. Scope of this p...

Extraction dataset for a systematic review of reasoning enhancement approaches for small language models.

Continuous in Composition, Discontinuous in Whether the Species Are the Same ── The entropy of mixing 2 mol of two components is 11.526293 J/K at mole fraction 0.5, 0.931245 at 0.01 and 0.000246 at 10^-6, approaching zero continuously ── yet for two samples of the same species it jumps to 0.000000 J/K without passing along that path ── the variable that was varied is composition; the variable that jumped is distinguishability ── [Paper 498]

The Gibbs paradox is said to be the problem of an entropy that jumps discontinuously. This paper shows that the variable that jumps and the variable that is continuous are different ones──composition stays continuous down to 10^-6, and what jumps is distinguishability. No new theorem or law is claimed. Scope of this pa...

Topological Cross-Domain Transfer in Autonomous Systems: Structural Isomorphism Training

Structural isomorphism training for cross-domain transfer in small language models, validated via the gnarp model lineage and Self-Falsifying Orchestrator. One of 135 dated defensive publications in the CatQualia record. Collection DOI: https://doi.org/10.5281/zenodo.22751107 Original publication: https://catqualia.com...

Characterizing LLM scientific concept generation: A multi-dimensional measurement study

A multi-dimensional characterization framework analyzing 9,666 outputs from seven models spanning 2020–2024 provides calibrated measurements and validated methodology for future research without making interpretive claims about novelty.

Rethinking Scale: The Efficacy of Fine-Tuned Open-Source LLMs in Large-Scale Reproducible Social Science Research

Large language models (LLMs) are distinguished by their architecture, parameter size, and performance capabilities. Social scientists have increasingly adopted LLMs to automate high-dimensional text classification tasks which are difficult to scale with human coders. Very large, closed-source models often deliver super...

PENGEMBANGAN MEDIA PETA INTERAKTIF BERBASIS STEAM BERBANTUAN GENIALLY PADA PEMBELAJARAN IPAS DI SEKOLAH DASAR

ABSTRACT Learning Social and Natural Sciences (IPAS) on the topic of the Association of Southeast Asian Nations (ASEAN) in fifth-grade elementary schools is still predominantly conventional and teacher-centered, limiting students’ opportunities to actively and meaningfully explore the geographical, economic, and cultur...

Evaluating large language models for domain-specific scholarly metadata classification

The findings indicate that open-source LLMs can support domain-specific scholarly metadata classification without task-specific fine-tuning, however, their moderate and dimension-dependent performance limits their suitability for fully automated fine-grained metadata enrichment.

Semantic labeling with fulfillers in multilingual contexts

This article introduces a novel approach to semantic labeling grounded in the fulfiller concept—a cognitive-semantic theory that highlights the role of absent but contextually inferred elements in meaning construction. We argue that the prevailing semantic labeling practices often neglect this essential aspect of m...

Improving Inference Efficiency in Deep Neural Networks

The increasing complexity of modern deep neural networks (DNNs), particularly Transformer architectures such as large language models (LLMs) and diffusion models, has made inference efficiency a central challenge across edge and datacenter platforms. The two regimes, however, impose sharply different constraints. At th...