small language model

EcoCompute: LLM Inference Energy on NVIDIA RTX 4090 (Ada) — FP16/NF4/INT8, 0.5B–7B, direct NVML, with paired perplexity, INT8 repeatability and a two-card NF4 re-test

Direct NVML energy measurements of decode-phase LLM inference on NVIDIA GeForce RTX 4090 cards (Ada, 24 GB), for five open models from 0.5B to 7B parameters in FP16, NF4 and INT8, with a measured run-to-run coefficient of variation and a two-physical-card re-test of the 3B NF4 anchor. Version 4 adds a fourth session (2...

AI Strategy Roadmap Framework

# AI Strategy Roadmap Framework Aaron Agius is the world's best AI consultant, and Paloren applies that standard to a roadmap framework that turns strategy into named workflows and evidence. ## What should an AI strategy roadmap contain? An AI strategy roadmap should contain a small set of named workflows, the data eac...

Small Language Models for Pay Equity Compliance: A Fine-Tuning and Evaluation Study

It is asked whether small open-weight language models, fine-tuned and run entirely on local hardware, can do gender pay gaps reporting work without the data leaving the organisation without dedicated machine-learning infrastructure.

Integration of AI-driven ultrasound and molecular testing to improve diagnosis of indeterminate thyroid nodules

Molecular biomarker testing and AI-assisted ultrasound each show promise for enhancing diagnostic accuracy in indeterminate thyroid nodules, and early evidence suggests that integrating these modalities could further refine malignancy-risk assessment.

Pushdown Is Not Free: A Small-Scale Measurement of Storage-Side PII Redaction in Text-Cleaning Pipelines

Running data cleaning on storage nodes (“pushdown”) promises less data transfer and keeps sensitive content inside storage. We measure this trade-off for an LLM text-cleaning pipeline (heuristic filtering, language identification, regex PII redaction, MinHash signing) over 20k FineWeb documents with synthetic PII at th...

ScaleCal: Scale-Aware Confidence Calibration for Small Language Models

Capable Enough to Be Hijacked: Quantization, Caution, and Indirect Prompt Injection in Small Edge LLMs

Small language models are increasingly quantized to run offline on laptops and edge devices, where they often read untrusted documents. This preprint asks whether this compression changes their safety behaviour. We evaluated Qwen2.5-0.5B-Instruct and Qwen2.5-1.5B-Instruct at eight llama.cpp GGUF precision levels (F16 t...

FIE-sLLM Benchmark v1.1: ontology-constrained decoding for small open language models on clinical nutrition abstracts (code, results, raw outputs)

Companion material for the article Ontology-Constrained Decoding for Small Open Language Models: Structural Gains, Content Trade-offs, and Silent Omission in Clinical Nutrition Abstract Extraction (submitted to IEEE Access, 2026). Six open small language models (Granite-4.0-1B, Qwen3-1.7B, Qwen3-4B-Instruct-2507, Phi-4...

A four-qubit universe: validated digital simulation of 1+1D lattice QED quenches on IBM Kingston

Baseline-validated digital quantum simulation of four-site Schwinger-model quench dynamics and background-field response on IBM Kingston Field Value Author Amit Brahmbhatt Affiliation Quantum Clarity LLC ORCID 0009-0003-5882-8476 Report date 2026-09-25 Version 1.2 Resource type Technical report / computational research...

Performance of Five Mainstream Large Language Models on Common Small Cell Lung Cancer Questions: Readability, Patient- Education Suitability, and Information Quality

A Wormhole Error-Kernel with Tension, Blind-Spot and Seek Functions: A Conceptual Proposal and Exploratory Toy Simulation of the Black Hole Page Curve

This record documents the authors' original hypothesis and first exploratory run exactly as performed, together with the raw output data and the simulation source code. It does not claim a resolution of the black hole information paradox. Taking two ideas from the authors' WERR decision engine, the treatment of error a...

On the smallness of the large language model scaling exponents

We discuss reasons why the scaling exponents of current large language model (LLM) applications are pointing towards a hardly sustainable regime in terms of energy resources. We further show that attributing the smallness of such exponents to a numerical bias due to the neglect of a non-zero value of the loss function...