QUELL — reproducibility package (v4.0). This repository accompanies the paper “QUELL: An On-Device Benchmark of Quantized Edge LLMs versus Classical Machine Learning for IoT Intrusion Detection.” QUELL fine-tunes decoder-only LLMs (GPT-2-medium, Qwen2.5-0.5B/1.5B, Llama-3.2-1B) on network-flow records serialized as text, quantizes the Qwen2.5-1.5B detector, deploys it on an NVIDIA Jetson Orin Nano Super (GGUF/llama.cpp), and compares it with random-forest and XGBoost detectors on Edge-IIoTset, CICIoT2023 and N-BaIoT under a leakage-controlled, multi-seed protocol. Axes: detection quality, on-device latency/energy/memory, quantization retentio
Şüyun, Süleyman Burçin, Aktürk, Fatih
Source code for the multimodal LLM evaluation pipeline described in "Multimodal Large Language Models for Clinical Scar Assessment: Agreement, Bias, and Reproducibility." Implements a five-stage pipeline (image ingestion, preprocessing, multi-provider API dispatch, JSON schema validation, and Excel/CSV export) used to evaluate five MLLMs (GPT-5.4, Claude Sonnet 5, Gemini 3.1 Pro, Qwen3.7-Plus, GLM-5V-Turbo) against blinded human observer scores on the POSAS Observer Scale. Ships without patient images, API keys, or run outputs; researchers supply their own image dataset and API credentials.
Abdullah, Ahmed
Reproducibility package for the Med3DVLM (MagicXin/Med3DVLM-Qwen-2.5-7B) clinical evaluation. Contains the preprocessing scripts (NIfTI-to-PNG conversion, PNG-to-prepared-NIfTI, mediastinal windowing transformation), the HPC inference pipelines, and the 15 evaluation prompts (5 closed pathology, 5 closed technical, 5 open report generation). Images are deposited separately: ChestPathCT5-S100 (doi:10.5281/zenodo.18256797). Local paths and HPC credentials have been anonymized as XXXXX placeholders.
Fahrni, Guillaume
AKBASCORE AI TECHNOLOGY — MAM SILICON HIPPOCAMPUS v1.4 Native Internal Tensor Memory Injection, Append-Only Cognitive Cartridge Consolidation and Multi-Hop Relational Synthesis Mustafa Akbaş | AkbasCore AI Teknoloji | Mersin, Türkiye | October 2026 A frozen language model can receive independently constructed numerical memory cartridges, consolidate their internal attention states, and answer relational questions without the original source facts being supplied again as text. AKBASCORE MAM v1.4 demonstrates this capability using Qwen2.5-7B-Instruct with unchanged model weights, no fine-tuning, no LoRA and no optimizer. The central technical d
ceceli33
Initial public release of the analysis code associated with the study: Explainable Transport Mode Choice Prediction: Integrating GIS, Machine Learning, Cross-Context Benchmarking, and LLM-Based Explanations This release includes: Machine-learning model benchmarking and SHAP analysis Qwen3-8B evidence-grounded explanations Sensitivity analysis excluding context-specific variables Study area: Pabna Municipality, Bangladesh.
Asif Rahman
Analysis code, development-cohort intermediate predictions, and machine-readable results supporting the manuscript "Fairness auditing of machine-learning and LLM stroke-risk estimation: development in NHANES 2007-2018 and external validation in CHARLS 2011". Contents: (1) full analysis pipeline scripts covering cohort construction, four-model training with subgroup fairness audits (LR/RF/GBM/MLP plus reweighing and adversarial debiasing), zero-shot LLM vignette audits of Qwen2.5-1.5B/7B, external validation in CHARLS 2011, and QLoRA distillation; (2) out-of-fold predictions for all tabular models on the NHANES development cohort (n = 34,719),
Xu, Wei, Guo, Yun-Ying
