small language model

The Tailor's Reading

A feature of a sparse-autoencoder dictionary is published with one natural-language label, written by one language model from the passages that most excite the feature in one corpus. That label is a reading, and its provenance, the explainer, the explainer's system prompt, the corpus, and the model through which the co...

Reliability Benchmark for Small Language model

Algebra Before Bit: A Computer Scientist's Account of a Finite Record Universe

Imagine inspecting a quantum experiment through a programmer's console: what must be stored, what changes when a detector is read, and what can its records tell us? This is a representation, not a hypothesis: nothing here says the universe is a simulation, and nothing here depends on it being one. We examine a proposed...

Reasoning Budget Sweep: truncation and answer-production measurements for 12 small reasoning LLMs

A per-response measurement dataset covering 12 open reasoning language models (0.6B to 7B) on AIME 2024, AIME 2025 and MATH-500 across six generation token budgets (512 to 16384) and three seeds. 11,118 scored responses. Each row records the token count, whether the response was truncated at the budget, whether an answ...

Pressure-Induced Directional Asymmetry in LLM Procedural Gating: DF-Core v1.3 Benchmark and Replication Package

Summary Benchmarks for policy-adherent language agents typically split tasks into those where the agent must refuse and those where it must act, and report accuracy on each. Such designs measure the joint effect of task difficulty and of whatever conversational dynamics the benchmark contains, without separating them....

A discrete relational model built from local difference, rotation, and closure — k-Flip Theory (A–E)

Dear Researcher, I am sharing a working research archive titled k-Flip Theory, now available on Zenodo as a five-part set (A–E). The project began from a deliberately simple question: If two physical contributions cancel to zero, what exactly has disappeared? The working premise is that zero net value need not erase th...

Algebra Before Bit: A Computer Scientist's Account of a Finite Record Universe

Imagine inspecting a quantum experiment through a programmer's console: what must be stored, what changes when a detector is read, and what can its records tell us? This is a representation, not a hypothesis: nothing here says the universe is a simulation, and nothing here depends on it being one. We examine a proposed...

Reasoning Budget Sweep: truncation and answer-production measurements for 12 small reasoning LLMs

A per-response measurement dataset covering 12 open reasoning language models (0.6B to 7B) on AIME 2024, AIME 2025 and MATH-500 across six generation token budgets (512 to 16384) and three seeds. 11,118 scored responses. Each row records the token count, whether the response was truncated at the budget, whether an answ...

The Tailor's Reading

A feature of a sparse-autoencoder dictionary is published with one natural-language label, written by one language model from the passages that most excite the feature in one corpus. That label is a reading, and its provenance, the explainer, the explainer's system prompt, the corpus, and the model through which the co...

Noēsis: Deterministic-First Retrieval with Two-Tier Context Hydration for Factuality-Critical Queries on Small Local Models

A wrong number is worse than no answer. Across factuality-critical domains—audience metrics, scheduling and rights in media; dosages and lab values in healthcare; figures and citations in finance and legal—a confident but fabricated value is more damaging than an honest admission of uncertainty. Yet this is the dominan...

Heterogeneous signal modeling and Spectral-Peephole LSTM architecture for resilient structural health monitoring in complex engineering systems

Structural health monitoring systems play a significant role in the early detection of nonlinear damage mechanisms caused by seismic activity and structural fatigue. Damage occurring in modern structures typically leads to complex deviations in time-series data and distortions in the fundamental frequency components th...

Accelerated LLM: A Fuzzy-Logic-Augmented Router Architecture for Efficient Multi-Domain Query Processing via Specialised Small Language Models

Accelerated LLM, a modular architecture that replaces a single general-purpose LLM with an ensemble of task-specialised small language models (SLMs) governed by a neural query router and a Mamdani fuzzy inference system, achieves competitive or superior task-specific performance at a fraction of the parameter count.