Find a story
Search Spins
Search titles, summaries, and missing voices across published articles — press releases, announcements, and media coverage.
4 results for “small language models”
MIITA: Memory-Induced Inference-Time Adaptation for Continual Learning with Small Language Models
MIITA is a new inference-time adaptation framework designed to enable continual learning in small language models without catastrophic forgetting, using memory-based semantic retrieval and gated hidden-state updates.
Jul 28, 2026
Task Competence Is Not Instruction Following: Evaluating Instruction-Conflicting Behavior in Small Language Models
A research paper on arXiv demonstrates that small instruction-tuned language models often ignore conflicting instructions while maintaining high task accuracy, revealing a fundamental decoupling between task competence and instruction following.
Jul 23, 2026
Dispersion loss counteracts embedding condensation in small language models
A Reddit post shares a technical observation about dispersion loss mitigating embedding condensation in small language models, highlighting an internal model behavior without empirical validation or real-world application context.
Published Jul 4, 2026 · Analyzed Jul 6, 2026
Dispersion loss counteracts embedding condensation in small language models
A technical observation about dispersion loss mitigating embedding condensation in small language models was posted as a comment on Hacker News, generating community discussion but lacking original research documentation or empirical validation.
Published Jul 3, 2026 · Analyzed Jul 6, 2026