#small language models
Small Language Models: 7 AI articles covering small language models news, analysis, and research
Articles
Liquid AI Releases LFM3: Six Open-Weight 2B-Class Multimodal⭐8
Liquid AI's LFM3 offers six open-weight 2B-class multimodal models on one shared backbone, running on phones and laptops at ~1GB memory with 256K context.
A Small Model Asks Better: Fine-Tuning Query Clarification for⭐9
ClariEdge, a new dataset of ~3,900 examples, shows a lightweight fine-tuned LoRA detects ambiguous RAG queries with 96.6% accuracy and generates better clarific...
How Specializing the Vocabulary of SLMs Can Cut Inference Cost⭐10
Specializing SLM vocabularies for domain-specific applications cuts token counts and inference cost by 77%. Learn how tokenizer retraining and embedding resizin...
Embeddings Are Overpriced? A Frozen 16 MB Vocabulary for Small⭐8
A frozen 16 MB embedding table for small and ternary language models matches learned embeddings on LAMBADA, HellaSwag, and PIQA while cutting parameters.
ChestPheNoT: Deployable, Auditable Label-Status-Evidence⭐9
ChestPheNoT extracts finding labels, three-class status, and verbatim evidence spans from radiology reports using a locally deployable, auditable 0.5–3B model.
Do Small Language Models Know What They Don't Know? New Research⭐9
New research shows token entropy fails as a confidence signal in sub-3B language models, while semantic entropy enables routing gains up to +50 percentage point...
PrismML's Tiny LLM Could 'Change How We All Use AI' — Here's Why⭐9
PrismML compresses reasoning LLMs to run on PCs and phones, shrinking a 27B model to under 6 GB — a 9x-10x memory reduction that could reshape AI.