# SmallThinker: A Family of Efficient Large Language Models Natively Trained for Local Deployment

arXiv: https://arxiv.org/abs/2507.20984 · HTML: https://mlsanity.com/paper/2507.20984/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning.
- Builds on 15 of its 21 references that are in the map.
- First posted on arXiv 28 Jul 2025.

## Placed in

- LLM reasoning / [Test-time compute and efficient thinking](https://mlsanity.com/lr/test-time-compute-and-efficient-thinking/)
- Mixture-of-experts / [Efficient MoE training and inference systems](https://mlsanity.com/moe/efficient-moe-training-and-inference-systems/)

## Builds on (most cited first, 15 of 15)

- [Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities](https://mlsanity.com/paper/2507.06261/) (2025)
- [DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models](https://mlsanity.com/paper/2401.06066/) (2024)
- [Scaling Synthetic Data Creation with 1,000,000,000 Personas](https://mlsanity.com/paper/2406.20094/) (2024)
- [SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model](https://mlsanity.com/paper/2502.02737/) (2025)
- [Nemotron-CC: Transforming Common Crawl into a Refined Long-Horizon Pretraining Dataset](https://mlsanity.com/paper/2412.02595/) (2024)
- [OpenCoder: The Open Cookbook for Top-Tier Code Large Language Models](https://mlsanity.com/paper/2411.04905/) (2024)
- [PowerInfer-2: Fast Large Language Model Inference on a Smartphone](https://mlsanity.com/paper/2406.06282/) (2024)
- [Nemotron-H: A Family of Accurate and Efficient Hybrid Mamba-Transformer Models](https://mlsanity.com/paper/2504.03624/) (2025)
- [ReLU2 Wins: Discovering Efficient Activation Functions for Sparse LLMs](https://mlsanity.com/paper/2402.03804/) (2024)
- [MegaMath: Pushing the Limits of Open Math Corpora](https://mlsanity.com/paper/2504.02807/) (2025)
- [Turbo Sparse: Achieving LLM SOTA Performance with Minimal Activated Parameters](https://mlsanity.com/paper/2406.05955/) (2024)
- [Sparsing Law: Towards Large Language Models with Greater Activation Sparsity](https://mlsanity.com/paper/2411.02335/) (2024)
- [SWAN-GPT: An Efficient and Scalable Approach for Long-Context Language Modeling](https://mlsanity.com/paper/2504.08719/) (2025)
- [Unearthing Large Scale Domain-Specific Knowledge from Public Corpora](https://mlsanity.com/paper/2401.14624/) (2024)
- [Reformulation for Pretraining Data Augmentation](https://mlsanity.com/paper/2502.04235/) (2025)

## Built on by (most cited first, 5 of 5)

- [FRISM: Fine-Grained Reasoning Injection via Subspace-Level Model Merging for Vision-Language Models](https://mlsanity.com/paper/2601.21187/) (2026)
- [Resource-Aware Neuro-Symbolic Reasoning for Local Small Language Models](https://mlsanity.com/paper/2606.27281/) (2026)
- [Retrieval-Augmented Large Language Models as Components of Cognitive Computing architecture for Regulatory Knowledge Management](https://mlsanity.com/paper/2607.24352/) (2026)
- [Budgeting Bytes: A Windowed Storage Roofline and Dual-Budget Architecture Ablations for Storage-Bound LLM Decoding](https://mlsanity.com/paper/2609.04238/) (2026)
- [Cascadia: A Control-Plane-Free Alternative to Hyperconverged AI Infrastructure](https://mlsanity.com/paper/2609.38697/) (2026)

## Abstract (arXiv)

While frontier large language models (LLMs) continue to push capability boundaries, their deployment remains confined to GPU-powered cloud infrastructure. We challenge this paradigm with SmallThinker, a family of LLMs natively designed - not adapted - for the unique constraints of local devices: weak computational power, limited memory, and slow storage. Unlike traditional approaches that mainly compress existing models built for clouds, we architect SmallThinker from the ground up to thrive within these limitations. Our innovation lies in a deployment-aware architecture that transforms constraints into design principles. First, We introduce a two-level sparse structure combining fine-grained Mixture-of-Experts (MoE) with sparse feed-forward networks, drastically reducing computational demands without sacrificing model capacity. Second, to conquer the I/O bottleneck of slow storage, we design a pre-attention router that enables our co-designed inference engine to prefetch expert parameters from storage while computing attention, effectively hiding storage latency that would otherwise cripple on-device inference. Third, for memory efficiency, we utilize NoPE-RoPE hybrid sparse attention mechanism to slash KV cache requirements. We release SmallThinker-4B-A0.6B and SmallThinker-21B-A3B, which achieve state-of-the-art performance scores and even outperform larger LLMs. Remarkably, our co-designed system mostly eliminates the need for expensive GPU hardware: with Q4_0 quantization, both models exceed 20 tokens/s on ordinary consumer CPUs, while consuming only 1GB and 8GB of memory respectively. SmallThinker is publicly available at hf.co/PowerInfer/SmallThinker-4BA0.6B-Instruct and hf.co/PowerInfer/SmallThinker-21BA3B-Instruct.