News Hub

Distil Labs compared large models versus fine-tuned small models in a production pipeline.

Three deployment methods exist for putting small models into production environments.

Distil Labs enables training and deployment of small models in 30 minutes via CLI or coding agent.

The best mid-size model to fine-tune was the third-best model to prompt.

Gemini 2.5 Flash-Lite model retirement date announced.

Study examines whether base model accuracy predicts fine-tuned performance using mid-size mixture-of-experts benchmark.

Agent Distillation with dltHub trains smaller models using traces from existing agents.

Distil Labs describes using small language models to enhance large language model capabilities.

Distil Labs released an autonomous bug fixing agent.

Distil Labs offers Claude skill to train small language models from production traces.

Distil Labs released Fine-Tuning Liquid.

Article compares small language models for fine-tuning applications.

A 0.6 billion parameter model outperformed a 120 billion parameter LLM by 29 points on a benchmark.

Title incomplete; cannot summarize without full content.

Distil Labs published guidance on combining expert model optimization with scalable GPU infrastructure for production language models.

Knowunity reduced their LLM costs by 68 percent using Distil Labs technology.

Distil Labs describes how to locally label emails using their fine-tuned model with n8n automation.

Small language models can enable retrieval-augmented generation on devices.

FunctionGemma enables multi-turn tool calling with 270 million parameters.

Article examines when reinforcement learning improves small language model performance.