Everything about post-training small models
124 explainers on fine-tuning, distillation, evaluation and deployment, grouped by topic.

Knowledge Distillation
How knowledge distillation transfers the capability of a large language model into a small one: teachers, students, and what makes the transfer work.
Fine-Tuning Small Language Models
What actually changes when you fine-tune a model: LoRA and adapters, hyperparameters that matter, and the failure modes that make a fine-tuned model worse than the base.
Small Language Models and Model Selection
Which small language model to fine-tune: how the 270M to 8B range differs in practice, and what the benchmark data says about picking a base model.
Task Types for Small Language Models
The two task types a small language model can be trained for, question answering and chat completion, and how classification, RAG and tool calling map onto them.
Training Data for Small Language Models
Three ways to get training data for a task-specific small language model: from your live traffic, from production traces you already have, or from a minimal hand-written dataset.
Evaluating Small Language Models
How to tell whether a fine-tuned small language model is good enough to ship: teacher evaluation, choosing metrics, building a test set, and why eval scores mislead.
Deploying Small Language Models
Running a fine-tuned small language model in production: llama.cpp, vLLM and Ollama, quantization formats, VRAM requirements, and on-device inference.
Cost and ROI of Small Language Models
What it actually costs to run a task-specific small language model versus a frontier API: inference economics, training cost, and how to calculate the return.
Alternatives to Training a Small Language Model
When to fine-tune, when to use retrieval, when to just write a better prompt, and when a task-specific small model is the wrong tool entirely.
Small Language Model Glossary
Short definitions of the terms that come up when training and deploying small language models: logits, KV cache, quantization, LoRA rank, and the rest.

Build your first specialist model
Start on the platform today, or talk to us about your use case.