Learn
Practical guides to fine-tuning, distillation, and deploying small language models.
LoRA vs Full Fine-Tuning: When to Use What
Compare LoRA and full fine-tuning for small language models. Learn the trade-offs in accuracy, speed, and memory so you can pick the right approach for your project.
Base Model vs Fine-Tuned Model: How to Compare Them
How to measure what fine-tuning actually bought you: get a base-model score on the same held-out set, train, read both numbers against the teacher, and check the comparison is fair before believing it.
Build vs Buy: Training Your Own SLM
Building means owning synthetic data generation, hyperparameter search and evaluation yourself. A stage-by-stage breakdown of what each path costs and the cases where building is clearly right.
Building a Test Set That Catches Real Failures
A step-by-step recipe for constructing a held-out evaluation set from production traffic: reserving it before the pipeline sees it, seeding it with known failures, and baselining the system you're replacing.
Calculating the ROI of a Task-Specific Model
A five-step procedure for turning a task-specific model into a break-even volume, using real CLI commands and a free evaluation gate before you commit any budget.
Deploy an SLM with Ollama
Loading a fine-tuned small language model into Ollama from a local GGUF or straight from Hugging Face, writing a Modelfile that preserves the training-time system prompt, and the chat-template trap specific to this runtime.
Deploy an SLM with vLLM
Serving a distilled small language model on a GPU with vLLM: environment setup, the tool-calling parser flags, querying the OpenAI-compatible endpoint, and the throughput and memory figures we measured on a single H100.
Do You Actually Need a Bigger Model?
Sometimes yes: broad reasoning and free-form generation still favour scale. But on bounded tasks the published rankings invert after fine-tuning, and this guide separates the two cases.
Fine-Tune, RAG, or Prompt: Which Should You Use?
Three techniques that fix three different problems: missing facts, wrong behaviour, and an under-specified task. A criteria table for telling them apart before you commit engineering time.