Home / Glossary / Base model

What is Base Model?

A base model is a pre-trained LLM before any task-specific fine-tuning or RLHF. osFoundry treats base models as the starting point for fine-tuning (LoRA / QLoRA) and as direct chat backends when no specialised behaviour is needed.

Detail

Base models are produced by an initial training run on a large text corpus. They predict next tokens but typically don’t follow instructions well — that’s what fine-tuning adds. Examples include Llama 3.1 (base) before its instruction-tuned variant.

For most chat applications you want the instruction-tuned version. Base models are useful as the starting point for your own LoRA fine-tuning — they accept new behaviours more readily than already-tuned models.

How osFoundry approaches Base Model

osFoundry’s training pipeline lets you fine-tune any of 60+ supported base models. Pick a base, point at your dataset, train, route. The adapter is auto-registered and routable from Maestro the moment it finishes.