Instruct model
Also known as: -Instruct · chat model
An instruct model is a base model that has gone through instruction tuning (and usually RLHF) so it follows instructions and behaves like a helpful assistant.
You’ll see it in names: “Llama-3-8B-Instruct”, “Mistral-Instruct”, “Qwen-Instruct”.
- Base model: predicts text.
- Instruct model: completes tasks, answers questions, stays on topic.
Practical rule: for chat and everyday use, grab the **Instruct** version. Reach for the base model only when you know what you're doing (e.g., specific fine-tuning).