Download and pick models

Cross-tool recommendations: Recommended models. This page focuses on Ollama commands.

Pull models

ollama pull <model-name>

Starter examples:

ollama pull llama3.2:3b
ollama pull qwen2.5:3b
ollama pull qwen2.5:7b
ollama pull qwen2.5:14b

List downloaded:

ollama list

Local chat:

ollama run qwen2.5:7b

Pick size

Rough size	RAM/VRAM rough guide	Notes
1B–3B	~4–8GB	Path check, light replies
7B–8B	~8–16GB	Common roleplay sweet spot
14B+	16GB+	Better quality, heavier hardware

Smaller quants use less memory and run faster with a quality trade-off. Ollama tags (e.g. :3b, :7b) already ship official default quants — beginners can use official names as-is.

Roleplay tips

• Prefer chat / instruct finetunes; raw completion bases may not fit chat format.
• The model name in MiniTavern must match the left column of ollama list exactly (including tag).
• For Chinese cards, try Qwen first; switch by feel afterward.

Next

Enable LAN access → Configure in MiniTavern