It decided to download Ollama first (1.3GB)
| Name | Tag | Size |
|---|---|---|
| Qwen (fast) | qwen3.5:4b | 3.4 GB |
| Qwen9 | qwen3.5:9b | 6.6 GB |
| Phi4 | phi4-mini | 2.5 GB |
| qwen3.5:2b | ~2.7 GB | |
| granite4.2:3b | ~2 GB | |
| gemma4:e2b | ~2–4 GB (tag-dependent) | |
| qwen3.5:27b | ~17 GB | |
| phi4 (14B) | ~9–12 GB |
RAG, add nomic-embed-text at ~0.3 GB.
Can't LoRA on CPU. Expect hours to a couple of days on CPU, not minutes. But this is an add-on to the model that allows training.