Phi Models — Hardware Requirements
9 Phi models from dphn and the community, from the smallest that runs in 0.9 GB of VRAM at Q4 up to 34.4B parameters. Every row links to full quantization tables and GPU compatibility.
All Phi Models by Size
| Model | Params | Runs from (Q4) | Context | Publisher | Quant downloads |
|---|---|---|---|---|---|
| TinyDolphin 2.8 1.1B | 1.1B | 1.1 GB | 4K | ||
| Phi 1 5 | 1.4B | 0.9 GB | 2K | ||
| Tofu Ft Phi 1.5 | 1.4B | 1.6 GB | 2K | ||
| Phi 1 | 1.4B | 0.9 GB | 2K | ||
| Phi Tiny MoE Instruct | 3.8B | 3.1 GB | 4K | ||
| MediPhi Instruct | 3.8B | 9.0 GB | 131K | ||
| Phi Mini MoE Instruct | 7.6B | 5.4 GB | 4K | ||
| Dolphin 2.9.1 Yi 1.5 34B | 34.4B | 22.9 GB | 8K |
Community variants and fine-tunes(3)
| Model | Params | Runs from (Q4) | Context | Publisher | Quant downloads |
|---|---|---|---|---|---|
| Dolphin X1 Trinity Nano | 6.1B | 4.9 GB | 131K | ||
| Dolphin Mistral 24B Venice Edition | 24.0B | 18.1 GB | 131K | ||
| KAT Coder V2.5 Dev 35B A3B ABLITERATED UNCENSORED PHILADELPHIA CLASS | 34.7B | 21.3 GB | 262K |
Frequently Asked Questions
- How much VRAM do I need to run a Phi model?
- The smallest Phi model, Phi 1 5, runs from 0.9 GB of VRAM at Q4_K_M (smaller low-bit quantizations start at 0.4 GB with a noticeable quality loss). Larger family members need proportionally more — see the table above for every model.
- Which Phi models can I run on a 16 GB GPU?
- 8 of 11 Phi models fit in 16 GB of VRAM at Q4_K_M, including Phi Mini MoE Instruct, Phi 1 5, Phi Tiny MoE Instruct.
- What is the most popular Phi model to run locally?
- Dolphin Mistral 24B Venice Edition is the most downloaded Phi model in local-friendly quantized formats. It runs from 18.1 GB of VRAM at Q4_K_M.