unsloth
phi 4
phi 4 is a 14.7B dense llama model from unsloth. Configured context length is 16,384 tokens. License: mit. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters14.7B
Context16K
Attentiongqa
Licensemit
40 layers · 10 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| F16 | 27.3 GB | Hugging Face file bytes |
| Q8_0 | 14.5 GB | Hugging Face file bytes |
| Q6_K | 11.2 GB | Hugging Face file bytes |
| Q5_K_M | 9.70 GB | Hugging Face file bytes |
| Q4_K_M | 8.28 GB | Hugging Face file bytes |
| Q3_K_M | 6.70 GB | Hugging Face file bytes |
| Q2_K | 10.6 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Partial offload | Q4_K_M | 10–14 |
| GeForce RTX 3060 12GB | Tight fit | Q4_K_M | 19–26 |
| GeForce RTX 4060 Ti 16GB | Tight fit | Q6_K | 12–16 |
| GeForce RTX 4090 | Runs | Q8_0 | 32–44 |
| GeForce RTX 5090 | Tight fit | F16 | 33–45 |
| Apple M4 Pro | Tight fit | Q8_0 | 8.7–12 |
| Apple M5 | Tight fit | Q4_K_M | 8.0–11 |
| CPU only · DDR5-5600 | Runs on CPU | Q8_0 | 2.9–3.9 |
Smallest devices that fit it fully
- GeForce RTX 3080 10GB — Tight fit, Q3_K_M, 47–64 tok/s
- Intel Arc B570 — Tight fit, Q3_K_M, 24–32 tok/s
- GeForce RTX 2080 Ti — Tight fit, Q4_K_M, 35–47 tok/s
- GeForce GTX 1080 Ti — Tight fit, Q4_K_M, 27–37 tok/s
- GeForce RTX 5070 — Tight fit, Q4_K_M, 35–48 tok/s
- GeForce RTX 4070 Ti — Tight fit, Q4_K_M, 26–36 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (Q6_K).
llama.cpp
llama-server -hf unsloth/phi-4-GGUF:Q6_K -c 8192 -ngl 40
Ollama
ollama run phi4
LM Studio
lms get unsloth/phi-4-GGUF # Then load quant Q6_K