Yes — with 12.0 GB to spare
Phi-4 14B at Q4_K_M fits your RTX A5000 entirely on the GPU at 8K context, at an estimated 56 tokens per second. There is room for its full 16K window.
Fully on GPU
8K context
Q4_K_M · 8.3 GB
MIT
Released Dec 2024
Trained heavily on synthetic reasoning data. Short 16K window is its main limitation.
The VRAM budget
weights 8.3 GB
Weights 8.3 GB
KV cache @ 8K 1.56 GB
Runtime overhead 0.6 GB
Free 12.0 GB of 22.4 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 27.4 GB | 29.5 GB | — | ~4.2 | Reference | 7.1 GB over |
| Q8_0 | 14.5 GB | 16.7 GB | 16K | 32 | −0.1% ppl | Fits |
| Q6_K | 11.2 GB | 13.4 GB | 16K | 41 | −0.4% ppl | Fits |
| Q5_K_M | 9.7 GB | 11.9 GB | 16K | 48 | −0.8% ppl | Fits |
| Q4_K_M | 8.3 GB | 10.4 GB | 16K | 56 | −1.9% ppl | Recommended |
| Q3_K_M | 6.7 GB | 8.9 GB | 16K | 69 | −5.4% ppl | Fits |
| Q2_K | 5.7 GB | 7.9 GB | 16K | 81 | −15% ppl | Fits |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it.
How to run it
$ ollama pull phi4:14b $ OLLAMA_CONTEXT_LENGTH=8192 \ ollama run phi4:14b
The default. One binary, a model registry, an OpenAI-compatible port. More on Ollama.
01Download is 8.3 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.
03There is room to go to the model's full 16K context on this card.