How much VRAM does Phi-4-mini-instruct need?

Q4_K_M is exactly 2,491,874,272 bytes (2.32 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Phi-4-mini-instruct's KV cache?

4.00 GiB at 32K context with an f16 cache, computed per layer from the model's architecture. Quantizing the cache to q8_0 roughly halves it.

microsoft · text

Phi-4-mini-instruct

microsoft/Phi-4-mini-instruct

Phi-4-mini-instruct at Q4_K_M is exactly 2,491,874,272 bytes (2.32 GiB / 2.49 GB) — an effective 5.197 bits per weight, not the nominal 4. Its KV cache at 32K is 4.00 GiB.

From the file· summed from 1 file(s)From the file· KV per layer

Parameters

3.8B

Architecture

phi3

32 layers

Context

131,072

native (config.json)

License

—

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ2_M	1.40 GiB	1,507,424,640	3.144	—	bartowski
IQ3_XXS	1.56 GiB	1,678,866,816	3.501	—	bartowski
Q2_K	1.57 GiB	1,682,635,744	3.509	—	unsloth
Q2_K_L	1.57 GiB	1,682,635,744	3.509	—	unsloth
Q2_K	1.57 GiB	1,682,636,160	3.509	—	bartowski
Q2_K_L	1.71 GiB	1,831,483,776	3.820	—	bartowski
IQ3_XS	1.71 GiB	1,840,708,992	3.839	—	bartowski
Q3_K_S	1.77 GiB	1,897,332,096	3.957	—	bartowski
IQ3_M	1.88 GiB	2,017,656,192	4.208	—	bartowski
Q3_K_M	1.97 GiB	2,117,532,640	4.416	196	unsloth
Q3_K_M	1.97 GiB	2,117,533,056	4.416	196	bartowski
IQ4_XS	2.07 GiB	2,224,487,808	4.639	196	bartowski
Q3_K_L	2.10 GiB	2,249,653,344	4.692	—	lmstudio-community
Q3_K_L	2.10 GiB	2,249,653,632	4.692	—	bartowski
IQ4_NL	2.17 GiB	2,325,151,104	4.849	—	bartowski
Q4_0	2.17 GiB	2,331,442,560	4.862	196	bartowski
Q4_K_S	2.18 GiB	2,337,734,016	4.875	—	bartowski
Q4_K_M	2.32 GiB	2,491,874,272	5.197	196	unsloth
Q4_K_M	2.32 GiB	2,491,874,400	5.197	196	lmstudio-community
Q4_K_M	2.32 GiB	2,491,874,688	5.197	—	bartowski
Q4_1	2.35 GiB	2,526,477,696	5.269	—	bartowski
Q4_K_L	2.46 GiB	2,640,722,304	5.507	—	bartowski
Q5_K_S	2.54 GiB	2,727,804,288	5.689	—	bartowski
Q5_K_M	2.65 GiB	2,848,127,968	5.940	196	unsloth
Q5_K_M	2.65 GiB	2,848,128,384	5.940	196	bartowski
Q5_K_L	2.79 GiB	2,996,976,000	6.250	—	bartowski
Q6_K	2.94 GiB	3,155,622,880	6.581	—	unsloth
Q6_K	2.94 GiB	3,155,623,008	6.581	196	lmstudio-community
Q6_K	2.94 GiB	3,155,623,296	6.581	196	bartowski
Q6_K_L	3.08 GiB	3,304,470,912	6.891	—	bartowski
Q8_0	3.80 GiB	4,084,611,040	8.518	196	unsloth
Q8_0	3.80 GiB	4,084,611,168	8.518	196	lmstudio-community
Q8_0	3.80 GiB	4,084,611,456	8.518	196	bartowski
BF16	7.15 GiB	7,680,694,240	16.018	—	unsloth

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.50 GiB	0.50 GiB	—	32 / 0 / 0
8,192	1.00 GiB	1.00 GiB	—	32 / 0 / 0
16,384	2.00 GiB	2.00 GiB	—	32 / 0 / 0
32,768	4.00 GiB	4.00 GiB	—	32 / 0 / 0
65,536	8.00 GiB	8.00 GiB	—	32 / 0 / 0
131,072	16.00 GiB	16.00 GiB	—	32 / 0 / 0

Compare with

same modality, comparable size

vs gemma-4-E4B-it vs Qwen3-4B vs Qwen3-8B vs HyperCLOVAX-SEED-Text-Instruct-1.5B vs Llama-3.2-1B-Instruct vs Qwen3-VL-8B-Instruct-abliterated-v1 All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 2.01 GiB. The real file is 2.32 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json

Layers

Attention heads

KV heads

Head dim

128

Hidden size

3072

Vocab

200,064

Sliding window

262144

SWA period

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Phi-4-mini-instruct need?: Q4_K_M is exactly 2,491,874,272 bytes (2.32 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Phi-4-mini-instruct's KV cache?: 4.00 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.
Which quantization of Phi-4-mini-instruct should I use?: Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.