How much VRAM does Qwen2.5-VL-72B-Instruct need?

Q4_K_M is exactly 47,415,713,248 bytes (44.16 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Qwen2.5-VL-72B-Instruct's KV cache?

10.00 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.

Qwen · vision language

Qwen2.5-VL-72B-Instruct

Qwen/Qwen2.5-VL-72B-Instruct

Qwen2.5-VL-72B-Instruct at Q4_K_M is exactly 47,415,713,248 bytes (44.16 GiB / 47.42 GB) — an effective 5.167 bits per weight, not the nominal 4. Its KV cache at 32K is 10.00 GiB.

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

Parameters

73.4B

Architecture

qwen2vl

80 layers

Context

128,000

native (config.json)

License

other

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ1_M	22.11 GiB	23,740,210,944	2.587	—	bartowski
IQ2_XXS	23.74 GiB	25,490,022,144	2.778	—	bartowski
UD-IQ2_XXS	23.94 GiB	25,700,426,144	2.801	—	unsloth
IQ2_XS	25.20 GiB	27,057,643,264	2.949	—	bartowski
IQ2_S	26.02 GiB	27,939,135,232	3.045	—	bartowski
IQ2_M	27.32 GiB	29,338,984,192	3.197	—	bartowski
UD-IQ2_M	27.60 GiB	29,631,275,424	3.229	—	unsloth
Q2_K	27.76 GiB	29,811,760,896	3.249	—	bartowski
Q2_K	27.76 GiB	29,811,761,568	3.249	—	unsloth
Q2_K_L	28.04 GiB	30,103,724,448	3.281	—	unsloth
Q2_K_L	28.90 GiB	31,028,272,896	3.381	—	bartowski
IQ3_XXS	29.66 GiB	31,845,080,832	3.470	—	bartowski
UD-IQ3_XXS	29.67 GiB	31,859,237,280	3.472	—	unsloth
IQ3_XS	30.59 GiB	32,842,178,304	3.579	—	bartowski
Q3_K_S	32.12 GiB	34,487,787,264	3.758	—	bartowski
Q3_K_S	32.12 GiB	34,487,787,936	3.758	—	unsloth
IQ3_M	33.07 GiB	35,503,595,264	3.869	—	bartowski
Q3_K_M	35.11 GiB	37,698,723,584	4.108	—	bartowski
Q3_K_M	35.11 GiB	37,698,724,256	4.108	—	unsloth
Q3_K_L	36.79 GiB	39,505,223,136	4.305	—	lmstudio-community
Q3_K_L	36.79 GiB	39,505,223,424	4.305	—	bartowski
IQ4_XS	36.98 GiB	39,709,073,152	4.327	—	bartowski
IQ4_XS	37.02 GiB	39,748,002,208	4.332	—	unsloth
IQ4_NL	38.48 GiB	41,315,622,656	4.502	—	bartowski
IQ4_NL	38.48 GiB	41,315,623,328	4.502	—	unsloth
Q4_0	38.54 GiB	41,383,124,736	4.510	—	bartowski
Q4_0	38.54 GiB	41,383,125,408	4.510	—	unsloth
Q4_K_S	40.88 GiB	43,889,221,376	4.783	—	bartowski
Q4_K_S	40.88 GiB	43,889,222,048	4.783	—	unsloth
Q4_1	42.56 GiB	45,697,883,904	4.980	—	bartowski
Q4_1	42.56 GiB	45,697,884,576	4.980	—	unsloth
Q4_K_M	44.16 GiB	47,415,713,248	5.167	—	lmstudio-community
Q4_K_M	44.16 GiB	47,415,713,536	5.167	—	bartowski
Q4_K_M	44.16 GiB	47,415,714,208	5.167	—	unsloth
Q4_K_L	45.02 GiB	48,340,262,656	5.268	—	bartowski
Q5_K_S2 shards	47.85 GiB	51,375,136,704	5.599	—	bartowski
Q5_K_S2 shards	47.85 GiB	51,375,137,408	5.599	—	unsloth
Q5_K_M2 shards	50.71 GiB	54,447,464,416	5.934	—	bartowski
Q5_K_M2 shards	50.71 GiB	54,447,465,056	5.934	—	unsloth
Q6_K2 shards	59.93 GiB	64,347,627,200	7.012	—	lmstudio-community

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	1.25 GiB	1.25 GiB	—	80 / 0 / 0
8,192	2.50 GiB	2.50 GiB	—	80 / 0 / 0
16,384	5.00 GiB	5.00 GiB	—	80 / 0 / 0
32,768	10.00 GiB	10.00 GiB	—	80 / 0 / 0
65,536	20.00 GiB	20.00 GiB	—	80 / 0 / 0
131,072	40.00 GiB	40.00 GiB	—	80 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3.6-35B-A3B vs Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP vs Qwen3.5-9B vs gemma-4-26B-A4B-it vs gemma-4-12B-it vs Qwythos-9B-Claude-Mythos-5-1M 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 38.46 GiB. The real file is 44.16 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

8192

Vocab

152,064

Sliding window

32768

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

false

This model declares a sliding window but sets use_sliding_window: false, so the window is not applied. Honouring the field without the flag understates KV for the whole family.

Questions people ask

How much VRAM does Qwen2.5-VL-72B-Instruct need?: Q4_K_M is exactly 47,415,713,248 bytes (44.16 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Qwen2.5-VL-72B-Instruct's KV cache?: 10.00 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 Qwen2.5-VL-72B-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.