How much VRAM does Qwen2.5-7B-Instruct need?

Q4_K_M is exactly 4,683,073,632 bytes (4.36 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Qwen2.5-7B-Instruct's KV cache?

1.75 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 · text

Qwen2.5-7B-Instruct

Qwen/Qwen2.5-7B-Instruct

Qwen2.5-7B-Instruct at Q4_K_M is exactly 4,683,073,632 bytes (4.36 GiB / 4.68 GB) — an effective 4.919 bits per weight, not the nominal 4. Its KV cache at 32K is 1.75 GiB.

From the file· summed from 2 file(s)From the file· KV per layer

Parameters

7.6B

Architecture

qwen2

28 layers

Context

32,768

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ2_M	2.59 GiB	2,780,342,976	2.921	—	bartowski
Q2_K	2.81 GiB	3,015,940,000	3.168	—	Qwen
Q2_K	2.81 GiB	3,015,940,800	3.168	—	bartowski
IQ3_XS	3.12 GiB	3,346,256,576	3.515	—	bartowski
Q3_K_S	3.25 GiB	3,492,369,088	3.669	—	bartowski
Q2_K_L	3.30 GiB	3,548,164,800	3.727	—	bartowski
IQ3_M	3.33 GiB	3,574,012,608	3.754	—	bartowski
Q3_K_M	3.55 GiB	3,808,391,072	4.001	339	Qwen
Q3_K_M	3.55 GiB	3,808,391,872	4.001	—	bartowski
Q3_K_L	3.81 GiB	4,088,459,680	4.295	—	lmstudio-community
Q3_K_L	3.81 GiB	4,088,459,968	4.295	—	bartowski
IQ4_XS	3.93 GiB	4,218,473,152	4.431	339	bartowski
Q4_02 shards	4.13 GiB	4,431,390,848	4.655	—	Qwen
Q4_0	4.14 GiB	4,444,121,792	4.668	339	bartowski
Q4_K_S	4.15 GiB	4,457,769,664	4.683	—	bartowski
Q4_K_M2 shards	4.36 GiB	4,683,073,632	4.919	—	Qwen
Q4_K_M	4.36 GiB	4,683,073,952	4.919	339	lmstudio-community
Q4_K_M	4.36 GiB	4,683,074,240	4.919	—	bartowski
Q4_K_L	4.74 GiB	5,087,564,480	5.344	—	bartowski
Q5_02 shards	4.95 GiB	5,315,176,576	5.583	—	Qwen
Q5_K_S	4.95 GiB	5,315,177,152	5.583	—	bartowski
Q5_K_M2 shards	5.07 GiB	5,444,831,360	5.720	—	Qwen
Q5_K_M	5.07 GiB	5,444,831,936	5.720	339	bartowski
Q5_K_L	5.38 GiB	5,781,197,504	6.073	—	bartowski
Q6_K2 shards	5.82 GiB	6,254,198,880	6.570	—	Qwen
Q6_K	5.82 GiB	6,254,199,200	6.570	339	lmstudio-community
Q6_K	5.82 GiB	6,254,199,488	6.570	339	bartowski
Q6_K_L	6.07 GiB	6,518,182,592	6.847	—	bartowski
Q8_03 shards	7.54 GiB	8,098,525,408	8.507	—	Qwen
Q8_0	7.54 GiB	8,098,525,600	8.507	339	lmstudio-community
Q8_0	7.54 GiB	8,098,525,888	8.507	339	bartowski
F16	14.19 GiB	15,237,853,600	16.007	339	bartowski

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.22 GiB	0.22 GiB	—	28 / 0 / 0
8,192	0.44 GiB	0.44 GiB	—	28 / 0 / 0
16,384	0.88 GiB	0.88 GiB	—	28 / 0 / 0
32,768	1.75 GiB	1.75 GiB	—	28 / 0 / 0
65,536	3.50 GiB	3.50 GiB	—	28 / 0 / 0
131,072	7.00 GiB	7.00 GiB	—	28 / 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 llama-3-youko-8b 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 3.99 GiB. The real file is 4.36 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

3584

Vocab

152,064

Sliding window

131072

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 Qwen2.5-7B-Instruct need?: Q4_K_M is exactly 4,683,073,632 bytes (4.36 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Qwen2.5-7B-Instruct's KV cache?: 1.75 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 Qwen2.5-7B-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.