How much VRAM does Qwen2.5-Math-1.5B-Instruct need?

Q4_K_M is exactly 986,048,544 bytes (0.92 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-Math-1.5B-Instruct's KV cache?

0.88 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-Math-1.5B-Instruct

Qwen/Qwen2.5-Math-1.5B-Instruct

Qwen2.5-Math-1.5B-Instruct at Q4_K_M is exactly 986,048,544 bytes (0.92 GiB / 0.99 GB) — an effective 5.110 bits per weight, not the nominal 4. Its KV cache at 32K is 0.88 GiB.

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

Parameters

1.5B

Architecture

qwen2

28 layers

Context

4,096

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ1_S	0.41 GiB	436,527,904	2.262	—	legraphista
IQ1_M	0.43 GiB	464,461,600	2.407	—	legraphista
IQ2_XXS	0.48 GiB	511,017,760	2.648	—	legraphista
IQ2_XS	0.51 GiB	550,327,072	2.852	—	legraphista
IQ2_S	0.53 GiB	563,810,080	2.922	—	legraphista
IQ2_M	0.56 GiB	601,054,944	3.115	—	liodon-ai
IQ2_M	0.56 GiB	601,055,008	3.115	—	legraphista
Q2_K_S	0.60 GiB	640,135,456	3.317	—	legraphista
IQ3_XXS	0.62 GiB	668,792,608	3.466	—	legraphista
Q2_K	0.63 GiB	676,305,184	3.505	—	legraphista
IQ3_XS	0.68 GiB	731,699,488	3.792	—	legraphista
Q3_K_S	0.71 GiB	760,944,928	3.943	—	legraphista
IQ3_S	0.71 GiB	762,407,200	3.951	—	legraphista
IQ3_M	0.72 GiB	776,664,288	4.025	—	liodon-ai
IQ3_M	0.72 GiB	776,664,352	4.025	—	legraphista
IQ3_M	0.72 GiB	776,664,384	4.025	—	bartowski
Q3_K	0.77 GiB	824,178,976	4.271	—	legraphista
Q3_K_L	0.82 GiB	880,162,848	4.561	—	lmstudio-community
Q3_K_L	0.82 GiB	880,163,104	4.561	—	legraphista
Q3_K_L	0.82 GiB	880,163,136	4.561	—	bartowski
IQ4_XS	0.83 GiB	895,731,936	4.642	—	liodon-ai
IQ4_XS	0.83 GiB	895,732,000	4.642	—	legraphista
IQ4_XS	0.83 GiB	895,732,032	4.642	—	bartowski
IQ4_NL	0.87 GiB	936,331,552	4.852	—	legraphista
Q4_0	0.87 GiB	937,535,808	4.859	—	bartowski
Q4_K_S	0.88 GiB	940,312,864	4.873	—	legraphista
Q4_K_S	0.88 GiB	940,312,896	4.873	—	bartowski
Q4_K_M	0.92 GiB	986,048,544	5.110	—	lmstudio-community
Q4_K_M	0.92 GiB	986,048,736	5.110	—	liodon-ai
Q4_K	0.92 GiB	986,048,800	5.110	—	legraphista
Q4_K_M	0.92 GiB	986,048,832	5.110	—	bartowski
Q4_K_L	0.97 GiB	1,042,569,024	5.403	—	bartowski
Q5_K_S	1.02 GiB	1,098,729,504	5.694	—	legraphista
Q5_K_S	1.02 GiB	1,098,729,792	5.694	—	bartowski
Q5_K	1.05 GiB	1,125,050,400	5.830	—	legraphista
Q5_K_M	1.05 GiB	1,125,050,592	5.830	—	liodon-ai
Q5_K_M	1.05 GiB	1,125,050,688	5.830	—	bartowski
Q5_K_L	1.10 GiB	1,181,570,880	6.123	—	bartowski
Q6_K	1.19 GiB	1,272,739,872	6.596	—	legraphista
Q6_K	1.19 GiB	1,272,739,872	6.596	—	lmstudio-community

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.11 GiB	0.11 GiB	—	28 / 0 / 0
8,192	0.22 GiB	0.22 GiB	—	28 / 0 / 0
16,384	0.44 GiB	0.44 GiB	—	28 / 0 / 0
32,768	0.88 GiB	0.88 GiB	—	28 / 0 / 0
65,536	1.75 GiB	1.75 GiB	—	28 / 0 / 0
131,072	3.50 GiB	3.50 GiB	—	28 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3-4B vs Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs Qwen3-1.7B vs embeddinggemma-300m 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 0.81 GiB. The real file is 0.92 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

1536

Vocab

151,936

Sliding window

4096

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-Math-1.5B-Instruct need?: Q4_K_M is exactly 986,048,544 bytes (0.92 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-Math-1.5B-Instruct's KV cache?: 0.88 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-Math-1.5B-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.