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

I1-IQ1_S is exactly 22,690,326,176 bytes (21.13 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Malaysian-Qwen2.5-72B-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.

mesolitica · text

Malaysian-Qwen2.5-72B-Instruct

mesolitica/Malaysian-Qwen2.5-72B-Instruct

Malaysian-Qwen2.5-72B-Instruct at I1-IQ1_S is exactly 22,690,326,176 bytes (21.13 GiB / 22.69 GB) — an effective 2.497 bits per weight, not the nominal 1. Its KV cache at 32K is 10.00 GiB.

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

Parameters

72.7B

Architecture

qwen2

80 layers

Context

32,768

native (config.json)

License

—

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
I1-IQ1_S	21.13 GiB	22,690,326,176	2.497	—	mradermacher
I1-IQ1_M	22.11 GiB	23,740,212,896	2.612	—	mradermacher
I1-IQ2_XXS	23.74 GiB	25,490,024,096	2.805	—	mradermacher
I1-IQ2_XS	25.20 GiB	27,057,645,216	2.977	—	mradermacher
I1-IQ2_S	26.02 GiB	27,939,137,184	3.074	—	mradermacher
I1-IQ2_M	27.32 GiB	29,338,986,144	3.228	—	mradermacher
I1-Q2_K_S	27.54 GiB	29,569,279,648	3.254	—	mradermacher
I1-Q2_K	27.76 GiB	29,811,762,848	3.280	—	mradermacher
I1-IQ3_XXS	29.66 GiB	31,845,082,784	3.504	—	mradermacher
I1-IQ3_XS	30.59 GiB	32,842,180,256	3.614	—	mradermacher
I1-Q3_K_S	32.12 GiB	34,487,789,216	3.795	—	mradermacher
I1-IQ3_S	32.12 GiB	34,487,789,216	3.795	—	mradermacher
I1-IQ3_M	33.07 GiB	35,503,597,216	3.906	—	mradermacher
I1-Q3_K_M	35.11 GiB	37,698,725,536	4.148	—	mradermacher
I1-Q3_K_L	36.79 GiB	39,505,225,376	4.347	—	mradermacher
I1-IQ4_XS	36.98 GiB	39,709,075,104	4.369	—	mradermacher
I1-Q4_0	38.54 GiB	41,383,126,688	4.553	—	mradermacher
I1-Q4_K_S	40.88 GiB	43,889,223,328	4.829	—	mradermacher
I1-Q4_1	42.56 GiB	45,697,885,856	5.028	—	mradermacher
I1-Q4_K_M	44.16 GiB	47,415,715,488	5.217	—	mradermacher

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-Coder-30B-A3B-Instruct vs Qwen3.6-27B vs Qwen3.8-27B vs gemma-4-E4B-it vs Qwen3-30B-A3B-Thinking-2507 vs Qwen3-8B All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts I1-IQ1_S at roughly 38.09 GiB. The real file is 21.13 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

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 Malaysian-Qwen2.5-72B-Instruct need?: I1-IQ1_S is exactly 22,690,326,176 bytes (21.13 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Malaysian-Qwen2.5-72B-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 Malaysian-Qwen2.5-72B-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.