How much VRAM does Hy-MT2-7B need?

Q4_K_M is exactly 4,624,648,896 bytes (4.31 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Hy-MT2-7B'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.

tencent · text

Hy-MT2-7B

tencent/Hy-MT2-7B

Hy-MT2-7B at Q4_K_M is exactly 4,624,648,896 bytes (4.31 GiB / 4.62 GB) — an effective 4.608 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

8.0B

Architecture

hunyuan-dense

32 layers

Context

262,144

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
UD-IQ2_M	2.63 GiB	2,828,577,856	2.818	—	unsloth
Q2_K	2.80 GiB	3,006,172,736	2.995	—	mradermacher
UD-IQ3_XXS	2.91 GiB	3,127,356,480	3.116	—	unsloth
Q3_K_S	3.20 GiB	3,438,186,048	3.425	—	mradermacher
Q3_K_S	3.20 GiB	3,438,186,272	3.425	—	unsloth
Q3_K_M	3.53 GiB	3,792,604,736	3.779	—	mradermacher
Q3_K_M	3.53 GiB	3,792,604,960	3.779	—	unsloth
Q3_K_L	3.81 GiB	4,095,643,200	4.081	—	mradermacher
IQ4_XS	3.88 GiB	4,167,995,168	4.153	—	unsloth
IQ4_XS	3.92 GiB	4,204,695,104	4.189	—	mradermacher
Q4_0	4.08 GiB	4,379,807,520	4.364	—	unsloth
IQ4_NL	4.08 GiB	4,381,904,672	4.366	—	unsloth
Q4_K_S	4.09 GiB	4,396,584,512	4.380	—	mradermacher
Q4_K_S	4.09 GiB	4,396,584,736	4.380	—	unsloth
Q4_K_M	4.31 GiB	4,624,648,896	4.608	—	tencent
Q4_K_M	4.31 GiB	4,624,649,792	4.608	—	mradermacher
Q4_K_M	4.31 GiB	4,624,650,016	4.608	—	unsloth
Q4_1	4.47 GiB	4,801,335,072	4.784	—	unsloth
Q5_K_S	4.88 GiB	5,237,542,464	5.218	—	mradermacher
Q5_K_S	4.88 GiB	5,237,542,688	5.218	—	unsloth
Q5_K_M	5.00 GiB	5,371,235,904	5.351	—	mradermacher
Q5_K_M	5.00 GiB	5,371,236,128	5.351	—	unsloth
Q6_K	5.74 GiB	6,164,482,720	6.142	—	tencent
Q6_K	5.74 GiB	6,164,483,648	6.142	—	mradermacher
Q6_K	5.74 GiB	6,164,483,872	6.142	—	unsloth
Q8_0	7.43 GiB	7,981,928,896	7.953	—	tencent
Q8_0	7.43 GiB	7,981,929,824	7.953	—	mradermacher
Q8_0	7.43 GiB	7,981,930,048	7.953	—	unsloth
BF16	13.99 GiB	15,017,205,312	14.962	—	unsloth
F16	13.99 GiB	15,017,205,344	14.962	—	mradermacher

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-4-12B-it-qat-q4_0-unquantized vs gemma-4-E4B-it vs Qwen3-4B vs Qwen3-8B vs HyperCLOVAX-SEED-Text-Instruct-1.5B vs Llama-3.2-1B-Instruct 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 4.21 GiB. The real file is 4.31 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

4096

Vocab

128,167

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Hy-MT2-7B need?: Q4_K_M is exactly 4,624,648,896 bytes (4.31 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Hy-MT2-7B'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 Hy-MT2-7B 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.