How much VRAM does MiniCPM5-1B-Uncensored need?

Q4_K_M is exactly 688,066,944 bytes (0.64 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is MiniCPM5-1B-Uncensored's KV cache?

0.75 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.

sahilchachra · text

MiniCPM5-1B-Uncensored

sahilchachra/MiniCPM5-1B-Uncensored

MiniCPM5-1B-Uncensored at Q4_K_M is exactly 688,066,944 bytes (0.64 GiB / 0.69 GB) — an effective 5.094 bits per weight, not the nominal 4. Its KV cache at 32K is 0.75 GiB.

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

Parameters

1.1B

Architecture

llama

24 layers

Context

131,072

native (config.json)

License

other

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
I1-IQ1_S	0.33 GiB	352,826,016	2.612	—	mradermacher
I1-IQ1_M	0.34 GiB	366,262,944	2.712	—	mradermacher
I1-IQ2_XXS	0.36 GiB	388,657,824	2.877	—	mradermacher
I1-IQ2_XS	0.38 GiB	408,933,024	3.027	—	mradermacher
I1-IQ2_S	0.41 GiB	442,073,760	3.273	—	mradermacher
I1-IQ2_M	0.43 GiB	459,989,664	3.405	—	mradermacher
I1-Q2_K_S	0.43 GiB	465,887,904	3.449	—	mradermacher
Q2_K	0.45 GiB	485,831,040	3.597	—	mradermacher
I1-Q2_K	0.45 GiB	485,831,328	3.597	—	mradermacher
I1-IQ3_XXS	0.46 GiB	489,517,728	3.624	—	mradermacher
I1-IQ3_XS	0.50 GiB	533,404,320	3.949	—	mradermacher
Q3_K_S	0.51 GiB	548,075,904	4.057	—	mradermacher
I1-Q3_K_S	0.51 GiB	548,076,192	4.057	—	mradermacher
I1-IQ3_S	0.51 GiB	549,329,568	4.067	—	mradermacher
I1-IQ3_M	0.52 GiB	562,176,672	4.162	—	mradermacher
Q3_K_M	0.54 GiB	582,900,096	4.315	—	mradermacher
I1-Q3_K_M	0.54 GiB	582,900,384	4.315	—	mradermacher
Q3_K_L	0.57 GiB	613,767,552	4.544	—	mradermacher
I1-Q3_K_L	0.57 GiB	613,767,840	4.544	—	mradermacher
I1-IQ4_XS	0.60 GiB	638,927,520	4.730	—	mradermacher
IQ4_XS	0.60 GiB	642,244,992	4.755	—	mradermacher
I1-IQ4_NL	0.62 GiB	666,133,152	4.931	—	mradermacher
I1-Q4_0	0.62 GiB	666,280,608	4.933	—	mradermacher
Q4_K_S	0.62 GiB	667,804,032	4.944	—	mradermacher
I1-Q4_K_S	0.62 GiB	667,804,320	4.944	—	mradermacher
Q4_K_M	0.64 GiB	688,066,944	5.094	—	mradermacher
I1-Q4_K_M	0.64 GiB	688,067,232	5.094	—	mradermacher
I1-Q4_1	0.67 GiB	719,954,592	5.330	—	mradermacher
Q5_K_S	0.72 GiB	774,955,392	5.737	—	mradermacher
I1-Q5_K_S	0.72 GiB	774,955,680	5.737	—	mradermacher
Q5_K_M	0.73 GiB	786,862,464	5.825	—	mradermacher
I1-Q5_K_M	0.73 GiB	786,862,752	5.825	—	mradermacher
Q6_K	0.83 GiB	891,832,704	6.602	—	mradermacher
I1-Q6_K	0.83 GiB	891,832,992	6.602	—	mradermacher
Q8_0	1.07 GiB	1,153,530,240	8.540	—	mradermacher
F16	2.02 GiB	2,166,552,960	16.039	—	mradermacher

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.09 GiB	0.09 GiB	—	24 / 0 / 0
8,192	0.19 GiB	0.19 GiB	—	24 / 0 / 0
16,384	0.38 GiB	0.38 GiB	—	24 / 0 / 0
32,768	0.75 GiB	0.75 GiB	—	24 / 0 / 0
65,536	1.50 GiB	1.50 GiB	—	24 / 0 / 0
131,072	3.00 GiB	3.00 GiB	—	24 / 0 / 0

Compare with

same modality, comparable size

vs Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs Qwen3-1.7B vs embeddinggemma-300m vs Qwen3-0.6B 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 0.57 GiB. The real file is 0.64 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

130,560

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does MiniCPM5-1B-Uncensored need?: Q4_K_M is exactly 688,066,944 bytes (0.64 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is MiniCPM5-1B-Uncensored's KV cache?: 0.75 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 MiniCPM5-1B-Uncensored 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.