How much VRAM does Nanbeige4.2-3B-heretic need?

Q4_K_M is exactly 2,574,808,288 bytes (2.40 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Nanbeige4.2-3B-heretic's KV cache?

2.75 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.

WaveCut · text

Nanbeige4.2-3B-heretic

WaveCut/Nanbeige4.2-3B-heretic

Nanbeige4.2-3B-heretic at Q4_K_M is exactly 2,574,808,288 bytes (2.40 GiB / 2.57 GB) — an effective 4.940 bits per weight, not the nominal 4. Its KV cache at 32K is 2.75 GiB.

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

Parameters

4.2B

Architecture

nanbeige

22 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
Q4_K_M	2.40 GiB	2,574,808,288	4.940	—	WaveCut
Q5_K_M	2.78 GiB	2,986,996,960	5.731	—	WaveCut
Q6_K	3.19 GiB	3,424,947,424	6.571	—	WaveCut
Q8_0	4.13 GiB	4,434,787,296	8.508	—	WaveCut

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.34 GiB	0.34 GiB	—	22 / 0 / 0
8,192	0.69 GiB	0.69 GiB	—	22 / 0 / 0
16,384	1.38 GiB	1.38 GiB	—	22 / 0 / 0
32,768	2.75 GiB	2.75 GiB	—	22 / 0 / 0
65,536	5.50 GiB	5.50 GiB	—	22 / 0 / 0
131,072	11.00 GiB	11.00 GiB	—	22 / 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 2.18 GiB. The real file is 2.40 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

3072

Vocab

166,144

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Nanbeige4.2-3B-heretic need?: Q4_K_M is exactly 2,574,808,288 bytes (2.40 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Nanbeige4.2-3B-heretic's KV cache?: 2.75 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 Nanbeige4.2-3B-heretic 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.