How much VRAM does Llama-3.2-1B-Instruct-abliterated need?

Q4_K_M is exactly 955,445,792 bytes (0.89 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Llama-3.2-1B-Instruct-abliterated's KV cache?

1.00 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.

huihui-ai · text

Llama-3.2-1B-Instruct-abliterated

huihui-ai/Llama-3.2-1B-Instruct-abliterated

Llama-3.2-1B-Instruct-abliterated at Q4_K_M is exactly 955,445,792 bytes (0.89 GiB / 0.96 GB) — an effective 5.101 bits per weight, not the nominal 4. Its KV cache at 32K is 1.00 GiB.

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

Parameters

1.5B

Architecture

llama

16 layers

Context

131,072

native (config.json)

License

llama3.2

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
I1-IQ1_S	0.45 GiB	479,740,704	2.561	—	mradermacher
I1-IQ1_M	0.47 GiB	499,794,720	2.668	—	mradermacher
I1-IQ2_XXS	0.50 GiB	533,218,080	2.847	—	mradermacher
I1-IQ2_XS	0.52 GiB	562,053,920	3.001	—	mradermacher
I1-IQ2_S	0.56 GiB	601,576,224	3.212	—	mradermacher
I1-IQ2_M	0.59 GiB	628,314,912	3.354	—	mradermacher
Q2_K	0.62 GiB	667,062,336	3.561	—	tensorblock
Q2_K	0.62 GiB	667,062,816	3.561	—	mradermacher
I1-Q2_K	0.62 GiB	667,063,072	3.561	—	mradermacher
I1-IQ3_XXS	0.63 GiB	674,976,544	3.603	—	mradermacher
IQ3_XS	0.68 GiB	733,979,168	3.918	—	mradermacher
I1-IQ3_XS	0.68 GiB	733,979,424	3.918	—	mradermacher
Q3_K_S	0.70 GiB	754,557,472	4.028	—	mradermacher
I1-Q3_K_S	0.70 GiB	754,557,728	4.028	—	mradermacher
IQ3_S	0.70 GiB	756,785,696	4.040	—	mradermacher
I1-IQ3_S	0.70 GiB	756,785,952	4.040	—	mradermacher
IQ3_M	0.72 GiB	770,155,040	4.112	—	mradermacher
I1-IQ3_M	0.72 GiB	770,155,296	4.112	—	mradermacher
Q3_K_M	0.75 GiB	803,708,992	4.291	—	tensorblock
Q3_K_M	0.75 GiB	803,709,472	4.291	—	mradermacher
I1-Q3_K_M	0.75 GiB	803,709,728	4.291	—	mradermacher
Q3_K_L	0.79 GiB	845,390,368	4.513	—	mradermacher
I1-Q3_K_L	0.79 GiB	845,390,624	4.513	—	mradermacher
I1-IQ4_XS	0.82 GiB	882,684,704	4.712	—	mradermacher
IQ4_XS	0.83 GiB	887,927,328	4.740	—	mradermacher
I1-Q4_0	0.86 GiB	920,777,504	4.916	—	mradermacher
Q4_K_S	0.86 GiB	923,398,688	4.930	—	mradermacher
I1-Q4_K_S	0.86 GiB	923,398,944	4.930	—	mradermacher
Q4_K_M	0.89 GiB	955,445,792	5.101	—	mradermacher
I1-Q4_K_M	0.89 GiB	955,446,048	5.101	—	mradermacher
Q5_K_S	1.00 GiB	1,073,148,448	5.729	—	mradermacher
I1-Q5_K_S	1.00 GiB	1,073,148,704	5.729	—	mradermacher
Q5_K_M	1.02 GiB	1,092,088,352	5.830	—	mradermacher
I1-Q5_K_M	1.02 GiB	1,092,088,608	5.830	—	mradermacher
Q6_K	1.15 GiB	1,237,271,072	6.606	—	mradermacher
I1-Q6_K	1.15 GiB	1,237,271,328	6.606	—	mradermacher
Q8_0	1.49 GiB	1,600,168,480	8.543	—	mradermacher
F16	2.80 GiB	3,004,932,640	16.043	—	mradermacher

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.13 GiB	0.13 GiB	—	16 / 0 / 0
8,192	0.25 GiB	0.25 GiB	—	16 / 0 / 0
16,384	0.50 GiB	0.50 GiB	—	16 / 0 / 0
32,768	1.00 GiB	1.00 GiB	—	16 / 0 / 0
65,536	2.00 GiB	2.00 GiB	—	16 / 0 / 0
131,072	4.00 GiB	4.00 GiB	—	16 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3-4B vs HyperCLOVAX-SEED-Text-Instruct-1.5B vs Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs Qwen3-1.7B 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 0.79 GiB. The real file is 0.89 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

Hidden size

2048

Vocab

128,256

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Llama-3.2-1B-Instruct-abliterated need?: Q4_K_M is exactly 955,445,792 bytes (0.89 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Llama-3.2-1B-Instruct-abliterated's KV cache?: 1.00 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 Llama-3.2-1B-Instruct-abliterated 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.