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

Q4_K_M is exactly 807,694,944 bytes (0.75 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-heretic'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.

saidutta69 · text

Llama-3.2-1B-Instruct-heretic

saidutta69/Llama-3.2-1B-Instruct-heretic

Llama-3.2-1B-Instruct-heretic at Q4_K_M is exactly 807,694,944 bytes (0.75 GiB / 0.81 GB) — an effective 5.229 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

1.2B

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	0.37 GiB	393,552,768	2.548	—	mradermacher
I1-IQ1_M	0.39 GiB	413,606,784	2.678	—	mradermacher
I1-IQ2_XXS	0.42 GiB	447,030,144	2.894	—	mradermacher
I1-IQ2_XS	0.44 GiB	475,865,984	3.080	—	mradermacher
I1-IQ2_S	0.46 GiB	488,711,040	3.164	—	mradermacher
I1-IQ2_M	0.48 GiB	515,449,728	3.337	—	mradermacher
I1-Q2_K_S	0.52 GiB	554,660,736	3.591	—	mradermacher
I1-IQ3_XXS	0.52 GiB	562,111,360	3.639	—	mradermacher
Q2_K	0.54 GiB	580,874,848	3.760	—	mradermacher
I1-Q2_K	0.54 GiB	580,875,136	3.760	—	mradermacher
I1-IQ3_XS	0.58 GiB	621,114,240	4.021	—	mradermacher
Q3_K_S	0.60 GiB	641,692,256	4.154	—	mradermacher
I1-Q3_K_S	0.60 GiB	641,692,544	4.154	—	mradermacher
I1-IQ3_S	0.60 GiB	643,920,768	4.168	—	mradermacher
I1-IQ3_M	0.61 GiB	657,290,112	4.255	—	mradermacher
Q3_K_M	0.64 GiB	690,844,256	4.472	—	mradermacher
I1-Q3_K_M	0.64 GiB	690,844,544	4.472	—	mradermacher
Q3_K_L	0.68 GiB	732,525,152	4.742	—	mradermacher
I1-Q3_K_L	0.68 GiB	732,525,440	4.742	—	mradermacher
I1-IQ4_XS	0.69 GiB	743,142,272	4.811	—	mradermacher
IQ4_XS	0.70 GiB	748,384,864	4.845	—	mradermacher
I1-Q4_0	0.72 GiB	773,026,688	5.004	—	mradermacher
I1-IQ4_NL	0.72 GiB	773,026,688	5.004	—	mradermacher
Q4_K_S	0.72 GiB	775,647,840	5.021	—	mradermacher
I1-Q4_K_S	0.72 GiB	775,648,128	5.021	—	mradermacher
Q4_K_M	0.75 GiB	807,694,944	5.229	—	mradermacher
I1-Q4_K_M	0.75 GiB	807,695,232	5.229	—	mradermacher
I1-Q4_1	0.77 GiB	831,746,944	5.384	—	mradermacher
Q5_K_S	0.83 GiB	892,564,064	5.778	—	mradermacher
I1-Q5_K_S	0.83 GiB	892,564,352	5.778	—	mradermacher
Q5_K_M	0.85 GiB	911,503,968	5.901	—	mradermacher
I1-Q5_K_M	0.85 GiB	911,504,256	5.901	—	mradermacher
Q6_K	0.95 GiB	1,021,801,056	6.615	—	mradermacher
I1-Q6_K	0.95 GiB	1,021,801,344	6.615	—	mradermacher
Q8_0	1.23 GiB	1,321,083,488	8.552	—	mradermacher
F16	2.31 GiB	2,479,596,128	16.052	—	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 Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs Qwen3-1.7B vs embeddinggemma-300m vs Llama-3.2-3B-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 0.65 GiB. The real file is 0.75 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-heretic need?: Q4_K_M is exactly 807,694,944 bytes (0.75 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-heretic'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-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.