How much VRAM does Llama-4-Scout-17B-16E-Instruct-abliterated-v2 need?

I1-IQ1_S is exactly 22,180,447,328 bytes (20.66 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

Is Llama-4-Scout-17B-16E-Instruct-abliterated-v2 a mixture-of-experts model?

Yes — 16 experts with 1 routed per token. Every expert must be resident in memory, but only the routed ones are read per token, so its memory requirement and its speed behave very differently.

jiangchengchengNLP · text · mixture of experts

Llama-4-Scout-17B-16E-Instruct-abliterated-v2

jiangchengchengNLP/Llama-4-Scout-17B-16E-Instruct-abliterated-v2

Llama-4-Scout-17B-16E-Instruct-abliterated-v2 at I1-IQ1_S is exactly 22,180,447,328 bytes (20.66 GiB / 22.18 GB) — an effective 1.633 bits per weight, not the nominal 1.

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

Parameters

109B

total, not active

Architecture

llama4

48 layers

Context

10,485,760

native (config.json)

License

—

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
I1-IQ1_S	20.66 GiB	22,180,447,328	1.633	—	mradermacher
I1-IQ1_M	22.88 GiB	24,572,183,648	1.809	—	mradermacher
I1-IQ2_XXS	26.60 GiB	28,558,410,848	2.103	—	mradermacher
I1-IQ2_XS	29.60 GiB	31,786,714,208	2.341	—	mradermacher
I1-IQ2_S	30.07 GiB	32,285,978,208	2.377	—	mradermacher
I1-IQ2_M	33.04 GiB	35,474,959,968	2.612	—	mradermacher
I1-Q2_K_S	34.42 GiB	36,962,186,848	2.722	—	mradermacher
Q2_K	36.85 GiB	39,563,310,432	2.913	—	mradermacher
I1-Q2_K	36.85 GiB	39,563,310,688	2.913	—	mradermacher
I1-IQ3_XXS	38.80 GiB	41,658,281,568	3.068	—	mradermacher
I1-IQ3_XS	41.25 GiB	44,296,527,968	3.262	—	mradermacher
Q3_K_S	43.53 GiB	46,740,365,152	3.442	—	mradermacher
I1-Q3_K_S	43.53 GiB	46,740,365,408	3.442	—	mradermacher
I1-IQ3_S	43.56 GiB	46,773,788,768	3.444	—	mradermacher
I1-IQ3_M	44.25 GiB	47,509,102,688	3.498	—	mradermacher

KV cache by context

unresolved

This model declares a 8,192-token sliding window, but we could not establish which layers use it. Its architecture publishes the layout as a per-layer array inside the model file rather than as a period in config.json, and we have not yet ingested that array.

A flat context × layers × heads figure would be substantially too high, so we are not showing one. This is tracked as a known gap rather than filled with a guess.

Compare with

same modality, comparable size

vs Qwen3-Coder-30B-A3B-Instruct vs Qwen3.6-27B vs Qwen3.8-27B vs gemma-4-E4B-it vs Qwen3-30B-A3B-Thinking-2507 vs Qwen3-8B All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts I1-IQ1_S at roughly 56.91 GiB. The real file is 20.66 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

5120

Vocab

202,048

Sliding window

8192

SWA period

—

MLA

Experts

Experts per token

use_sliding_window

—

Questions people ask

How much VRAM does Llama-4-Scout-17B-16E-Instruct-abliterated-v2 need?: I1-IQ1_S is exactly 22,180,447,328 bytes (20.66 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Is Llama-4-Scout-17B-16E-Instruct-abliterated-v2 a mixture-of-experts model?: Yes — 16 experts, 1 routed per token. Every expert must be resident, but only the routed ones are read per token, which is why its memory requirement and its speed behave very differently.
Which quantization of Llama-4-Scout-17B-16E-Instruct-abliterated-v2 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.