How much VRAM does Falcon3-7B-Instruct need?

Q4_K_M is exactly 4,569,726,368 bytes (4.26 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Falcon3-7B-Instruct's KV cache?

3.50 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.

tiiuae · text

Falcon3-7B-Instruct

tiiuae/Falcon3-7B-Instruct

Falcon3-7B-Instruct at Q4_K_M is exactly 4,569,726,368 bytes (4.26 GiB / 4.57 GB) — an effective 4.903 bits per weight, not the nominal 4. Its KV cache at 32K is 3.50 GiB.

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

Parameters

7.5B

Architecture

llama

28 layers

Context

32,768

native (config.json)

License

other

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ2_M	2.47 GiB	2,647,145,920	2.841	—	bartowski
Q2_K	2.69 GiB	2,886,983,072	3.098	—	tiiuae
Q2_K	2.69 GiB	2,886,983,104	3.098	—	bartowski
IQ3_XS	2.99 GiB	3,210,354,112	3.445	—	bartowski
Q2_K_L	3.05 GiB	3,280,199,104	3.520	—	bartowski
Q3_K_S	3.14 GiB	3,366,215,104	3.612	—	bartowski
IQ3_M	3.20 GiB	3,441,208,768	3.692	—	bartowski
Q3_K_M	3.43 GiB	3,685,850,528	3.955	—	tiiuae
Q3_K_M	3.43 GiB	3,685,850,560	3.955	—	bartowski
Q3_K_L	3.70 GiB	3,967,983,040	4.258	—	bartowski
IQ4_XS	3.80 GiB	4,082,507,200	4.381	—	bartowski
Q4_0	4.01 GiB	4,302,904,736	4.617	—	tiiuae
IQ4_NL	4.01 GiB	4,302,904,768	4.617	—	bartowski
Q4_0	4.02 GiB	4,316,175,808	4.631	—	bartowski
Q4_K_S	4.03 GiB	4,331,019,712	4.647	—	bartowski
Q4_K_M	4.26 GiB	4,569,726,368	4.903	—	tiiuae
Q4_K_M	4.26 GiB	4,569,726,400	4.903	—	bartowski
Q4_K_L	4.53 GiB	4,868,570,560	5.224	—	bartowski
Q5_0	4.83 GiB	5,184,495,008	5.563	—	tiiuae
Q5_K_S	4.83 GiB	5,184,495,040	5.563	—	bartowski
Q5_K_M	4.96 GiB	5,321,948,576	5.711	—	tiiuae
Q5_K_M	4.96 GiB	5,321,948,608	5.711	—	bartowski
Q5_K_L	5.19 GiB	5,570,461,120	5.977	—	bartowski
Q6_K	5.70 GiB	6,121,184,672	6.568	—	tiiuae
Q6_K	5.70 GiB	6,121,184,704	6.568	—	bartowski
Q6_K_L	5.88 GiB	6,316,219,840	6.777	—	bartowski
Q8_0	7.38 GiB	7,926,783,392	8.506	—	tiiuae
Q8_0	7.38 GiB	7,926,783,424	8.506	—	bartowski
F16	13.89 GiB	14,916,197,568	16.006	—	bartowski
F16	13.89 GiB	14,916,197,792	16.006	—	tiiuae

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.44 GiB	0.44 GiB	—	28 / 0 / 0
8,192	0.88 GiB	0.88 GiB	—	28 / 0 / 0
16,384	1.75 GiB	1.75 GiB	—	28 / 0 / 0
32,768	3.50 GiB	3.50 GiB	—	28 / 0 / 0
65,536	7.00 GiB	7.00 GiB	—	28 / 0 / 0
131,072	14.00 GiB	14.00 GiB	—	28 / 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 3.91 GiB. The real file is 4.26 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

256

Hidden size

3072

Vocab

131,072

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Falcon3-7B-Instruct need?: Q4_K_M is exactly 4,569,726,368 bytes (4.26 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Falcon3-7B-Instruct's KV cache?: 3.50 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 Falcon3-7B-Instruct 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.