How much VRAM does Mistral-7B-v0.3 need?

Q4_K_M is exactly 4,372,811,680 bytes (4.07 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is Mistral-7B-v0.3's KV cache?

4.00 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.

mistralai · text

Mistral-7B-v0.3

mistralai/Mistral-7B-v0.3

Mistral-7B-v0.3 at Q4_K_M is exactly 4,372,811,680 bytes (4.07 GiB / 4.37 GB) — an effective 4.827 bits per weight, not the nominal 4. Its KV cache at 32K is 4.00 GiB.

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

Parameters

7.2B

Architecture

llama

32 layers

Context

32,768

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ1_S	1.50 GiB	1,615,318,720	1.783	—	legraphista
IQ1_M	1.64 GiB	1,757,662,912	1.940	—	legraphista
IQ2_XXS	1.86 GiB	1,994,903,232	2.202	—	legraphista
IQ2_XS	2.05 GiB	2,201,472,704	2.430	—	legraphista
IQ2_XS	2.05 GiB	2,201,481,536	2.430	—	bartowski
IQ2_S	2.16 GiB	2,314,456,768	2.555	—	legraphista
IQ2_S	2.16 GiB	2,314,466,432	2.555	—	bartowski
IQ2_M	2.33 GiB	2,504,249,024	2.764	—	legraphista
IQ2_M	2.33 GiB	2,504,258,688	2.764	—	bartowski
Q2_K_S	2.36 GiB	2,532,560,576	2.795	—	legraphista
Q2_K	2.54 GiB	2,722,877,120	3.005	—	legraphista
Q2_K	2.54 GiB	2,722,877,344	3.005	—	mradermacher
Q2_K	2.54 GiB	2,722,887,040	3.005	—	bartowski
IQ3_XXS	2.64 GiB	2,830,880,448	3.125	—	legraphista
IQ3_XXS	2.64 GiB	2,830,890,112	3.125	—	bartowski
IQ3_XS	2.82 GiB	3,022,769,856	3.336	—	legraphista
IQ3_XS	2.82 GiB	3,022,770,080	3.336	—	mradermacher
IQ3_XS	2.82 GiB	3,022,780,608	3.336	—	bartowski
Q3_K_S	2.95 GiB	3,168,521,920	3.497	—	legraphista
Q3_K_S	2.95 GiB	3,168,522,144	3.497	—	mradermacher
Q3_K_S	2.95 GiB	3,168,532,672	3.497	—	bartowski
IQ3_S	2.97 GiB	3,186,347,712	3.517	—	legraphista
IQ3_S	2.97 GiB	3,186,347,936	3.517	—	mradermacher
IQ3_M	3.06 GiB	3,288,846,016	3.630	—	legraphista
IQ3_M	3.06 GiB	3,288,846,240	3.630	—	mradermacher
IQ3_M	3.06 GiB	3,288,856,768	3.630	—	bartowski
Q3_K	3.28 GiB	3,522,940,608	3.888	—	legraphista
Q3_K_M	3.28 GiB	3,522,940,832	3.888	—	mradermacher
Q3_K_M	3.28 GiB	3,522,951,360	3.889	—	bartowski
Q3_K_L	3.56 GiB	3,825,979,072	4.223	—	legraphista
Q3_K_L	3.56 GiB	3,825,979,296	4.223	—	mradermacher
Q3_K_L	3.56 GiB	3,825,989,824	4.223	—	bartowski
IQ4_XS	3.64 GiB	3,911,962,304	4.318	—	legraphista
IQ4_XS	3.64 GiB	3,911,973,888	4.318	—	bartowski
IQ4_XS	3.68 GiB	3,948,662,688	4.358	—	mradermacher
IQ4_NL	3.85 GiB	4,130,066,112	4.559	—	legraphista
Q4_K_S	3.86 GiB	4,144,746,176	4.575	—	legraphista
Q4_K_S	3.86 GiB	4,144,746,400	4.575	—	mradermacher
Q4_K_S	3.86 GiB	4,144,758,016	4.575	—	bartowski
Q4_K	4.07 GiB	4,372,811,456	4.827	—	legraphista

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.50 GiB	0.50 GiB	—	32 / 0 / 0
8,192	1.00 GiB	1.00 GiB	—	32 / 0 / 0
16,384	2.00 GiB	2.00 GiB	—	32 / 0 / 0
32,768	4.00 GiB	4.00 GiB	—	32 / 0 / 0
65,536	8.00 GiB	8.00 GiB	—	32 / 0 / 0
131,072	16.00 GiB	16.00 GiB	—	32 / 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 Qwen3-VL-8B-Instruct-abliterated-v1 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 3.80 GiB. The real file is 4.07 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

4096

Vocab

32,768

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Mistral-7B-v0.3 need?: Q4_K_M is exactly 4,372,811,680 bytes (4.07 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Mistral-7B-v0.3's KV cache?: 4.00 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 Mistral-7B-v0.3 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.