How much VRAM does whisper-base need?

Q4_K_M is exactly 58,870,848 bytes (0.05 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

openai · audio asr

whisper-base

openai/whisper-base

whisper-base at Q4_K_M is exactly 58,870,848 bytes (0.05 GiB / 0.06 GB) — an effective 6.488 bits per weight, not the nominal 4.

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

Parameters

73M

Architecture

whisper

6 layers

Context

—

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q2_K	0.03 GiB	29,925,346	3.298	—	oxide-lab
Q3_K	0.03 GiB	37,095,158	4.088	—	oxide-lab
Q4_0	0.04 GiB	42,225,024	4.653	—	oxide-lab
Q4_0	0.04 GiB	46,471,066	5.121	—	oxide-lab
Q4_K	0.04 GiB	46,471,066	5.121	—	oxide-lab
Q4_1	0.04 GiB	46,705,312	5.147	—	oxide-lab
Q4_1	0.05 GiB	50,883,258	5.607	—	oxide-lab
Q5_K	0.05 GiB	55,295,450	6.094	—	oxide-lab
Q5_0	0.05 GiB	55,295,450	6.094	—	oxide-lab
Q4_K_M	0.05 GiB	58,870,848	6.488	—	handy-computer
Q5_1	0.06 GiB	59,707,642	6.580	—	oxide-lab
Q5_K_M	0.06 GiB	63,786,048	7.029	—	handy-computer
Q6_K	0.06 GiB	64,671,358	7.127	—	oxide-lab
Q6_K	0.06 GiB	67,865,664	7.479	—	handy-computer
Q8_0	0.07 GiB	78,067,328	8.603	—	oxide-lab
Q8_0	0.08 GiB	81,768,602	9.011	—	oxide-lab
Q8_0	0.08 GiB	84,962,880	9.363	—	handy-computer
F16	0.14 GiB	151,145,760	16.657	—	handy-computer
F32	0.27 GiB	292,335,904	32.216	—	handy-computer

Compare with

same modality, comparable size

vs whisper-tiny vs whisper-base.en vs moonshine-streaming-tiny vs moonshine-base 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 0.04 GiB. The real file is 0.05 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

—

Vocab

51,865

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does whisper-base need?: Q4_K_M is exactly 58,870,848 bytes (0.05 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of whisper-base 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.