How much VRAM does Qwen3-ASR-1.7B need?

Q4_K_M is exactly 1,319,830,496 bytes (1.23 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

Qwen · audio asr

Qwen3-ASR-1.7B

Qwen/Qwen3-ASR-1.7B

Qwen3-ASR-1.7B at Q4_K_M is exactly 1,319,830,496 bytes (1.23 GiB / 1.32 GB) — an effective 4.495 bits per weight, not the nominal 4.

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

Parameters

2.3B

Architecture

qwen3_asr

null 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
Q4_K_M	1.23 GiB	1,319,830,496	4.495	709	handy-computer
Q4_0	1.39 GiB	1,490,915,200	5.077	710	cstr
Q4_K	1.39 GiB	1,490,915,200	5.077	—	cstr
Q5_K_M	1.41 GiB	1,517,290,464	5.167	709	handy-computer
Q6_K	1.58 GiB	1,692,554,208	5.764	709	handy-computer
Q5_K	1.63 GiB	1,744,867,200	5.942	—	cstr
Q8_0	2.02 GiB	2,165,034,944	7.373	311	ggml-org
Q8_0	2.03 GiB	2,185,030,624	7.441	—	handy-computer
Q8_0	2.33 GiB	2,506,723,200	8.536	—	cstr
BF16	3.79 GiB	4,069,674,944	13.859	—	ggml-org
BF16	3.80 GiB	4,083,087,904	13.905	—	handy-computer
F16	3.81 GiB	4,091,390,944	13.933	709	handy-computer
F16	4.38 GiB	4,704,800,576	16.022	710	cstr

Measured

published by a third party, attributed below

Metric	Value	What it means
rtf	394.1
RTFx	394.1	Higher is better — audio seconds processed per second of compute.
Word error rate	1.24%	Lower is better — the share of words transcribed incorrectly.
Word error rate	9.88%	Lower is better — the share of words transcribed incorrectly.
Word error rate	8.31%	Lower is better — the share of words transcribed incorrectly.
Word error rate	2.92%	Lower is better — the share of words transcribed incorrectly.
Word error rate	2.58%	Lower is better — the share of words transcribed incorrectly.
Word error rate	3.01%	Lower is better — the share of words transcribed incorrectly.
Word error rate	9.26%	Lower is better — the share of words transcribed incorrectly.
Word error rate	7.25%	Lower is better — the share of words transcribed incorrectly.
Word error rate	5.99%	Lower is better — the share of words transcribed incorrectly.
Word error rate	7.22%	Lower is better — the share of words transcribed incorrectly.
Word error rate	5.02%	Lower is better — the share of words transcribed incorrectly.

Benchmarked· by open-asr-leaderboard-english-short-latest

RTFx measured by the Open ASR Leaderboard on a single datacenter GPU at a large batch size. It ranks models against each other; it says nothing about throughput on consumer hardware. We reproduce these figures with attribution; they are not ours and we have not verified the runs. Source: open-asr-leaderboard-english-short-latest.

Compare with

same modality, comparable size

vs nemotron-3.5-asr-streaming-0.6b vs parakeet-unified-en-0.6b vs cohere-transcribe-03-2026 vs parakeet-tdt-0.6b-v3 vs whisper-medium vs Voxtral-Mini-4B-Realtime-2602 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 1.23 GiB. The real file is 1.23 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

—

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does Qwen3-ASR-1.7B need?: Q4_K_M is exactly 1,319,830,496 bytes (1.23 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of Qwen3-ASR-1.7B 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.