ByteDance · video

Bernini-R

ByteDance/Bernini-R

Bernini-R at Q4_K_M is exactly 9,650,100,736 bytes (8.99 GiB / 9.65 GB) — an effective 5.403 bits per weight, not the nominal 4.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
14.3B
Architecture
wan
null layers
Context
—
native (config.json)
License
apache-2.0

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
Q2_K4.94 GiB5,299,329,5362.967—vantagewithai
Q2_K4.94 GiB5,299,329,5362.967—vantagewithai
Q3_K_S6.07 GiB6,513,383,9363.647—vantagewithai
Q3_K_S6.07 GiB6,513,383,9363.647—vantagewithai
Q3_K_M6.68 GiB7,174,478,3364.0171095vantagewithai
Q3_K_M6.68 GiB7,174,478,3364.017—vantagewithai
Q4_07.97 GiB8,556,468,7364.791—vantagewithai
Q4_07.97 GiB8,556,468,7364.7911095vantagewithai
Q4_K_S8.15 GiB8,746,523,1364.897—vantagewithai
Q4_K_S8.15 GiB8,746,523,1364.897—vantagewithai
Q4_18.62 GiB9,257,703,9365.183—vantagewithai
Q4_18.62 GiB9,257,703,9365.183—vantagewithai
Q4_K_M8.99 GiB9,650,100,7365.4031095vantagewithai
Q4_K_M8.99 GiB9,650,100,7365.403—vantagewithai
Q5_K_S9.44 GiB10,135,886,3365.675—vantagewithai
Q5_K_S9.44 GiB10,135,886,3365.675—vantagewithai
Q5_09.60 GiB10,312,833,5365.774—vantagewithai
Q5_09.60 GiB10,312,833,5365.774—vantagewithai
Q5_K_M10.05 GiB10,790,427,1366.0421095vantagewithai
Q5_K_M10.05 GiB10,790,427,1366.042—vantagewithai
Q5_110.26 GiB11,014,068,7366.167—vantagewithai
Q5_110.26 GiB11,014,068,7366.167—vantagewithai
Q6_K11.18 GiB12,002,023,9366.7201095vantagewithai
Q6_K11.18 GiB12,002,023,9366.720—vantagewithai
Q8_014.35 GiB15,404,980,7368.625—vantagewithai
Q8_014.35 GiB15,404,980,7368.6251095vantagewithai
Q4_K_M2 shards17.99 GiB19,319,862,52810.817—FenomAI
Q4_K_M2 shards17.99 GiB19,319,862,52810.817—neuregex
Q5_K_M2 shards20.12 GiB21,600,515,32812.094—neuregex
Q5_K_M2 shards20.12 GiB21,600,515,32812.094—FenomAI
Q8_02 shards28.71 GiB30,829,622,52817.261—FenomAI
Q8_02 shards28.71 GiB30,829,622,52817.261—neuregex

Compare with

same modality, comparable size
vs MiniMax-H3vs LTX-2.3vs Wan2.2-I2V-A14Bvs LTX-2vs LTX2.3-10Erosvs Wan2.2-T2V-A14BAll alternatives →Will it fit on my card? →

Will it run on your card?

full quant x context sweep
By VRAM:8GB12GB16GB24GB32GB48GB
Radeon RX 6500 XT 4GBGeForce RTX 3050 6GBGeForce RTX 5050 8GBGeForce RTX 3080 10GBGeForce RTX 2080 Ti 11GBGeForce RTX 5070 12GBGeForce RTX 5060 Ti 16GBApple M3 Pro 18GBGeForce RTX 3080 Ti 20GBGeForce RTX 5090 D V2 24GBGeForce RTX 5090 D 32GBApple M5 Max 36GBApple M5 Pro 48GBApple M5 Max 64GBApple M3 Ultra 96GBApple M5 Max 128GBApple M2 Ultra 192GBApple M3 Ultra 256GBApple M3 Ultra 512GB

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 7.49 GiB. The real file is 8.99 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
no
Experts
—
Experts per token
—
use_sliding_window
—

Questions people ask

How much VRAM does Bernini-R need?
Q4_K_M is exactly 9,650,100,736 bytes (8.99 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of Bernini-R 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.