unsloth · text

gemma-4-E4B-it

unsloth/gemma-4-E4B-it

gemma-4-E4B-it at Q4_K_M is exactly 4,977,164,544 bytes (4.64 GiB / 4.98 GB) — an effective 4.980 bits per weight, not the nominal 4. Its KV cache at 32K is 0.51 GiB, not the 2.63 GiB a flat formula predicts.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
8.0B
Architecture
gemma4
42 layers
Context
131,072
native (config.json)
License
gemma

Shipped quantizations

exact bytes, summed from published files
QuantSizeExact bytesEffective bpwTensorsPublisher
UD-IQ2_M3.29 GiB3,530,227,9683.532dahara1
UD-IQ3_XXS3.45 GiB3,702,259,9683.704dahara1
Q3_K_S3.60 GiB3,862,372,6083.864dahara1
Q3_K_M3.78 GiB4,058,130,6884.060dahara1
IQ4_XS4.39 GiB4,715,409,6644.718dahara1
IQ4_NL4.50 GiB4,835,832,0644.838dahara1
Q4_04.50 GiB4,835,995,9044.838dahara1
Q4_K_S4.51 GiB4,844,843,2644.847dahara1
Q4_K_M4.64 GiB4,977,164,5444.980dahara1
Q4_14.73 GiB5,074,383,1045.077dahara1
Q4_K_M4.97 GiB5,335,275,9365.338pankajpandey-dev
Q4_K_M4.97 GiB5,335,290,2405.338trjxter
Q5_K_S5.03 GiB5,404,848,3845.407dahara1
Q5_K_M5.11 GiB5,481,791,7445.484dahara1
Q5_K_M5.37 GiB5,762,898,3365.766pankajpandey-dev
Q5_K_M5.37 GiB5,762,912,6405.766trjxter
Q6_K5.79 GiB6,217,261,4406.220trjxter
Q6_K5.95 GiB6,392,299,7766.395dahara1
Q8_07.48 GiB8,031,226,2728.035pankajpandey-dev
Q8_07.48 GiB8,031,240,5768.035trjxter
Q8_07.63 GiB8,192,946,4328.197dahara1

KV cache by context

computed per layer — this model uses sliding-window attention
ContextKV cache (f16)Flat formulaOverstated byFull / windowed / recurrent
4,0960.12 GiB0.33 GiB2.67×7 / 35 / 0
8,1920.18 GiB0.66 GiB3.69×7 / 35 / 0
16,3840.29 GiB1.31 GiB4.57×7 / 35 / 0
32,7680.51 GiB2.63 GiB5.19×7 / 35 / 0
65,5360.94 GiB5.25 GiB5.57×7 / 35 / 0
131,0721.82 GiB10.50 GiB5.77×7 / 35 / 0

35 of 42 layers cache only a 512-token window rather than the full context, on a period of . Figures assume the default configuration; --swa-full disables the saving entirely.

Compare with

same modality, comparable size

Will it run on your card?

full quant x context sweep

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 4.19 GiB. The real file is 4.64 GiB, because a quantization is a mixture and some tensors are always kept at higher precision. The larger discrepancy is the cache: a flat formula gives 2.63 GiB at 32K context where the real figure is 0.51 GiB, because most of this model's layers cache a fixed window rather than the whole context.

Architecture

from config.json
Layers
42
Attention heads
8
KV heads
2
Head dim
256
Hidden size
2560
Vocab
262,144
Sliding window
512
SWA period
MLA
no
Experts
Experts per token
use_sliding_window

Questions people ask

How much VRAM does gemma-4-E4B-it need?
Q4_K_M is exactly 4,977,164,544 bytes (4.64 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is gemma-4-E4B-it's KV cache?
0.51 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 gemma-4-E4B-it 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.