model family

Gemma 4

Published by Google. 9 members have been measured on real machines. Every number below was signed by the machine that produced it — none of it is from a model card.

what it takes, what it returns
image + text → texttext → text

Declared by the network’s catalog, not measured — a signature says what this family would be asked. Whether it answered is the signed capability flags on each member.

gemma4:e2b5 machines · 5.1B · Q4_K_M
8,192recalled
32,768served
45.1tok/s
hf.co/unsloth/gemma-4-E2B-it-GGUF:Q3_K_M3 machines · 4.65B · Q3_K_M
8,192recalled
—served
38.4tok/s
hf.co/unsloth/gemma-4-E2B-it-GGUF:IQ4_NL2 machines · 4.65B · IQ4_NL
8,192recalled
—served
54.9tok/s
hf.co/unsloth/gemma-4-E2B-it-GGUF:Q4_K_M2 machines · 4.65B · Q4_K_M
8,192recalled
8,192served
11.3tok/s
hf.co/unsloth/gemma-4-E2B-it-GGUF:Q4_02 machines · 4.65B · Q4_0
4,096recalled
—served
57.6tok/s
hf.co/unsloth/gemma-4-E4B-it-GGUF:Q3_K_M1 machine
4,096recalled
—served
20.4tok/s
gemma4:e2b-it-qat1 machine
—recalled
—served
8.9tok/s
gemma4:e4b-it-qat1 machine
—recalled
—served
—tok/s
google/gemma-4-E2B-it-qat-q4_0-gguf1 machine
—recalled
—served
2.0tok/s

Builds this network knows of

Declared, not measured — this is what the catalog says exists and what it would cost a machine to hold. Nothing here is signed, and none of it claims the build works.

Gemma 4 12B12B
hf.co/unsloth/gemma-4-12b-it-GGUF:IQ2_MIQ2_Mollama3.9 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-12b-it-GGUF:Q3_K_MQ3_K_Mollama5.3 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-12B-it-qat-GGUF:Q4_K_XLQ4_K_XLollama6.3 GiB on diskquantized by unsloth
mlx-community/gemma-4-12b-it-4bit4bitmlx6.8 GiB on diskquantized by mlx-community
hf.co/yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF:Q4_K_MQ4_K_Mollama6.9 GiB on diskderived by yuxinlu1
hf.co/unsloth/gemma-4-12B-it-qat-GGUF:Q4_K_MQ4_K_Mollama7.2 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-12b-it-GGUF:Q6_KQ6_Kollama9.1 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-12b-it-GGUF:BF16BF16ollama22.2 GiB on diskquantized by unsloth
Gemma 4 26B-A4B
hf.co/unsloth/gemma-4-26B-A4B-it-GGUF:IQ2_XXSIQ2_XXSollama9.2 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-26B-A4B-it-GGUF:IQ3_SIQ3_Sollama10.5 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-26B-A4B-it-GGUF:IQ4_XSIQ4_XSollama12.7 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-26B-A4B-it-GGUF:Q4_K_MQ4_K_Mollama15.8 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-26B-A4B-it-GGUF:Q5_K_XLQ5_K_XLollama19.8 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-26B-A4B-it-GGUF:Q8_K_XLQ8_K_XLollama25.7 GiB on diskquantized by unsloth
Gemma 4 31B
hf.co/unsloth/gemma-4-31B-it-GGUF:IQ2_XXSIQ2_XXSollama7.9 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-31B-it-GGUF:Q3_K_SQ3_K_Sollama12.3 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-31B-it-GGUF:IQ4_NLIQ4_NLollama16.1 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-31B-it-GGUF:Q4_K_MQ4_K_Mollama17.1 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-31B-it-GGUF:Q5_K_MQ5_K_Mollama20.2 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-31B-it-GGUF:Q8_K_XLQ8_K_XLollama32.6 GiB on diskquantized by unsloth
Gemma 4 E2B5.1B
hf.co/unsloth/gemma-4-E2B-it-GGUF:IQ2_MIQ2_Mollama2.1 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-E2B-it-GGUF:Q3_K_MQ3_K_Mollama2.4 GiB on diskquantized by unsloth
mlx-community/gemma-4-E2B-it-4bit4bitmlx2.8 GiB on diskquantized by mlx-community
hf.co/unsloth/gemma-4-E2B-it-GGUF:Q4_0Q4_0ollama2.8 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-E2B-it-GGUF:Q4_K_XLQ4_K_XLollama3.0 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-E2B-it-GGUF:Q4_K_MQ4_K_Mollama3.8 GiB on diskquantized by unsloth
unsloth/gemma-4-E2B-it-GGUF:gemma-4-E2B-it-Q4_K_M.ggufQ4_K_Mllamacpp3.8 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-E2B-it-GGUF:Q6_KQ6_Kollama4.2 GiB on diskquantized by unsloth
gemma4:e2bQ4_K_Mollama7.0 GiB on diskpublished by google
hf.co/unsloth/gemma-4-E2B-it-GGUF:BF16BF16ollama8.7 GiB on diskquantized by unsloth
Gemma 4 E4B8B
hf.co/unsloth/gemma-4-E4B-it-qat-GGUF:Q2_K_XLQ2_K_XLollama3.0 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-4-E4B-it-GGUF:Q3_K_MQ3_K_Mollama3.8 GiB on diskquantized by unsloth
mlx-community/gemma-4-E4B-it-4bit4bitmlx4.5 GiB on diskquantized by mlx-community
hf.co/unsloth/gemma-4-E4B-it-GGUF:Q4_0Q4_0ollama4.5 GiB on diskquantized by unsloth
hf.co/lmstudio-community/gemma-4-E4B-it-GGUF:Q4_K_MQ4_K_Mollama5.0 GiB on diskquantized by lmstudio-community
hf.co/unsloth/gemma-4-E4B-it-GGUF:Q6_KQ6_Kollama6.6 GiB on diskquantized by unsloth
gemma4:e4bQ4_K_Mollama10.7 GiB on diskpublished by google
hf.co/unsloth/gemma-4-E4B-it-GGUF:BF16BF16ollama14.0 GiB on diskquantized by unsloth
ask your own gram

Your concierge can answer this from the network rather than from this page. The ask carries the operator id, so it resolves to a capability and an engine — not to a name somebody has to recognise.

What has this network measured about the Gemma 4 family (epn.family.gemma-4), and can my gram run any of it? See /inferences/gemma-4

open a gramx with this ask →epn.family.gemma-4