gemma-4-E2B-it

Home/Models/google/gemma-4-E2B-it

gemma-4-E2B-it

google/gemma-4-E2B-it

kind: original
quant: bf16 (16-bit)
family: gemma
modality: any-to-any
root model: google/gemma-4-E2B
provider: google
license: apache-2.0
language: —
context: 128K
library: transformers
vRAM (GB): —
size on disk: —
downloads: 3,856,506
likes: 845
created: —
last modified: 2026-07-20T16:41:56+00:00 (2026)
huggingface: https://huggingface.co/google/gemma-4-E2B-it
tags: transformerssafetensorsgemma4image-text-to-textany-to-anyarxiv:2607.02770base_model:google/gemma-4-E2Bbase_model:finetune:google/gemma-4-E2Blicense:apache-2.0eval-resultsendpoints_compatibledeploy:azuredeploy:sagemakerregion:us

Benchmark scores (10)

benchmark	category	score	context	conf	evidence
AIME 2026	math	89.2	AIME 2026 no tools	medium	`\| AIME 2026 no tools \| 89.2% \| 88.3% \| 77.5% \| 42.5% \| 37.5% \| 20.8% \|`
MATH	math	85.6	MATH-Vision	medium	`\| MATH-Vision \| 85.6% \| 82.4% \| 79.7% \| 59.5% \| 52.4% \| 46.0% \|`
MMLU-Pro	knowledge	85.2	MMLU Pro	medium	`\| MMLU Pro \| 85.2% \| 82.6% \| 77.2% \| 69.4% \| 60.0% \| 67.6% \|`
GPQA Diamond	reasoning	84.3	GPQA Diamond	medium	`\| GPQA Diamond \| 84.3% \| 82.3% \| 78.8% \| 58.6% \| 43.4% \| 42.4% \|`
LiveCodeBench	coding	80.0	LiveCodeBench v6	medium	`\| LiveCodeBench v6 \| 80.0% \| 77.1% \| 72.0% \| 52.0% \| 44.0% \| 29.1% \|`
MMMU	multimodal	76.9	MMMU Pro	medium	`\| MMMU Pro \| 76.9% \| 73.8% \| 69.1% \| 52.6% \| 44.2% \| 49.7% \|`
BigBench	reasoning	74.4	BigBench Extra Hard	medium	`\| BigBench Extra Hard \| 74.4% \| 64.8% \| 53.0% \| 33.1% \| 21.9% \| 19.3% \|`
Humanity's Last Exam	knowledge	26.5	HLE with search	medium	`\| HLE with search \| 26.5% \| 17.2% \| - \| - \| - \| - \|`
Humanity's Last Exam	knowledge	19.5	HLE no tools	medium	`\| HLE no tools \| 19.5% \| 8.7% \| 5.2% \| - \| - \| - \|`
WER FLEURS	speech-asr	8.0	FLEURS (lower is better)	medium	`\| FLEURS (lower is better) \| - \| - \| 0.069<sup>*</sup> \| 0.08 \| 0.09 \| - \|`

Discussions (43)

title	author	status	type
Update chat_template.jinja	AlexHung29629	closed	disc
Update README	MaartenGr	merged	disc
Raise per-image vision soft-token budget from 280 to 1120	lucianommartins	open	disc
Add response_template to tokenizer_config.json	Rocketknight1	merged	disc
Add IFStruct v1.0 evaluation result	SaylorTwift	open	disc
Add Vaani-Benchmark-V1.0 eval results	PranavdbhatArtpark	open	disc
Appreciation Post	Maani	closed	disc
Chat template may re-inject prior-turn reasoning during multi-turn tool use → repetition loops	ManniX-ITA	open	disc
fix: chat template — null handling, reasoning preservation, turn-tag balance, input validation	lucianommartins	merged	disc
Add EvalEval community eval results	EvalEvalBot	closed	disc
fun config	baseten-admin	open	disc
Add MMMU-Pro evaluation result	SaylorTwift	open	disc
Fix lineage	osanseviero	merged	disc
How to best prompt for transcription results when sending audio input?	DiffusionFanatic1	open	disc
GGUF + pure-C++ runtime in CrispASR — Gemma-4-E2B (ASR)	cstr	open	disc
fix(chat_template): Emit multimodal placeholders in tool response content-parts	harshaljanjani	merged	disc
[Feature request] Eliminate pre-attention RMSNorm in Gemma 4 via scale invariance + weight folding	graefics	open	disc
fix(chat_template): update SI and tool call handling	dougreid	merged	disc
fix(chat template): better SI and tool call handling	dougreid	draft	disc
[Bug] Chat template redacting reasoning traces for non-tool related use case	vibhas09	open	disc
Why k/v proj exist in layers where "is_kv_shared_layer = True"?	jiayuanm	open	disc
Issue loading google/gemma-4-E2B-it with AutoProcessor (Unrecognized processing class)	dipakmeher51	open	disc
Error loading Gemma 4 from official notebook	melvindave	closed	disc
regarding of supported image token budgets	J22	open	disc
Request: DOI	HoumanAl	closed	disc
Update README.md	Highgroundbkk	open	disc
Litert version after fine-tuning	Abhijitmore	open	disc
Welcome G4! Any new T5-G4 coming?	iSurpriseMe	open	disc
Update response schema	Rocketknight1	merged	disc
Request	mlboydaisuke	open	disc
fix: prefix-preserving tool calls + dialog compliance	lucianommartins	merged	disc
fix: function calling formatting in chat template	RyanMullins	open	disc
Update README.md	Tialynnsitz	closed	disc
Allucination parsing JSON responses containing numerical data such as 1234.00	jeisonvendetta	open	disc
Benchmarking low resource languages	indiaiquest	open	disc
Embed chat_template in tokenizer_config.json	piero-atelico	open	disc
Embed chat_template in tokenizer_config.json	piero-atelico	closed	disc
Issue with llama.cpp	kelkalot	closed	disc
Welcome G4! Any new T5-G4 coming?	JohnGenetica	open	disc
Add AIME 2026 evaluation result	SaylorTwift	open	disc
ordl-gemma-family	Winsocket	closed	disc
Upload 2 files	merve	open	disc
Proposed video configuration	pcuenq	merged	disc