2 Apr 2026
gemma-4-E4B-it
More effective capacity, with tables still to place
8B reported total · text + image + video + audio
Gemma / Dense
gemma-4-E2B-it
“Effective” is a compute label, not the download size
New to model sizes or GPU memory? Start with weights, parameters and quantization →
E2B is interesting for constrained devices because it separates effective computation from per-layer embedding storage. Read its 2.3B effective and 5.1B inclusive counts together before selecting hardware.
5.1Breported total parameters
Denseactive parameters per token
131,072configured token positions
Includes per-layer embeddings; effective compute is smaller. Not a MoE active-parameter count.
Evaluate it for
Small-device text, vision and audio experiments with a supported Gemma runtime.
Choose another path when
A budget that reserves memory for only two billion parameters or copies a text-only conversion’s footprint onto the full model.
Release context
Why this release matters
Per-layer embedding tables are the reason the E label differs from a conventional dense parameter label. Placement and quantization of those tables can change the practical footprint substantially.
Deployment starting points
Which hardware, how many, at what precision?
Start with an exact artifact and an engine that supports it. A configuration below is evidence of a documented path; its device count does not promise a particular throughput or concurrency.
No certified hardware configuration in this edition
A budget that reserves memory for only two billion parameters or copies a text-only conversion’s footprint onto the full model.
The pinned model card establishes the architecture. It does not, by itself, prove that a particular GPU count runs this checkpoint at your target context. The capacity screen below helps rule out undersized allocations; a successful load and workload test are still required.
Capacity screen: would the weights fit?
This estimates inference memory for a hypothetical uniform precision. It is useful for rejecting an allocation that is too small. It does not establish a working deployment or estimate training memory.
A uniform-precision GPU count would hide this model’s component layout. Includes per-layer embeddings; effective compute is smaller. Not a MoE active-parameter count. Use the documented artifact and placement path above; no GPU count is inferred here.
Architecture in practice
What the serving engine has to do
Shared tables and request cache have different lifetimes.
Model tables do not multiply with every request.
The lookup and host/device placement are model-specific.
Mechanism schematic based on the pinned model card and configuration. It explains a design principle; it is not a full implementation graph.
Inspect the full checkpoint specifications
| Exact checkpoint | google/gemma-4-E2B-it |
|---|---|
| Text model type | gemma4_text |
| Layers | 35 |
| Hidden width | 1,536 |
| Attention / KV heads | 8 / 1 |
| Head dimension | 256 |
| Routed / selected experts | Not verified / Not verified |
| Configured positions | 131,072 |
| Documented extension | Not recorded for this checkpoint |
| Inputs → output | text, image, video, audio → text |
| License metadata | apache-2.0 |
28 sliding attention layers; 7 full attention layers. These counts describe the configured pattern.
Open weights do not imply unrestricted use. Read the applicable license terms linked from the pinned card.
A useful comparison
What else belongs on the shortlist?
E4B tests additional capacity with the same design; Qwen2B provides a more conventional small-model comparison.
2 Mar 2026
Qwen3.5-2B
A small step up when the tiniest model misses too much
2B reported total · text + image + video
These are editorial comparison candidates. We have not run a matched quality or serving benchmark, so this is not a ranking.
Prepare to explain it
The interview lesson
Contrast shared model tables with per-request cache. Explain why offloading the former and the latter solve different problems.
Attention layouts · Quantization · Courses and worked examples
For a deployment evaluation, record exact weights, precision, engine, device count, interconnect, prompt/output lengths and concurrency. Report task success, errors, TTFT, TPOT and useful throughput together.
Sources and citation
Our assessment is editorial judgment based on the linked architecture and deployment evidence. Model facts are publisher-reported or attributed to runtime maintainers; no independent GPU benchmark was run.
AI Infra Interviews, “gemma-4-E2B-it: release, architecture and deployment”, checked 2026-09-13. Preserve this date and the exact checkpoint when citing.