Gemma 4 (E2B)
Google's edge-optimized Gemma 4 at ~2B effective parameters. Multimodal (text + image + audio), 256K context, native tool calling. Safe small-model default for any modern laptop.
At a glance
- Parameters
- 2.3B
- Quantization
- Q8_0
- Context window
- 128K tokens
- Approx. size
- 5.0 GB
- Engines
- llama.cpp, MLX, Ollama
- License
- Apache-2.0 (open)
- Version
- v1.1.0 , released 2026-04-26
- Family
- gemma
Tuning
Defaults a gezel client applies out of the box, from real evaluation runs against this exact quantization.
- sampling.temperature
1- sampling.topP
0.95- sampling.topK
64- sampling.repetitionPenalty
1.1- sampling.repetitionContext
20- reasoning.thinkingBudget
96- profiles
- thinking-general, thinking-coding, thinking-precise, instruct, creative
Model behaviors
Client-side behavior modules the catalog enables for this model (reasoning-tag handling, fabrication detection, prompt shaping).
- reasoning.strip-channel-tags
- prompt.private-reasoning-guidance
- prompt.tool-cookbook-full
- fabrication.detect-past-tense-no-tools
- fabrication.detect-claim-without-tool
- turn.auto-acknowledge-tool-errors
- mcp.relax-required-fields
- mcp.default-missing-fields
- turn.continuation-budget
- turn.preamble-folding
- turn.ramble-detection
- parse.gemma-special-token
- prompt.prefer-writefile-edits
- prompt.derive-by-execution
- tools.mlx-grammar
Sources
- llama.cpp (GGUF)
- ggml-org/gemma-4-E2B-it-GGUF ·
gemma-4-E2B-it-Q8_0.gguf· Q8_0 - MLX
- mlx-community/gemma-4-e2b-it-4bit
- Ollama
gemma4:e2b- Upstream
- https://huggingface.co/google/gemma-4-E2B-it
- Manifest
- View on GitHub
Tags
- multimodal
- vision
- audio
- tools
- small