Gemma 4 E2B-it GGUF — Quantized by BatiAI

Optimized GGUF quantizations of google/gemma-4-E2B-it — Google's tiniest fully-multimodal Gemma 4 (text + image + audio) for on-device AI on Mac. Built directly from official Google BF16 weights by BatiAI for BatiFlow.

The "E" stands for Edge. Tiny enough to run on basically any Mac (8 GB+) while still understanding speech and images alongside text — the only Gemma 4 size with audio support shared with E4B.

Quick Start

# Recommended
ollama pull batiai/gemma4-e2b:q4

# Higher quality
ollama pull batiai/gemma4-e2b:q6

Available Quantizations

Tag	Quant	Size	Recommended For
`:q4`	Q4_K_M	~2.1 GB	balanced (recommended default)
`:q6`	Q6_K	~2.8 GB	higher quality, near-lossless

Sub-3 B models stay sharp at Q4 — quantization noise is tiny relative to model capacity.

Two modes — text-only by default, multimodal opt-in

Upstream Gemma 4 E2B-it is fully multimodal — text + image + audio in one model. In the GGUF ecosystem this is delivered as two files: a main model.gguf (text tower) and a separate mmproj.gguf that holds both vision and audio encoders together (a single 1411-tensor projector covering image and speech input).

	Text-only (default)	Multimodal (opt-in)
Files needed	main GGUF only	main GGUF + `mmproj-BF16.gguf`
Capabilities	Q&A, coding, tool calling, agents	+ image (OCR, captioning, visual reasoning) + audio (speech understanding)
`ollama pull`	✅ single command	⚠ Ollama mmproj integration is still rough — use llama.cpp directly
Disk / RAM	smaller (no projector weights)	larger (+ ~940 MB)
Recommended for	most users (chat, code, agents)	OCR, image / speech understanding

Multimodal usage (llama.cpp)

# Pick a main model (text tower)
wget https://huggingface.co/batiai/Gemma-4-E2B-it-GGUF/resolve/main/google-gemma-4-E2B-it-Q4_K_M.gguf

# Get the multimodal projector (vision + audio in one file)
wget https://huggingface.co/batiai/Gemma-4-E2B-it-GGUF/resolve/main/mmproj-BF16.gguf

Server mode (image input via OpenAI-compatible Vision API):

llama-server \
  -m google-gemma-4-E2B-it-Q4_K_M.gguf \
  --mmproj mmproj-BF16.gguf \
  -c 32768 --host 127.0.0.1 --port 8080

curl http://127.0.0.1:8080/v1/chat/completions -d '{
  "messages": [{
    "role": "user",
    "content": [
      {"type": "image_url", "image_url": {"url": "data:image/jpeg;base64,..."}},
      {"type": "text", "text": "What does this show?"}
    ]
  }]
}'

One-shot CLI with image:

llama-mtmd-cli \
  -m google-gemma-4-E2B-it-Q4_K_M.gguf \
  --mmproj mmproj-BF16.gguf \
  --image ~/Desktop/photo.jpg \
  -p "describe this image"

One-shot CLI with audio:

llama-mtmd-cli \
  -m google-gemma-4-E2B-it-Q4_K_M.gguf \
  --mmproj mmproj-BF16.gguf \
  --audio ~/Downloads/voice.wav \
  -p "transcribe and summarize"

mmproj quantization

File	Quant	Size	Why this only?
`mmproj-BF16.gguf`	BF16	~940 MB	The vision + audio projector tensors don't satisfy K-quant block-alignment requirements (108+ tensors require fallback), so Q6_K reliably aborts on this combined projector. BF16 is the only safe choice today — applies to every quantizer of this model. The main text GGUF is unaffected.

Why E series (E2B / E4B) vs 26B / 31B?

	E2B / E4B	26B-A4B / 31B
Audio support	✅	❌ (vision only)
Min RAM	8 GB+	24 GB+
Speed	very fast (small)	slower (larger)
Reasoning depth	lower	higher
Use case	edge / mobile-class Mac / voice + image	desktop chat + agents

If you need voice + image + text in one model on a small Mac, E series is the only Gemma 4 option.

Why BatiAI?

	BatiAI	Third-party (unsloth, etc.)
Source	Quantized directly from official Google weights	Often re-quantized from other GGUFs
Compatibility	✅ Verified on Ollama 0.20+	❌ Known issues with Ollama 0.20+
Tested on	Real Mac mini M4 (16 GB) + MacBook Pro M4 Max	Often untested
Tool calling	✅ Verified with BatiFlow's 57 tool functions	Often broken
Korean	✅ Validated	Not tested
Multimodal	✅ Vision + audio mmproj available	Often missing
Signing	`general.author: BatiAI` for provenance	Varies

About BatiFlow

flow.bati.ai

BatiFlow is a macOS-native AI desktop automation app — just 5 MB, built with Swift.

Free & Unlimited — On-device AI via Ollama, no API costs
100 % Private — All data stays on your Mac
Ultra Lightweight — Native macOS app, only 5 MB
57 built-in tools — calendar, notes, reminders, files, email, browser, messaging, and more

Related models in the BatiAI Gemma 4 lineup

Model	Modalities	Min RAM	Repo
Gemma 4 E2B-it	text + image + audio	8 GB	this repo
Gemma 4 E4B-it	text + image + audio	8 GB	batiai/Gemma-4-E4B-it-GGUF
Gemma 4 26B-A4B-it	text + image / video (no audio)	24 GB	batiai/Gemma-4-26B-A4B-it-GGUF
Gemma 4 31B-it	text + image / video (no audio)	24 GB	batiai/Gemma-4-31B-it-GGUF

Technical Details

Original Model: google/gemma-4-E2B-it
Architecture: Gemma 4 (Edge variant)
Modalities: Text (primary) + Image + Audio (via opt-in mmproj)
License: Gemma (commercial use permitted under terms)
Quantized with: llama.cpp
Quantized by: BatiAI
GGUF metadata: general.author: BatiAI, general.url: https://flow.bati.ai

License

Mirrors the upstream Gemma license. Commercial use permitted per Google's Gemma terms.

BatiAI's quantization pipeline is provided under MIT.

batiai/gemma-4-E2B-it-GGUF