Instructions to use AsadIsmail/gemma-4-E4B-it-ternary with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AsadIsmail/gemma-4-E4B-it-ternary with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="AsadIsmail/gemma-4-E4B-it-ternary")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("AsadIsmail/gemma-4-E4B-it-ternary", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use AsadIsmail/gemma-4-E4B-it-ternary with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AsadIsmail/gemma-4-E4B-it-ternary"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AsadIsmail/gemma-4-E4B-it-ternary",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/AsadIsmail/gemma-4-E4B-it-ternary

SGLang

How to use AsadIsmail/gemma-4-E4B-it-ternary with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "AsadIsmail/gemma-4-E4B-it-ternary" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AsadIsmail/gemma-4-E4B-it-ternary",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "AsadIsmail/gemma-4-E4B-it-ternary" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AsadIsmail/gemma-4-E4B-it-ternary",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use AsadIsmail/gemma-4-E4B-it-ternary with Docker Model Runner:
```
docker model run hf.co/AsadIsmail/gemma-4-E4B-it-ternary
```

Gemma 4-E4B-it — Ternary Quantized

Ternary-quantized version of google/gemma-4-E4B-it, produced with ternary-quant.

Gemma 4-E4B is Google's multimodal model that processes both images and text. This ternary-quantized version retains multimodal capabilities while reducing model size through a HuggingFace-native, component-aware PTQ workflow.

Quantization details

Metric	Value
Scheme	tritplane3 (3-plane progressive ternary)
Components quantized	text_backbone, multimodal_connector (342 linear layers)
Vision encoder	Kept in FP16 (preserving image understanding quality)
Group size	32
Calibration iterations	10
Stored size	4234 MB (~4.2 GB)
FP16 size	~16 GB
Compression ratio	~3.8x

Usage

from ternary_quant.inference import load_ternary_model

# Load model (auto-detects best device: CUDA > MPS > CPU)
model, processor = load_ternary_model(
    "AsadIsmail/gemma-4-E4B-it-ternary",
    runtime_mode="metal"  # Use "cached" for NVIDIA GPU or CPU
)

# Image understanding
from PIL import Image
image = Image.open("photo.jpg")
inputs = processor(text="Describe this image", images=image, return_tensors="pt")
inputs = {k: v.to(model.device) for k, v in inputs.items()}
outputs = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(outputs[0], skip_special_tokens=True))

Hardware requirements

Runtime	Min RAM	Speed	Notes
`metal` (Apple Silicon)	~8 GB	Good	Native Metal kernels
`cached` (any device)	~12 GB	Fastest	Dequantizes once at load
`triton_memory` (NVIDIA)	~6 GB	Moderate	Packed ternary in VRAM

Reproduce

pip install ternary-quant
ternary-quant quantize-broad google/gemma-4-E4B-it \
    --output ./gemma-4-E4B-it-ternary \
    --components text_backbone multimodal_connector \
    --scheme tritplane3 \
    --dtype float16 \
    --eval

Part of the ternary-models collection

See github.com/Asad-Ismail/ternary-models for the full collection of ternary-quantized multimodal models.

Citation

@software{ternary_quant,
  author = {Ismail, Asad},
  title = {ternary-quant: Post-training ternary quantization for HuggingFace generative models},
  url = {https://github.com/Asad-Ismail/ternary-quant},
  year = {2026}
}