zeromodels/gemma-3n-e2b

See our collection for all Gemma 3n sizes and variants.

Run Gemma 3n with Keras 3: JAX, PyTorch, or TensorFlow

zeromodels/gemma-3n-e2b

Pure-Keras 3 conversion of google/gemma-3n-E2B for zeromodels. One implementation runs unmodified on TensorFlow / Torch / JAX. This is a base (pretrained) checkpoint, served here as image + audio + text -> text via Gemma3nConditionalGenerate; weights are stored in bfloat16.

For model details, license, and usage terms, see Google's model card.

✨ Quick start

Text-only

import os
os.environ["KERAS_BACKEND"] = "torch"  # or "jax" / "tensorflow"

from zeromodels.models.gemma3n import Gemma3nTextGenerate, Gemma3nTokenizer

model = Gemma3nTextGenerate.from_weights("zeromodels/gemma-3n-e2b")
tokenizer = Gemma3nTokenizer.from_weights("zeromodels/gemma-3n-e2b")

inputs = tokenizer([{"role": "user", "content": "Hello, who are you?"}])
outputs = model.generate(**inputs, max_new_tokens=64)
print(tokenizer.decode(outputs[0]))

Image + audio + text

from zeromodels.models.gemma3n import Gemma3nConditionalGenerate, Gemma3nProcessor

model = Gemma3nConditionalGenerate.from_weights("zeromodels/gemma-3n-e2b")
processor = Gemma3nProcessor.from_weights("zeromodels/gemma-3n-e2b")

conversation = [
    {"role": "user", "content": [
        {"type": "image", "url": "https://.../image.jpg"},
        {"type": "text", "text": "Describe this image."},
    ]},
]
inputs = processor(conversation)
outputs = model.generate(**inputs, max_new_tokens=64)
print(processor.decode(outputs[0]))

Load any Gemma 3n variant the same way with from_weights("zeromodels/<variant>"):

Variant	Hub
gemma-3n-e2b	zeromodels/gemma-3n-e2b
gemma-3n-e2b-it	zeromodels/gemma-3n-e2b-it
gemma-3n-e4b	zeromodels/gemma-3n-e4b
gemma-3n-e4b-it	zeromodels/gemma-3n-e4b-it

Tips

• Set KERAS_BACKEND before importing Keras / zeromodels.
• Loads in bfloat16 by default. Pass load_dtype="float32" for full precision, or quantization="int8" to shrink further.
• See the Gemma 3n docs.
• Community / upstream weights still work via the hf: prefix: Gemma3nConditionalGenerate.from_weights("hf:google/gemma-3n-E2B").

Special Thanks

A huge thank you to the Google Gemma authors for creating and releasing these models.

License: Gemma (gated). Accept the license on the upstream Hub card before downloading.