EmbeddingGemma 2 — Text + Image — FP32 — Transformers / SentenceTransformers

A modular deployment export of Google DeepMind's EmbeddingGemma 2. Only the text backbone, vision encoder and vision projection are retained; the audio encoder and audio projection are absent. No training, distillation, or quantization was applied.

FP32 weights are a lossless expansion of the upstream BF16 values. This does not recover precision absent from the original trained checkpoint.

Weight size is not total runtime memory. 270M/440M in repository names are rounded deployment sizes.

Install

pip install "transformers>=5.19.0" "sentence-transformers>=6.1.0"

Tested with Transformers 5.19.0, SentenceTransformers 6.1.0, and PyTorch 2.14.1.

Text and code

import torch
from sentence_transformers import SentenceTransformer

model = SentenceTransformer(
    "jayyun98/embeddinggemma-2-text-image-440m-fp32",
    device="cpu",
    model_kwargs={"dtype": torch.float32},
)
queries = model.encode(
    ["What causes the northern lights?", "로컬 코드 검색 모델을 찾고 싶어요."],
    prompt_name="SearchQuery",
    normalize_embeddings=True,
)
documents = model.encode(
    ["Charged particles from the sun cause the northern lights."],
    prompt_name="Document",
    normalize_embeddings=True,
)
print(model.similarity(queries, documents))

code_query = model.encode(
    "Find a Python function that sorts a list.",
    prompt_name="CodeRetrieval",
    truncate_dim=256,
    normalize_embeddings=True,
)

Use SearchQuery for search queries, CodeRetrieval for code-search queries, and Document for corpus items. For the MLX API, prepend the corresponding literal prefix from config_sentence_transformers.json, as shown above. For titled documents, use title: {title} | text: {content} without another prefix. Use matching dimensions for queries and documents, and normalize after truncation. This FP32 package stores FP32 weights; loading it as BF16 changes runtime precision.

Images and mixed text/image inputs

from PIL import Image

image = Image.open("photo.jpg").convert("RGB")
image_vector = model.encode({"image": image}, normalize_embeddings=True)
mixed_vector = model.encode(
    {"text": "A description of this photo.", "image": image},
    normalize_embeddings=True,
)

Verification

CPU FP32 inference was tested in the package's named precision. Outputs were exactly equal to the full source model running in the same precision. Checks cover English/Korean search and document text, code queries, and 128/256/512-dimensional normalized vectors. Image-only, mixed text/image, and interleaved two-image inputs were also checked.

Measurements compare with the pinned full Google checkpoint. They are small numerical and loading checks, not MTEB results, a retrieval-quality evaluation, or a speed benchmark. The original prompts, pooling behavior, tokenizer and processor assets are retained. Processor metadata does not restore the removed encoder weights. This package targets text/image; audio is unavailable and video workflows were not evaluated.

• conversion.json: storage precision, removal and conversion checks.
• verification.json: runtime, numerical results and tested environment.
• Original source revision: 914f7f89142e33e77833254d9c9b90c3cef7303b.

Attribution and license

Original weights and tokenizer/processor assets: Google DeepMind.

This is an independent derivative deployment export, not an official Google release. The Apache 2.0 LICENSE and NOTICE are included. See the original model card for training, intended use and limitations.