embeddinggemma-300m-mic

Slim model weights, executable usage and framework expectations in one repository.

• Contract manifest
• Transformers usage
• Transformers expectations
• Fixture provenance

Contract

EmbeddingGemma embedding

Runnable usage and the framework expectations directory are linked above.

Slim model

Slim fixture: embeddinggemma-300m-mic

A 16.9M-parameter fixture for the EmbeddingGemma MIC, covering google/embeddinggemma-300m. It is a test artifact, not a usable model, tagged test-artifact to keep it out of candidate-model search.

• Fixture key cf79e3a783d5. It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
• Same architecture, smaller config: Gemma3TextModel, shrunk by the generic rule in sweep.py (num_hidden_layers 2).
• Processor files copied byte for byte from google/embeddinggemma-300m: added_tokens.json, config_sentence_transformers.json, generation_config.json, modules.json, notebook.ipynb, sentence_bert_config.json, special_tokens_map.json, tokenizer.json, tokenizer.model, tokenizer_config.json.
• Random weights (seed 17). They passed the input-sensitivity gate: every perturbation of the contract's inputs moved the logits by at least 0.001 (weakest: input_ids:reverse, 1.62), so no training was needed.
• Same tensor names as the real checkpoint. The real checkpoint has every parameter, and so does the fixture.
• Saved in float32, the dtype google/embeddinggemma-300m loads in by default.
• Finite in float32, bfloat16, and float16 (forward, loss, gradients).

Built by build_fixtures.py; build.json records the pinned sources, gate results, and file hashes. This repository includes the slim weights, usage and framework expectations; rebuild the other files with python example/build_fixtures.py --fixture embeddinggemma-300m-mic.

Sentence Transformers

This fixture supports the public Sentence Transformers API. The existing backbone weights and Transformers checks are unchanged. Missing pooling and tiny projection assets complete the pipeline already declared in modules.json.

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("huggingface-mic/embeddinggemma-300m-mic")
queries = ["Which planet is known as the Red Planet?", "Where do penguins live?"]
documents = ["Mars is called the Red Planet.", "Penguins live in Antarctica."]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
scores = model.similarity(query_embeddings, document_embeddings)
assert scores.shape == (2, 2)

Executable ST usage and reviewed ST baselines are included. The scores are test outputs from tiny weights and do not measure retrieval quality.