EmbeddingGemma 2 Artemis

TurinTech models, built with Artemis. 768-dimensional text embeddings from Google’s EmbeddingGemma 2, for retrieval and similarity.

The default has a measured size/retrieval trade-off: on our exposed development panel it had slightly higher nDCG and lower Recall than Unsloth Q4. The larger alternative scored higher on both at a larger file size. These are selection-panel observations, not independent evidence of general superiority. Measurements, exact bytes and SHA-256 checksums.

Use the embedding API

Start Ollama, then pull the exact default:

ollama pull turintech/embeddinggemma-2-artemis:q4_k_s

curl http://localhost:11434/api/embed -H 'Content-Type: application/json' -d '{
  "model": "turintech/embeddinggemma-2-artemis:q4_k_s",
  "input": [
    "task: search result | query: How does photosynthesis work?",
    "title: Photosynthesis | text: Plants use light energy to convert carbon dioxide and water into sugars."
  ],
  "truncate": false,
  "options": {"num_ctx": 512, "num_thread": 2, "num_gpu": 0}
}'

Keep the query/document prefixes, and replace the text after them. Use q5_k_s for the larger file. Both files passed a two-input CPU check with Ollama 0.40.4, returning finite, normalized 768-dimensional vectors at a 512-token context setting. Longer inputs are rejected by this request rather than silently truncated. This is an embedding model; use /api/embed, not chat. GPU operation, speed and retrieval quality through Ollama have not been established here.

The tag labels are size classes, not stock tensor recipes. Both files contain mixed tensor types and retain a broad Q4_K_M header, which Ollama also displays in its runtime listing. That metadata label does not identify the selected artifact. Exact file hashes identify each variant. File MB are decimal download sizes, not runtime memory; compression does not shrink the 768-dimensional vector index.

Original model: Google EmbeddingGemma 2. Apache 2.0; license and notice.

Hugging Face files and evidence · TurinTech embedding models