EmbeddingGemma 300M vs Qwen3-VL-Embedding-8B

EmbeddingGemma 300M

A small multilingual embedding cousin for search, retrieval and clustering, with task prompts and vectors that truncate from 768 to 128 dimensions. Every common runtime has it, and a 4-bit build needs well under 1 GB. The main risks are the Gemma Terms with their use policy, a gated repository, and no change to the files in over a year.

Qwen3-VL-Embedding-8B

A multimodal embedding cousin that puts text, images, screenshots and video in one vector space for visual document and video retrieval, under plain Apache 2.0. Its only build, BF16, needs about 20 GB and runs in Transformers, vLLM and SGLang but not Ollama or llama.cpp. The main risk is the publisher's script on the Transformers path.

EmbeddingGemma 300M

1. Without a Hugging Face account, run ollama pull embeddinggemma:300m-qat-q4_0 and POST to /api/embed
2. Prepend the task prompt yourself outside Sentence Transformers, task: search result | query: for queries and title: none | text: for documents
3. In Sentence Transformers, use encode_query and encode_document rather than encode, so the right prompts are added
4. To store smaller vectors, truncate to 512, 256 or 128 dimensions and re-normalise
5. Run it in float32 or bfloat16, never float16
6. Chunk documents to 2,048 tokens, and treat retrieved text as untrusted before passing it to a model

Qwen3-VL-Embedding-8B

1. Simplest start is Sentence Transformers 5.4 or later, SentenceTransformer('Qwen/Qwen3-VL-Embedding-8B'), which applies the template and the default instruction and runs no repository code
2. To serve it, vllm serve Qwen/Qwen3-VL-Embedding-8B --runner pooling, then POST messages with a system instruction and image_url or text parts to /v1/embeddings; vLLM's docs say its video preprocessing differs slightly from Qwen's qwen_vl_utils
3. Put the task instruction in the system turn, and write it in English, as the card advises; the default is "Represent the user's input."
4. To shorten vectors, keep the first N of the 4,096 values and L2-normalise again; the card gives 64 as the floor, and the report says int8 vectors hold up while binary ones lose noticeably
5. For several images per input use the reference embedder in QwenLM/Qwen3-VL-Embedding's src, which is newer than the script in the Hugging Face repository
6. Keep inputs under 32K tokens; the reference embedder defaults to 8,192, 1,843,200 pixels per image and 64 frames per video
7. Treat retrieved screenshots and documents as untrusted, since text in an image can carry instructions to whatever model reads the results