Granite Embedding Small English R2 vs Qwen3-VL-Embedding-8B

Granite Embedding Small English R2

A tiny English embedding cousin for retrieval over long documents and code, 47M parameters with 384-dimension vectors and an 8,192-token input, under Apache 2.0 with a permissive data story. It runs on any hardware through Sentence Transformers, vLLM or llama.cpp. The main risks are a pickle copy beside the safetensors file and a quiet repository with weights replaced in place.

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.

Granite Embedding Small English R2

1. Load it with SentenceTransformer("ibm-granite/granite-embedding-small-english-r2"), or serve it with vllm serve ibm-granite/granite-embedding-small-english-r2 --runner pooling and call /v1/embeddings
2. Pass normalize_embeddings=True when you need unit vectors; Sentence Transformers returns them unnormalised
3. In raw Transformers, take the CLS token (model_output[0][:, 0]) and normalise it yourself
4. Don't add query or passage prefixes; the model takes plain text
5. Load the safetensors file and ignore pytorch_model.bin, and chunk text past 8,192 tokens
6. 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