nomic-ai/Nomic Embed Text v2 MoE 475M

• License: Apache-2.0
• Release date: Not explicitly documented in the original model card.
• Modalities: Input: text; Output: embeddings
• Original model: nomic-ai/nomic-embed-text-v2-moe

Description

Nomic Embed Text v2 MoE converts text in roughly 100 languages into vectors for multilingual retrieval. It uses eight experts with top-two routing, with 475 million total parameters and 305 million active during inference. Matryoshka training supports reducing its full 768-dimensional embeddings to 256 dimensions for smaller representations.

Run

self serve nomic-embed-text-v2-moe:475m

Here is a link to the API docs: Embeddings API.

Example Request

With the server running, open another terminal and send a request:

curl http://localhost:8080/v1/embeddings \
	-H 'Content-Type: application/json' \
	-d '{"input":"search_query: What is multilingual retrieval?"}'

The server returns a JSON response containing the embedding vector.

Engine Parameters

Served by the upstream llama-server through the embedding adapter. These are startup settings, not per-request API fields. Settings not specified in the registry are left to the bundled engine's defaults.

Parameter	Default	What it controls
context_size	8192	Engine context size in tokens; accepts integers from 512 to 1048576. The upstream model card specifies a 512-token maximum; a larger engine setting does not establish model support.
pooling	mean	Embedding pooling; accepts none, mean, cls, last, or rank. Keep mean for this model's sentence embeddings.
threads	Engine default (not set in registry)	Number of CPU workers; accepts integers from 1 to 1024.
gpu_layers	Engine default (not set in registry)	Layers offloaded to GPU; accepts integers from -1 to 10000, with -1 meaning all layers.
flash_attn	Engine default (not set in registry)	Flash Attention mode: auto, on, or off; availability depends on the backend and model.
parallel	Engine default (not set in registry)	Number of parallel engine slots; accepts integers from 1 to 256.

Override startup settings with --set <parameter>=<value>.

Quantizations and Files

Default variant: q4. Sizes and SHA-256 hashes are pinned in the model registry.

q4 (default)

File	Role	Size	SHA-256	Origin
nomic-embed-text-v2-moe.Q4_K_M.gguf	model	328.2 MiB (344120288 bytes)	b5fb2811647b8ef461519a68a3bf67014a84a66a130c8a2af9413ff9f06d3f22	nomic-ai/nomic-embed-text-v2-moe-GGUF

Total download: 328.2 MiB (344120288 bytes).

Notes

• Task prefixes are required: use search_query: for retrieval queries and search_document: for indexed documents, including non-English text. Include the prefix in every input string; the adapter does not add it.
• The upstream model card specifies a maximum sequence length of 512 tokens, despite the registry's context_size: 8192. Keep inputs within the documented model limit; the runtime setting does not extend the trained context.
• The API returns full vectors and does not support the dimensions request field. For downstream Matryoshka reduction, follow the upstream procedure: truncate to 256 dimensions, then L2-normalize. Process queries and documents consistently.
• Download size is not the total runtime memory requirement.
• The origin link uses main because the registry pins the file hash, not a repository revision.

Quality benchmarks

MTEB: Listed; rank not imported

nomic-ai/nomic-embed-text-v2-moe

Registered in MTEB; no suite-specific rank imported

Checked 2026-10-13

Benchmarks

No submitted benchmark reports yet.