Class EmbeddingPresets

Namespace

Qavren.Edge.Embeddings.Onnx

Assembly

Qavren.Edge.Embeddings.Onnx.dll

The four pinned presets. Only half of this catalogue is hand-written: every OnnxModelManifest - repo, full commit SHA, per-file path, role, size and SHA-256, and SPDX licence - lives in EmbeddingPresets.g.cs, generated from one paths-info call per repo and committed a wave earlier. Nothing here retypes a digest.

public static class EmbeddingPresets

Inheritance

object

EmbeddingPresets

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Properties

All

Every shipped preset, in catalogue order.

BgeSmallEnV15

bge-small-en-v1.5, 133,093,490 B fp32, 384-d, CLS pooling, 512 tokens, MIT. BAAI publishes no quantized ONNX. QueryPrefix is set and applied to queries only.

MiniLmL6V2Fp32

all-MiniLM-L6-v2 fp32, 90,405,214 B. Same numbers, no quantization loss.

MiniLmL6V2Int8

DEFAULT. all-MiniLM-L6-v2 int8, 23,026,053 B, 384-d, mean + L2, 256 tokens, no prefixes, Apache-2.0. Its qint8_arm64 / qint8_avx512 / qint8_avx512_vnni files are ONE blob under three names, so a single download covers every RID. 384 matches SP1's own VecTable example, so the sample needs no schema change.

NomicEmbedTextV15Int8

OPT-IN. nomic-embed-text-v1.5 int8, 137,296,292 B, 768-d, Apache-2.0, mandatory search_query: / search_document: prefixes. Doubles vec0 storage. Pooling is mean + PostPoolLayerNorm = true + L2: this model's modules.json has no Normalize module and its reference pipeline inserts F.layer_norm over the 768 dim between pooling and L2. It is the only shipped preset that sets the flag, and the preset-catalogue table test guards it.

Methods

ById(string)

Resolves a preset by Id.