bge-large-e8-snap

bge-large-en-v1.5 fine-tuned with RF-Snap to align embeddings to the E8 lattice.

Part of the LatticeMemory project. GitHub →

What Makes This Different

Standard embedding models output float32 vectors. This model is trained so its outputs naturally snap to the nearest point in the E8 lattice — the densest sphere packing in 8 dimensions. The result: a 32x smaller index with better STS quality than the float32 baseline.

Benchmarks

Metric	Float baseline (bge-large-en-v1.5)	This model
STSBenchmark (Spearman)	0.8637	0.8714 (+0.0077)
STS13	—	0.8826
Index compression	1x	32x
Retrieval @ 100K docs	20.8 ms (scan)	O(1) on E8 key hit
Recall@10 (MS-MARCO 1K)	100%	100%

Compression basis: 1 address byte per 8-dim block × 128 blocks = 128 bytes for a 1024-dim embedding, vs 4,096 bytes for float32 = 32x. This applies to E8 key storage; see LatticeMemory for hybrid mode, which also stores a dense fallback index for asymmetric retrieval.

Usage

from sentence_transformers import SentenceTransformer
import torch, math, torch.nn.functional as F

model = SentenceTransformer("dfrokido/bge-large-e8-snap")
embeddings = model.encode(["What is the capital of France?"], convert_to_tensor=True)
embeddings = F.normalize(embeddings.float(), p=2, dim=1)
# Embeddings are now ready for LatticeMemory indexing — 32x smaller index

Training

Fine-tuned from BAAI/bge-large-en-v1.5 using RF-Snap training:

• Loss: cosine similarity + MNRL + E8 address cross-entropy + teacher anchor
• Data: NLI 50K pairs
• Config: freeze_until=10/24, lr=3e-6, batch=8, grad_accum=4, 1 epoch
• Hardware: GTX 1660 Ti (6GB VRAM)

LatticeMemory

This model powers LatticeMemory (GitHub, pip install latticememory) — a semantic cache, dedup, and hybrid memory library for LLM applications. 32x compressed E8 keys for instant repeat-query cache hits, dense fallback for novel retrieval.

Design partner inquiries: dfrokido@gmail.com